SHARE
TWEET

do you trust this computer eng

sofiasari Feb 5th, 2019 1,374 Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
  1. 0
  2. 00:00:05,000 --> 00:00:15,000
  3. INDOXXI
  4. Support dengan like & share :)
  5.  
  6. 1
  7. 00:00:30,000 --> 00:00:33,500
  8. Correction and synchronisation:
  9. Mazrim Taim
  10.  
  11. 2
  12. 00:00:35,078 --> 00:00:41,302
  13. What we're on the brink of is
  14. a world of increasingly intense,
  15.  
  16. 3
  17. 00:00:41,345 --> 00:00:45,219
  18. sophisticated artificial intelligence.
  19.  
  20. 4
  21. 00:00:45,262 --> 00:00:48,396
  22. Technology is evolving
  23. so much faster than our society
  24.  
  25. 5
  26. 00:00:48,439 --> 00:00:51,181
  27. has the ability
  28. to protect us as citizens.
  29.  
  30. 6
  31. 00:00:51,486 --> 00:00:55,707
  32. The robots are coming,
  33. and they will destroy our livelihoods.
  34.  
  35. 7
  36. 00:01:01,887 --> 00:01:04,238
  37. You have a networked intelligence
  38. that watches us,
  39.  
  40. 8
  41. 00:01:04,281 --> 00:01:08,590
  42. knows everything about us,
  43. and begins to try to change us.
  44.  
  45. 9
  46. 00:01:08,633 --> 00:01:12,768
  47. Twitter has become
  48. the world's number-one news site.
  49.  
  50. 10
  51. 00:01:12,811 --> 00:01:15,205
  52. Technology is never good or bad.
  53.  
  54. 11
  55. 00:01:15,249 --> 00:01:18,948
  56. It's what we do
  57. with the technology.
  58.  
  59. 12
  60. 00:01:18,991 --> 00:01:22,734
  61. Eventually, millions of people
  62. are gonna be thrown out of jobs
  63.  
  64. 13
  65. 00:01:22,778 --> 00:01:25,737
  66. because their skills
  67. are going to be obsolete.
  68.  
  69. 14
  70. 00:01:25,781 --> 00:01:27,435
  71. Mass unemployment...
  72.  
  73. 15
  74. 00:01:27,478 --> 00:01:31,727
  75. greater inequalities,
  76. even social unrest.
  77.  
  78. 16
  79. 00:01:32,570 --> 00:01:35,530
  80. Regardless of whether
  81. to be afraid or not afraid,
  82.  
  83. 17
  84. 00:01:35,573 --> 00:01:38,185
  85. the change is coming,
  86. and nobody can stop it.
  87.  
  88. 18
  89. 00:01:44,582 --> 00:01:48,146
  90. We've invested huge amounts of money,
  91. and so it stands to reason
  92.  
  93. 19
  94. 00:01:48,148 --> 00:01:50,893
  95. that the military,
  96. with their own desires,
  97.  
  98. 20
  99. 00:01:50,936 --> 00:01:53,330
  100. are gonna start to use
  101. these technologies.
  102.  
  103. 21
  104. 00:01:53,374 --> 00:01:57,552
  105. Autonomous weapons systems
  106. could lead to a global arms race
  107.  
  108. 22
  109. 00:01:57,595 --> 00:02:00,032
  110. to rival the Nuclear Era.
  111.  
  112. 23
  113. 00:02:02,339 --> 00:02:05,429
  114. We know what the answer is.
  115. They'll eventually be killing us.
  116.  
  117. 24
  118. 00:02:10,826 --> 00:02:15,874
  119. These technology leaps
  120. are gonna yield incredible miracles
  121.  
  122. 25
  123. 00:02:15,918 --> 00:02:18,181
  124. and incredible horrors.
  125.  
  126. 26
  127. 00:02:24,274 --> 00:02:29,323
  128. We created it,
  129. so I think, as we move forward,
  130.  
  131. 27
  132. 00:02:29,366 --> 00:02:33,762
  133. this intelligence
  134. will contain parts of us.
  135.  
  136. 28
  137. 00:02:33,805 --> 00:02:35,981
  138. And I think the question is:
  139.  
  140. 29
  141. 00:02:36,025 --> 00:02:39,463
  142. Will it contain
  143. the good parts...
  144.  
  145. 30
  146. 00:02:39,507 --> 00:02:41,378
  147. or the bad parts?
  148.  
  149. 31
  150. 00:03:04,836 --> 00:03:08,840
  151. The survivors
  152. called the war "Judgment Day."
  153.  
  154. 32
  155. 00:03:08,884 --> 00:03:12,583
  156. They lived only to face
  157. a new nightmare:
  158.  
  159. 33
  160. 00:03:12,627 --> 00:03:14,319
  161. The war against the machines.
  162.  
  163. 34
  164. 00:03:15,456 --> 00:03:18,023
  165. I think
  166. we've completely fucked this up.
  167.  
  168. 35
  169. 00:03:18,067 --> 00:03:21,549
  170. I think Hollywood has managed
  171. to inoculate the general public
  172.  
  173. 36
  174. 00:03:21,592 --> 00:03:24,247
  175. against this question.
  176.  
  177. 37
  178. 00:03:24,291 --> 00:03:28,251
  179. The idea of machines
  180. that will take over the world.
  181.  
  182. 38
  183. 00:03:28,295 --> 00:03:30,645
  184. Open the pod bay doors, HAL.
  185.  
  186. 39
  187. 00:03:30,688 --> 00:03:33,561
  188. I'm sorry, Dave.
  189.  
  190. 40
  191. 00:03:33,604 --> 00:03:35,911
  192. I'm afraid I can't do that.
  193.  
  194. 41
  195. 00:03:37,434 --> 00:03:38,696
  196. HAL?
  197.  
  198. 42
  199. 00:03:38,740 --> 00:03:40,437
  200. We've cried wolf enough times...
  201.  
  202. 43
  203. 00:03:40,481 --> 00:03:42,483
  204. ...that the public
  205. has stopped paying attention,
  206.  
  207. 44
  208. 00:03:42,484 --> 00:03:44,120
  209. because it feels like
  210. science fiction.
  211.  
  212. 45
  213. 00:03:44,121 --> 00:03:46,001
  214. Even sitting here talking
  215. about it right now,
  216.  
  217. 46
  218. 00:03:46,002 --> 00:03:48,301
  219. it feels a little bit silly,
  220. a little bit like,
  221.  
  222. 47
  223. 00:03:48,302 --> 00:03:51,697
  224. "Oh, this is an artifact
  225. of some cheeseball movie."
  226.  
  227. 48
  228. 00:03:51,709 --> 00:03:56,584
  229. The WOPR spends all its time
  230. thinking about World War III.
  231.  
  232. 49
  233. 00:03:56,627 --> 00:03:59,064
  234. But it's not.
  235.  
  236. 50
  237. 00:03:59,108 --> 00:04:02,111
  238. The general public is about
  239. to get blindsided by this.
  240.  
  241. 51
  242. 00:04:11,555 --> 00:04:13,514
  243. As a society and as individuals,
  244.  
  245. 52
  246. 00:04:13,557 --> 00:04:18,954
  247. we're increasingly surrounded
  248. by machine intelligence.
  249.  
  250. 53
  251. 00:04:18,997 --> 00:04:22,653
  252. We carry this pocket device
  253. in the palm of our hand
  254.  
  255. 54
  256. 00:04:22,697 --> 00:04:24,829
  257. that we use to make
  258. a striking array
  259.  
  260. 55
  261. 00:04:24,873 --> 00:04:26,831
  262. of life decisions right now,
  263.  
  264. 56
  265. 00:04:26,875 --> 00:04:29,007
  266. aided by a set
  267. of distant algorithms
  268.  
  269. 57
  270. 00:04:29,051 --> 00:04:30,748
  271. that we have no understanding.
  272.  
  273. 58
  274. 00:04:34,186 --> 00:04:36,537
  275. We're already pretty jaded
  276. about the idea
  277.  
  278. 59
  279. 00:04:36,580 --> 00:04:37,929
  280. that we can talk to our phone,
  281.  
  282. 60
  283. 00:04:37,973 --> 00:04:40,062
  284. and it mostly understands us.
  285.  
  286. 61
  287. 00:04:40,105 --> 00:04:42,456
  288. I found quite a number
  289. of action films.
  290.  
  291. 62
  292. 00:04:42,499 --> 00:04:44,327
  293. Five years ago -- no way.
  294.  
  295. 63
  296. 00:04:44,371 --> 00:04:47,678
  297. Robotics.
  298. Machines that see and speak...
  299.  
  300. 64
  301. 00:04:47,722 --> 00:04:48,897
  302. ...and listen.
  303.  
  304. 65
  305. 00:04:48,940 --> 00:04:50,202
  306. All that's real now.
  307.  
  308. 66
  309. 00:04:50,246 --> 00:04:51,639
  310. And these technologies
  311.  
  312. 67
  313. 00:04:51,682 --> 00:04:54,886
  314. are gonna fundamentally
  315. change our society.
  316.  
  317. 68
  318. 00:04:55,730 --> 00:05:00,212
  319. Now we have this great
  320. movement of self-driving cars.
  321.  
  322. 69
  323. 00:05:00,256 --> 00:05:01,953
  324. Driving a car autonomously
  325.  
  326. 70
  327. 00:05:01,997 --> 00:05:05,688
  328. can move people's lives
  329. into a better place.
  330.  
  331. 71
  332. 00:05:06,131 --> 00:05:09,570
  333. I've lost a number of family members,
  334. including my mother,
  335.  
  336. 72
  337. 00:05:09,613 --> 00:05:11,876
  338. my brother and sister-in-law
  339. and their kids,
  340.  
  341. 73
  342. 00:05:11,920 --> 00:05:14,009
  343. to automobile accidents.
  344.  
  345. 74
  346. 00:05:14,052 --> 00:05:18,405
  347. It's pretty clear we could
  348. almost eliminate car accidents
  349.  
  350. 75
  351. 00:05:18,448 --> 00:05:20,102
  352. with automation.
  353.  
  354. 76
  355. 00:05:20,145 --> 00:05:21,843
  356. 30,000 lives in the U.S. alone.
  357.  
  358. 77
  359. 00:05:21,886 --> 00:05:24,955
  360. About a million around the world
  361. per year.
  362.  
  363. 78
  364. 00:05:25,499 --> 00:05:27,501
  365. In healthcare, early indicators
  366.  
  367. 79
  368. 00:05:27,544 --> 00:05:29,503
  369. are the name of the game
  370. in that space,
  371.  
  372. 80
  373. 00:05:29,546 --> 00:05:33,158
  374. so that's another place where
  375. it can save somebody's life.
  376.  
  377. 81
  378. 00:05:33,202 --> 00:05:35,726
  379. Here in
  380. the breast-cancer center,
  381.  
  382. 82
  383. 00:05:35,770 --> 00:05:38,381
  384. all the things that
  385. the radiologist's brain
  386.  
  387. 83
  388. 00:05:38,425 --> 00:05:43,386
  389. does in two minutes,
  390. the computer does instantaneously.
  391.  
  392. 84
  393. 00:05:43,430 --> 00:05:47,303
  394. The computer has looked
  395. at 1 billion mammograms,
  396.  
  397. 85
  398. 00:05:47,347 --> 00:05:49,261
  399. and it takes that data
  400. and applies it
  401.  
  402. 86
  403. 00:05:49,305 --> 00:05:51,438
  404. to this image instantaneously,
  405.  
  406. 87
  407. 00:05:51,481 --> 00:05:54,441
  408. so the medical application
  409. is profound.
  410.  
  411. 88
  412. 00:05:56,399 --> 00:05:59,402
  413. Another really exciting area that
  414. we're seeing a lot of development in
  415.  
  416. 89
  417. 00:05:59,446 --> 00:06:03,275
  418. is actually understanding
  419. our genetic code
  420.  
  421. 90
  422. 00:06:03,319 --> 00:06:06,104
  423. and using that
  424. to both diagnose disease
  425.  
  426. 91
  427. 00:06:06,148 --> 00:06:07,758
  428. and create
  429. personalized treatments.
  430.  
  431. 92
  432. 00:06:11,675 --> 00:06:14,112
  433. The primary application
  434. of all these machines
  435.  
  436. 93
  437. 00:06:14,156 --> 00:06:17,246
  438. will be to extend
  439. our own intelligence.
  440.  
  441. 94
  442. 00:06:17,289 --> 00:06:19,422
  443. We'll be able to make
  444. ourselves smarter,
  445.  
  446. 95
  447. 00:06:19,466 --> 00:06:22,543
  448. and we'll be better
  449. at solving problems.
  450.  
  451. 96
  452. 00:06:22,586 --> 00:06:25,075
  453. We don't have to age.
  454. We'll actually understand aging.
  455.  
  456. 97
  457. 00:06:25,076 --> 00:06:27,126
  458. We'll be able to stop it.
  459.  
  460. 98
  461. 00:06:27,169 --> 00:06:29,519
  462. There's really no limit
  463. to what intelligent machines
  464.  
  465. 99
  466. 00:06:29,563 --> 00:06:30,868
  467. can do for the human race.
  468.  
  469. 100
  470. 00:06:36,308 --> 00:06:39,399
  471. How could a smarter machine
  472. not be a better machine?
  473.  
  474. 101
  475. 00:06:42,053 --> 00:06:44,708
  476. It's hard to say exactly
  477. when I began to think
  478.  
  479. 102
  480. 00:06:44,752 --> 00:06:46,971
  481. that that was a bit naive.
  482.  
  483. 103
  484. 00:06:56,503 --> 00:07:00,898
  485. Stuart Russell, he's basically a god
  486. in the field of artificial intelligence.
  487.  
  488. 104
  489. 00:07:00,942 --> 00:07:04,380
  490. He wrote the book that almost
  491. every university uses.
  492.  
  493. 105
  494. 00:07:04,424 --> 00:07:06,948
  495. I used to say it's the
  496. best-selling AI textbook.
  497.  
  498. 106
  499. 00:07:06,991 --> 00:07:10,255
  500. Now I just say "It's the PDF
  501. that's stolen most often."
  502.  
  503. 107
  504. 00:07:13,694 --> 00:07:17,306
  505. Artificial intelligence is
  506. about making computers smart,
  507.  
  508. 108
  509. 00:07:17,349 --> 00:07:19,830
  510. and from the point
  511. of view of the public,
  512.  
  513. 109
  514. 00:07:19,874 --> 00:07:21,484
  515. what counts as AI
  516. is just something
  517.  
  518. 110
  519. 00:07:21,528 --> 00:07:23,268
  520. that's surprisingly intelligent
  521.  
  522. 111
  523. 00:07:23,312 --> 00:07:25,488
  524. compared to what
  525. we thought computers
  526.  
  527. 112
  528. 00:07:25,532 --> 00:07:28,004
  529. would typically be able to do.
  530.  
  531. 113
  532. 00:07:28,448 --> 00:07:33,801
  533. AI is a field of research
  534. to try to basically simulate
  535.  
  536. 114
  537. 00:07:33,844 --> 00:07:36,717
  538. all kinds of human capabilities.
  539.  
  540. 115
  541. 00:07:36,760 --> 00:07:38,719
  542. We're in the AI era.
  543.  
  544. 116
  545. 00:07:38,762 --> 00:07:40,503
  546. Silicon Valley
  547. has the ability to focus
  548.  
  549. 117
  550. 00:07:40,547 --> 00:07:42,462
  551. on one bright, shiny thing.
  552.  
  553. 118
  554. 00:07:42,505 --> 00:07:45,508
  555. It was social networking and
  556. social media over the last decade,
  557.  
  558. 119
  559. 00:07:45,552 --> 00:07:48,119
  560. and it's pretty clear
  561. that the bit has flipped.
  562.  
  563. 120
  564. 00:07:48,163 --> 00:07:50,557
  565. And it starts
  566. with machine learning.
  567.  
  568. 121
  569. 00:07:50,600 --> 00:07:54,343
  570. When we look back at this moment,
  571. what was the first AI?
  572.  
  573. 122
  574. 00:07:54,386 --> 00:07:57,389
  575. It's not sexy, and it isn't the thing
  576. we could see at the movies,
  577.  
  578. 123
  579. 00:07:57,433 --> 00:08:00,741
  580. but you'd make a great case
  581. that Google created,
  582.  
  583. 124
  584. 00:08:00,784 --> 00:08:03,395
  585. not a search engine,
  586. but a godhead.
  587.  
  588. 125
  589. 00:08:03,439 --> 00:08:06,486
  590. A way for people to ask
  591. any question they wanted
  592.  
  593. 126
  594. 00:08:06,529 --> 00:08:08,270
  595. and get the answer they needed.
  596.  
  597. 127
  598. 00:08:08,313 --> 00:08:11,273
  599. Most people are not
  600. aware that what Google is doing
  601.  
  602. 128
  603. 00:08:11,316 --> 00:08:13,710
  604. is actually a form of
  605. artificial intelligence.
  606.  
  607. 129
  608. 00:08:13,754 --> 00:08:16,234
  609. They just go there,
  610. they type in a thing.
  611.  
  612. 130
  613. 00:08:16,278 --> 00:08:18,323
  614. Google gives them the answer.
  615.  
  616. 131
  617. 00:08:18,367 --> 00:08:21,444
  618. With each search,
  619. we train it to be better.
  620.  
  621. 132
  622. 00:08:21,445 --> 00:08:24,108
  623. Sometimes we're typing a search,
  624. and it tell us the answer
  625.  
  626. 133
  627. 00:08:24,109 --> 00:08:27,434
  628. before you've finished
  629. asking the question.
  630.  
  631. 134
  632. 00:08:27,463 --> 00:08:29,944
  633. You know, who is the president
  634. of Kazakhstan?
  635.  
  636. 135
  637. 00:08:29,987 --> 00:08:31,685
  638. And it'll just tell you.
  639.  
  640. 136
  641. 00:08:31,728 --> 00:08:34,818
  642. You don't have to go to the Kazakhstan
  643. national website to find out.
  644.  
  645. 137
  646. 00:08:34,862 --> 00:08:37,081
  647. You didn't used to be able to do that.
  648.  
  649. 138
  650. 00:08:37,125 --> 00:08:39,475
  651. That is artificial intelligence.
  652.  
  653. 139
  654. 00:08:39,519 --> 00:08:42,783
  655. Years from now when we try
  656. to understand, we will say,
  657.  
  658. 140
  659. 00:08:42,826 --> 00:08:44,567
  660. "How did we miss it?"
  661.  
  662. 141
  663. 00:08:44,611 --> 00:08:48,484
  664. It's one of the striking contradictions
  665. that we're facing.
  666.  
  667. 142
  668. 00:08:48,528 --> 00:08:52,053
  669. Google and Facebook, et al,
  670. have built businesses on giving us,
  671.  
  672. 143
  673. 00:08:52,096 --> 00:08:54,185
  674. as a society, free stuff.
  675.  
  676. 144
  677. 00:08:54,229 --> 00:08:56,013
  678. But it's a Faustian bargain.
  679.  
  680. 145
  681. 00:08:56,057 --> 00:09:00,017
  682. They're extracting something
  683. from us in exchange,
  684.  
  685. 146
  686. 00:09:00,061 --> 00:09:01,628
  687. but we don't know
  688.  
  689. 147
  690. 00:09:01,671 --> 00:09:03,760
  691. what code is running
  692. on the other side and why.
  693.  
  694. 148
  695. 00:09:03,804 --> 00:09:05,846
  696. We have no idea.
  697.  
  698. 149
  699. 00:09:06,589 --> 00:09:08,591
  700. It does strike
  701. right at the issue
  702.  
  703. 150
  704. 00:09:08,635 --> 00:09:11,028
  705. of how much we should
  706. trust these machines.
  707.  
  708. 151
  709. 00:09:14,162 --> 00:09:18,166
  710. I use computers
  711. literally for everything.
  712.  
  713. 152
  714. 00:09:18,209 --> 00:09:21,386
  715. There are so many
  716. computer advancements now,
  717.  
  718. 153
  719. 00:09:21,430 --> 00:09:23,824
  720. and it's become such
  721. a big part of our lives.
  722.  
  723. 154
  724. 00:09:23,867 --> 00:09:26,174
  725. It's just incredible
  726. what a computer can do.
  727.  
  728. 155
  729. 00:09:26,217 --> 00:09:29,090
  730. You can actually carry
  731. a computer in your purse.
  732.  
  733. 156
  734. 00:09:29,133 --> 00:09:31,571
  735. I mean, how awesome is that?
  736.  
  737. 157
  738. 00:09:31,614 --> 00:09:35,052
  739. I think most technology is meant
  740. to make things easier
  741.  
  742. 158
  743. 00:09:35,096 --> 00:09:37,315
  744. and simpler for all of us,
  745.  
  746. 159
  747. 00:09:37,359 --> 00:09:40,362
  748. so hopefully that just
  749. remains the focus.
  750.  
  751. 160
  752. 00:09:40,405 --> 00:09:43,147
  753. I think everybody loves
  754. their computers.
  755.  
  756. 161
  757. 00:09:51,721 --> 00:09:53,810
  758. People don't realize
  759. they are constantly
  760.  
  761. 162
  762. 00:09:53,854 --> 00:09:59,076
  763. being negotiated with
  764. by machines;
  765.  
  766. 163
  767. 00:09:59,120 --> 00:10:02,993
  768. whether that's the price
  769. of products in your Amazon cart,
  770.  
  771. 164
  772. 00:10:03,037 --> 00:10:05,517
  773. whether you can get
  774. on a particular flight,
  775.  
  776. 165
  777. 00:10:05,561 --> 00:10:08,912
  778. whether you can reserve
  779. a room at a particular hotel.
  780.  
  781. 166
  782. 00:10:08,956 --> 00:10:11,959
  783. What you're experiencing
  784. are machine-learning algorithms
  785.  
  786. 167
  787. 00:10:12,002 --> 00:10:14,265
  788. that have determined
  789. that a person like you
  790.  
  791. 168
  792. 00:10:14,309 --> 00:10:17,791
  793. is willing to pay 2 cents more
  794. and is changing the price.
  795.  
  796. 169
  797. 00:10:21,795 --> 00:10:24,014
  798. Now, a computer looks
  799. at millions of people
  800.  
  801. 170
  802. 00:10:24,058 --> 00:10:28,105
  803. simultaneously for
  804. very subtle patterns.
  805.  
  806. 171
  807. 00:10:28,149 --> 00:10:31,369
  808. You can take seemingly
  809. innocent digital footprints,
  810.  
  811. 172
  812. 00:10:31,413 --> 00:10:34,677
  813. such as someone's playlist
  814. on Spotify,
  815.  
  816. 173
  817. 00:10:34,721 --> 00:10:37,201
  818. or stuff that they
  819. bought on Amazon,
  820.  
  821. 174
  822. 00:10:37,245 --> 00:10:40,291
  823. and then use algorithms
  824. to translate this
  825.  
  826. 175
  827. 00:10:40,335 --> 00:10:44,513
  828. into a very detailed and a
  829. very accurate, intimate profile.
  830.  
  831. 176
  832. 00:10:47,603 --> 00:10:50,911
  833. There is a dossier on
  834. each of us that is so extensive
  835.  
  836. 177
  837. 00:10:50,954 --> 00:10:52,695
  838. it would be possibly
  839. accurate to say
  840.  
  841. 178
  842. 00:10:52,739 --> 00:10:55,698
  843. that they know more about you
  844. than your mother does.
  845.  
  846. 179
  847. 00:11:04,098 --> 00:11:06,883
  848. The major cause
  849. of the recent AI breakthrough
  850.  
  851. 180
  852. 00:11:06,927 --> 00:11:08,580
  853. isn't just that some dude
  854.  
  855. 181
  856. 00:11:08,624 --> 00:11:11,583
  857. had a brilliant insight
  858. all of a sudden,
  859.  
  860. 182
  861. 00:11:11,627 --> 00:11:14,325
  862. but simply that we have
  863. much bigger data
  864.  
  865. 183
  866. 00:11:14,369 --> 00:11:18,242
  867. to train them on
  868. and vastly better computers.
  869.  
  870. 184
  871. 00:11:18,286 --> 00:11:19,940
  872. The magic is in the data.
  873.  
  874. 185
  875. 00:11:19,983 --> 00:11:21,463
  876. It's a ton of data.
  877.  
  878. 186
  879. 00:11:21,506 --> 00:11:23,726
  880. I mean, it's data
  881. that's never existed before.
  882.  
  883. 187
  884. 00:11:23,770 --> 00:11:26,686
  885. We've never had
  886. this data before.
  887.  
  888. 188
  889. 00:11:26,729 --> 00:11:30,733
  890. We've created technologies
  891. that allow us to capture
  892.  
  893. 189
  894. 00:11:30,777 --> 00:11:33,040
  895. vast amounts of information.
  896.  
  897. 190
  898. 00:11:33,083 --> 00:11:35,738
  899. If you think of a billion
  900. cellphones on the planet
  901.  
  902. 191
  903. 00:11:35,782 --> 00:11:38,393
  904. with gyroscopes
  905. and accelerometers
  906.  
  907. 192
  908. 00:11:38,436 --> 00:11:39,786
  909. and fingerprint readers...
  910.  
  911. 193
  912. 00:11:39,829 --> 00:11:42,005
  913. couple that with the GPS
  914. and the photos they take
  915.  
  916. 194
  917. 00:11:42,049 --> 00:11:43,964
  918. and the tweets that you send,
  919.  
  920. 195
  921. 00:11:44,007 --> 00:11:47,750
  922. we're all giving off huge
  923. amounts of data individually.
  924.  
  925. 196
  926. 00:11:47,794 --> 00:11:50,274
  927. Cars that drive as the cameras
  928. on them suck up information
  929.  
  930. 197
  931. 00:11:50,318 --> 00:11:52,059
  932. about the world around them.
  933.  
  934. 198
  935. 00:11:52,102 --> 00:11:54,844
  936. The satellites that are now
  937. in orbit the size of a toaster.
  938.  
  939. 199
  940. 00:11:54,888 --> 00:11:57,629
  941. The infrared about
  942. the vegetation on the planet.
  943.  
  944. 200
  945. 00:11:57,673 --> 00:12:01,024
  946. The buoys that are out in the oceans
  947. to feed into the climate models.
  948.  
  949. 201
  950. 00:12:05,072 --> 00:12:08,902
  951. And the NSA, the CIA,
  952. as they collect information
  953.  
  954. 202
  955. 00:12:08,945 --> 00:12:12,644
  956. about the
  957. geopolitical situations.
  958.  
  959. 203
  960. 00:12:12,688 --> 00:12:15,604
  961. The world today is literally
  962. swimming in this data.
  963.  
  964. 204
  965. 00:12:20,609 --> 00:12:22,480
  966. Back in 2012,
  967.  
  968. 205
  969. 00:12:22,524 --> 00:12:25,875
  970. IBM estimated
  971. that an average human being
  972.  
  973. 206
  974. 00:12:25,919 --> 00:12:31,098
  975. leaves 500 megabytes
  976. of digital footprints every day.
  977.  
  978. 207
  979. 00:12:31,141 --> 00:12:34,841
  980. If you wanted to back up
  981. on the one day worth of data
  982.  
  983. 208
  984. 00:12:34,884 --> 00:12:36,494
  985. that humanity produces
  986.  
  987. 209
  988. 00:12:36,538 --> 00:12:39,062
  989. and imprint it out
  990. on a letter-sized paper,
  991.  
  992. 210
  993. 00:12:39,106 --> 00:12:43,806
  994. double-sided, font size 12,
  995. and you stack it up,
  996.  
  997. 211
  998. 00:12:43,850 --> 00:12:46,113
  999. it would reach from
  1000. the surface of the Earth
  1001.  
  1002. 212
  1003. 00:12:46,156 --> 00:12:49,116
  1004. to the sun four times over.
  1005.  
  1006. 213
  1007. 00:12:49,159 --> 00:12:51,292
  1008. That's every day.
  1009.  
  1010. 214
  1011. 00:12:51,335 --> 00:12:53,816
  1012. The data itself
  1013. is not good or evil.
  1014.  
  1015. 215
  1016. 00:12:53,860 --> 00:12:55,470
  1017. It's how it's used.
  1018.  
  1019. 216
  1020. 00:12:55,513 --> 00:12:58,342
  1021. We're relying, really,
  1022. on the goodwill of these people
  1023.  
  1024. 217
  1025. 00:12:58,386 --> 00:13:01,171
  1026. and on the policies
  1027. of these companies.
  1028.  
  1029. 218
  1030. 00:13:01,215 --> 00:13:03,870
  1031. There is no legal requirement
  1032. for how they can
  1033.  
  1034. 219
  1035. 00:13:03,913 --> 00:13:06,307
  1036. and should use
  1037. that kind of data.
  1038.  
  1039. 220
  1040. 00:13:06,350 --> 00:13:09,266
  1041. That, to me, is at the heart
  1042. of the trust issue.
  1043.  
  1044. 221
  1045. 00:13:11,007 --> 00:13:13,793
  1046. Right now there's a
  1047. giant race for creating machines
  1048.  
  1049. 222
  1050. 00:13:13,836 --> 00:13:15,751
  1051. that are as smart as humans.
  1052.  
  1053. 223
  1054. 00:13:15,795 --> 00:13:18,071
  1055. Google -- They're working on
  1056. what's really the kind of
  1057.  
  1058. 224
  1059. 00:13:18,072 --> 00:13:20,074
  1060. Manhattan Project
  1061. of artificial intelligence.
  1062.  
  1063. 225
  1064. 00:13:20,075 --> 00:13:22,686
  1065. They've got the most money.
  1066. They've got the most talent.
  1067.  
  1068. 226
  1069. 00:13:22,714 --> 00:13:27,067
  1070. They're buying up AI companies
  1071. and robotics companies.
  1072.  
  1073. 227
  1074. 00:13:27,110 --> 00:13:29,069
  1075. People still think
  1076. of Google as a search engine
  1077.  
  1078. 228
  1079. 00:13:29,112 --> 00:13:30,722
  1080. and their e-mail provider
  1081.  
  1082. 229
  1083. 00:13:30,766 --> 00:13:33,943
  1084. and a lot of other things
  1085. that we use on a daily basis,
  1086.  
  1087. 230
  1088. 00:13:33,987 --> 00:13:39,383
  1089. but behind that search box
  1090. are 10 million servers.
  1091.  
  1092. 231
  1093. 00:13:39,427 --> 00:13:43,910
  1094. That makes Google the most powerful
  1095. computing platform in the world.
  1096.  
  1097. 232
  1098. 00:13:43,953 --> 00:13:47,217
  1099. Google is now working
  1100. on an AI computing platform
  1101.  
  1102. 233
  1103. 00:13:47,261 --> 00:13:50,133
  1104. that will have
  1105. 100 million servers.
  1106.  
  1107. 234
  1108. 00:13:52,179 --> 00:13:53,963
  1109. So when you're interacting
  1110. with Google,
  1111.  
  1112. 235
  1113. 00:13:54,007 --> 00:13:56,052
  1114. we're just seeing
  1115. the toenail of something
  1116.  
  1117. 236
  1118. 00:13:56,096 --> 00:13:58,881
  1119. that is a giant beast
  1120. in the making.
  1121.  
  1122. 237
  1123. 00:13:58,925 --> 00:14:00,622
  1124. And the truth is,
  1125. I'm not even sure
  1126.  
  1127. 238
  1128. 00:14:00,665 --> 00:14:02,798
  1129. that Google knows
  1130. what it's becoming.
  1131.  
  1132. 239
  1133. 00:14:11,546 --> 00:14:15,811
  1134. If you look inside of what algorithms
  1135. are being used at Google,
  1136.  
  1137. 240
  1138. 00:14:15,855 --> 00:14:20,076
  1139. it's technology
  1140. largely from the '80s.
  1141.  
  1142. 241
  1143. 00:14:20,120 --> 00:14:23,863
  1144. So these are models that you
  1145. train by showing them a 1, a 2,
  1146.  
  1147. 242
  1148. 00:14:23,906 --> 00:14:27,344
  1149. and a 3, and it learns not
  1150. what a 1 is or what a 2 is --
  1151.  
  1152. 243
  1153. 00:14:27,388 --> 00:14:30,434
  1154. It learns what the difference
  1155. between a 1 and a 2 is.
  1156.  
  1157. 244
  1158. 00:14:30,478 --> 00:14:32,436
  1159. It's just a computation.
  1160.  
  1161. 245
  1162. 00:14:32,480 --> 00:14:35,396
  1163. In the last half decade,
  1164. where we've made this rapid progress,
  1165.  
  1166. 246
  1167. 00:14:35,439 --> 00:14:38,268
  1168. it has all been
  1169. in pattern recognition.
  1170.  
  1171. 247
  1172. 00:14:38,312 --> 00:14:41,184
  1173. Most of the good, old-fashioned AI
  1174.  
  1175. 248
  1176. 00:14:41,228 --> 00:14:44,057
  1177. was when we would tell our computers
  1178.  
  1179. 249
  1180. 00:14:44,100 --> 00:14:46,798
  1181. how to play a game like chess...
  1182.  
  1183. 250
  1184. 00:14:46,842 --> 00:14:49,584
  1185. from the old paradigm where
  1186. you just tell the computer
  1187.  
  1188. 251
  1189. 00:14:49,627 --> 00:14:51,895
  1190. exactly what to do.
  1191.  
  1192. 252
  1193. 00:14:54,502 --> 00:14:57,505
  1194. This is "Jeopardy!"
  1195.  
  1196. 253
  1197. 00:14:59,420 --> 00:15:02,510
  1198. "The IBM Challenge"!
  1199.  
  1200. 254
  1201. 00:15:02,553 --> 00:15:05,730
  1202. No one at the time
  1203. had thought that a machine
  1204.  
  1205. 255
  1206. 00:15:05,774 --> 00:15:08,298
  1207. could have the precision
  1208. and the confidence
  1209.  
  1210. 256
  1211. 00:15:08,342 --> 00:15:11,475
  1212. and the speed to play "Jeopardy!"
  1213. well enough against the best humans.
  1214.  
  1215. 257
  1216. 00:15:11,519 --> 00:15:14,609
  1217. Let's play "Jeopardy!"
  1218.  
  1219. 258
  1220. 00:15:18,569 --> 00:15:20,354
  1221. What is "shoe"?
  1222.  
  1223. 259
  1224. 00:15:20,397 --> 00:15:21,877
  1225. You are right.
  1226. You get to pick.
  1227.  
  1228. 260
  1229. 00:15:21,921 --> 00:15:24,836
  1230. Literary Character APB
  1231. for $800.
  1232.  
  1233. 261
  1234. 00:15:24,880 --> 00:15:28,014
  1235. Answer --
  1236. the Daily Double.
  1237.  
  1238. 262
  1239. 00:15:28,057 --> 00:15:31,539
  1240. Watson actually got its
  1241. knowledge by reading Wikipedia
  1242.  
  1243. 263
  1244. 00:15:31,582 --> 00:15:34,672
  1245. and 200 million pages
  1246. of natural-language documents.
  1247.  
  1248. 264
  1249. 00:15:34,716 --> 00:15:36,674
  1250. You can't program every line
  1251.  
  1252. 265
  1253. 00:15:36,718 --> 00:15:38,502
  1254. of how the world works.
  1255.  
  1256. 266
  1257. 00:15:38,546 --> 00:15:40,722
  1258. The machine has to learn
  1259. by reading.
  1260.  
  1261. 267
  1262. 00:15:40,765 --> 00:15:42,202
  1263. Now we come to Watson.
  1264.  
  1265. 268
  1266. 00:15:42,245 --> 00:15:43,986
  1267. "Who is Bram Stoker?"
  1268.  
  1269. 269
  1270. 00:15:44,030 --> 00:15:45,988
  1271. And the wager?
  1272.  
  1273. 270
  1274. 00:15:46,032 --> 00:15:49,165
  1275. Hello! $17,973.
  1276.  
  1277. 271
  1278. 00:15:49,209 --> 00:15:50,993
  1279. $41,413.
  1280.  
  1281. 272
  1282. 00:15:51,037 --> 00:15:53,343
  1283. And a two-day total
  1284. of $77--
  1285.  
  1286. 273
  1287. 00:15:53,387 --> 00:15:56,694
  1288. Watson's trained
  1289. on huge amounts of text,
  1290.  
  1291. 274
  1292. 00:15:56,738 --> 00:15:59,628
  1293. but it's not like it
  1294. understands what it's saying.
  1295.  
  1296. 275
  1297. 00:15:59,671 --> 00:16:02,309
  1298. It doesn't know that water makes
  1299. things wet by touching water
  1300.  
  1301. 276
  1302. 00:16:02,352 --> 00:16:04,441
  1303. and by seeing the way
  1304. things behave in the world
  1305.  
  1306. 277
  1307. 00:16:04,485 --> 00:16:06,182
  1308. the way you and I do.
  1309.  
  1310. 278
  1311. 00:16:06,226 --> 00:16:10,143
  1312. A lot of language AI today
  1313. is not building logical models
  1314.  
  1315. 279
  1316. 00:16:10,186 --> 00:16:11,622
  1317. of how the world works.
  1318.  
  1319. 280
  1320. 00:16:11,666 --> 00:16:15,365
  1321. Rather, it's looking at
  1322. how the words appear
  1323.  
  1324. 281
  1325. 00:16:15,409 --> 00:16:18,238
  1326. in the context of other words.
  1327.  
  1328. 282
  1329. 00:16:18,281 --> 00:16:20,196
  1330. David Ferrucci
  1331. developed IBM's Watson,
  1332.  
  1333. 283
  1334. 00:16:20,240 --> 00:16:23,547
  1335. and somebody asked him,
  1336. "Does Watson think?"
  1337.  
  1338. 284
  1339. 00:16:23,591 --> 00:16:26,660
  1340. And he said,
  1341. "Does a submarine swim?"
  1342.  
  1343. 285
  1344. 00:16:26,903 --> 00:16:29,331
  1345. And what they meant was,
  1346. when they developed submarines,
  1347.  
  1348. 286
  1349. 00:16:29,332 --> 00:16:32,949
  1350. they borrowed basic principles
  1351. of swimming from fish.
  1352.  
  1353. 287
  1354. 00:16:33,035 --> 00:16:36,525
  1355. But a submarine swims farther and faster
  1356. than fish and can carry a huge payload.
  1357.  
  1358. 288
  1359. 00:16:36,569 --> 00:16:39,411
  1360. It out-swims fish.
  1361.  
  1362. 289
  1363. 00:16:39,955 --> 00:16:43,741
  1364. Watson winning the game of "Jeopardy!"
  1365. will go down in the history of AI
  1366.  
  1367. 290
  1368. 00:16:43,785 --> 00:16:46,370
  1369. as a significant milestone.
  1370.  
  1371. 291
  1372. 00:16:46,614 --> 00:16:49,269
  1373. We tend to be amazed
  1374. when the machine does so well.
  1375.  
  1376. 292
  1377. 00:16:49,312 --> 00:16:52,663
  1378. I'm even more amazed when the
  1379. computer beats humans at things
  1380.  
  1381. 293
  1382. 00:16:52,707 --> 00:16:55,188
  1383. that humans are
  1384. naturally good at.
  1385.  
  1386. 294
  1387. 00:16:55,231 --> 00:16:58,060
  1388. This is how we make progress.
  1389.  
  1390. 295
  1391. 00:16:58,104 --> 00:17:00,671
  1392. In the early days of
  1393. the Google Brain project,
  1394.  
  1395. 296
  1396. 00:17:00,715 --> 00:17:02,804
  1397. I gave the team a very
  1398. simple instruction,
  1399.  
  1400. 297
  1401. 00:17:02,847 --> 00:17:05,807
  1402. which was, "Build the biggest
  1403. neural network possible,
  1404.  
  1405. 298
  1406. 00:17:05,850 --> 00:17:08,157
  1407. like 1,000 computers."
  1408.  
  1409. 299
  1410. 00:17:08,201 --> 00:17:12,161
  1411. A neural net is something very close
  1412. to a simulation of how the brain works.
  1413.  
  1414. 300
  1415. 00:17:12,205 --> 00:17:16,818
  1416. It's very probabilistic,
  1417. but with contextual relevance.
  1418.  
  1419. 301
  1420. 00:17:16,819 --> 00:17:18,456
  1421. In your brain,
  1422. you have long neurons
  1423.  
  1424. 302
  1425. 00:17:18,457 --> 00:17:20,372
  1426. that connect to thousands
  1427. of other neurons,
  1428.  
  1429. 303
  1430. 00:17:20,373 --> 00:17:22,592
  1431. and you have these pathways
  1432. that are formed and forged
  1433.  
  1434. 304
  1435. 00:17:22,593 --> 00:17:24,769
  1436. based on what
  1437. the brain needs to do.
  1438.  
  1439. 305
  1440. 00:17:24,782 --> 00:17:28,960
  1441. When a baby tries something and
  1442. it succeeds, there's a reward,
  1443.  
  1444. 306
  1445. 00:17:29,004 --> 00:17:32,312
  1446. and that pathway that created
  1447. the success is strengthened.
  1448.  
  1449. 307
  1450. 00:17:32,355 --> 00:17:34,662
  1451. If it fails at something,
  1452. the pathway is weakened,
  1453.  
  1454. 308
  1455. 00:17:34,705 --> 00:17:36,794
  1456. and so, over time,
  1457. the brain becomes honed
  1458.  
  1459. 309
  1460. 00:17:36,838 --> 00:17:40,120
  1461. to be good at
  1462. the environment around it.
  1463.  
  1464. 310
  1465. 00:17:40,363 --> 00:17:43,279
  1466. Really, it's just getting
  1467. machines to learn by themselves.
  1468.  
  1469. 311
  1470. 00:17:43,323 --> 00:17:45,538
  1471. This is called "deep learning,"
  1472. and "deep learning"
  1473.  
  1474. 312
  1475. 00:17:45,539 --> 00:17:48,834
  1476. and "neural networks"
  1477. mean roughly the same thing.
  1478.  
  1479. 313
  1480. 00:17:48,835 --> 00:17:52,391
  1481. Deep learning
  1482. is a totally different approach
  1483.  
  1484. 314
  1485. 00:17:52,419 --> 00:17:55,161
  1486. where the computer learns
  1487. more like a toddler,
  1488.  
  1489. 315
  1490. 00:17:55,204 --> 00:17:56,466
  1491. by just getting a lot of data
  1492.  
  1493. 316
  1494. 00:17:56,510 --> 00:18:00,340
  1495. and eventually
  1496. figuring stuff out.
  1497.  
  1498. 317
  1499. 00:18:00,383 --> 00:18:03,125
  1500. The computer just gets
  1501. smarter and smarter
  1502.  
  1503. 318
  1504. 00:18:03,169 --> 00:18:05,997
  1505. as it has more experiences.
  1506.  
  1507. 319
  1508. 00:18:06,041 --> 00:18:09,697
  1509. Imagine, if you will, a neural network,
  1510. you know, like 1,000 computers.
  1511.  
  1512. 320
  1513. 00:18:09,740 --> 00:18:11,438
  1514. And it wakes up
  1515. not knowing anything.
  1516.  
  1517. 321
  1518. 00:18:11,481 --> 00:18:14,093
  1519. And we made it watch YouTube
  1520. for a week.
  1521.  
  1522. 322
  1523. 00:18:25,408 --> 00:18:28,194
  1524. Charlie!
  1525. That really hurt!
  1526.  
  1527. 323
  1528. 00:18:36,245 --> 00:18:38,508
  1529. And so, after watching
  1530. YouTube for a week,
  1531.  
  1532. 324
  1533. 00:18:38,552 --> 00:18:39,988
  1534. what would it learn?
  1535.  
  1536. 325
  1537. 00:18:40,031 --> 00:18:42,103
  1538. We had a hypothesis that
  1539. it would learn to detect
  1540.  
  1541. 326
  1542. 00:18:42,146 --> 00:18:44,384
  1543. commonly occurring objects
  1544. in videos.
  1545.  
  1546. 327
  1547. 00:18:44,427 --> 00:18:47,517
  1548. And so, we know that human faces
  1549. appear a lot in videos,
  1550.  
  1551. 328
  1552. 00:18:47,561 --> 00:18:49,302
  1553. so we looked,
  1554. and, lo and behold,
  1555.  
  1556. 329
  1557. 00:18:49,345 --> 00:18:52,008
  1558. there was a neuron that had
  1559. learned to detect human faces.
  1560.  
  1561. 330
  1562. 00:18:52,052 --> 00:18:55,865
  1563. Leave Britney alone!
  1564.  
  1565. 331
  1566. 00:18:56,309 --> 00:18:58,354
  1567. Well, what else
  1568. appears in videos a lot?
  1569.  
  1570. 332
  1571. 00:19:00,095 --> 00:19:01,792
  1572. So, we looked,
  1573. and to our surprise,
  1574.  
  1575. 333
  1576. 00:19:01,836 --> 00:19:05,082
  1577. there was actually a neuron
  1578. that had learned to detect cats.
  1579.  
  1580. 334
  1581. 00:19:14,892 --> 00:19:17,068
  1582. I still remember seeing recognition.
  1583.  
  1584. 335
  1585. 00:19:17,112 --> 00:19:20,071
  1586. "Wow, that's a cat. Okay, cool.
  1587. Great."
  1588.  
  1589. 336
  1590. 00:19:23,162 --> 00:19:26,295
  1591. It's all pretty innocuous when
  1592. you're thinking about the future.
  1593.  
  1594. 337
  1595. 00:19:26,339 --> 00:19:29,733
  1596. It all seems kind of
  1597. harmless and benign.
  1598.  
  1599. 338
  1600. 00:19:29,777 --> 00:19:33,520
  1601. But we're making cognitive architectures
  1602. that will fly farther and faster than us
  1603.  
  1604. 339
  1605. 00:19:33,563 --> 00:19:37,437
  1606. and carry a bigger payload,
  1607. and they won't be warm and fuzzy.
  1608.  
  1609. 340
  1610. 00:19:37,480 --> 00:19:41,702
  1611. I think that, in three to five years,
  1612. you will see a computer system
  1613.  
  1614. 341
  1615. 00:19:41,745 --> 00:19:45,401
  1616. that will be able
  1617. to autonomously learn
  1618.  
  1619. 342
  1620. 00:19:45,445 --> 00:19:49,013
  1621. how to understand,
  1622. how to build understanding,
  1623.  
  1624. 343
  1625. 00:19:49,057 --> 00:19:51,364
  1626. not unlike the way
  1627. the human mind works.
  1628.  
  1629. 344
  1630. 00:19:53,931 --> 00:19:56,891
  1631. Whatever that lunch was,
  1632. it was certainly delicious.
  1633.  
  1634. 345
  1635. 00:19:56,934 --> 00:19:59,807
  1636. Simply some of
  1637. Robby's synthetics.
  1638.  
  1639. 346
  1640. 00:19:59,850 --> 00:20:01,635
  1641. He's your cook, too?
  1642.  
  1643. 347
  1644. 00:20:01,678 --> 00:20:04,551
  1645. Even manufactures
  1646. the raw materials.
  1647.  
  1648. 348
  1649. 00:20:04,594 --> 00:20:06,944
  1650. Come around here, Robby.
  1651.  
  1652. 349
  1653. 00:20:06,988 --> 00:20:09,773
  1654. I'll show you
  1655. how this works.
  1656.  
  1657. 350
  1658. 00:20:11,122 --> 00:20:13,342
  1659. One introduces
  1660. a sample of human food
  1661.  
  1662. 351
  1663. 00:20:13,386 --> 00:20:15,344
  1664. through this aperture.
  1665.  
  1666. 352
  1667. 00:20:15,388 --> 00:20:17,738
  1668. Down here there's a small
  1669. built-in chemical laboratory,
  1670.  
  1671. 353
  1672. 00:20:17,781 --> 00:20:19,218
  1673. where he analyzes it.
  1674.  
  1675. 354
  1676. 00:20:19,261 --> 00:20:21,263
  1677. Later, he can reproduce
  1678. identical molecules
  1679.  
  1680. 355
  1681. 00:20:21,307 --> 00:20:22,482
  1682. in any shape or quantity.
  1683.  
  1684. 356
  1685. 00:20:22,525 --> 00:20:24,614
  1686. Why, it's a housewife's dream.
  1687.  
  1688. 357
  1689. 00:20:24,958 --> 00:20:26,834
  1690. Meet Baxter,
  1691.  
  1692. 358
  1693. 00:20:26,877 --> 00:20:30,490
  1694. revolutionary new category of robots,
  1695. with common sense.
  1696.  
  1697. 359
  1698. 00:20:30,533 --> 00:20:31,839
  1699. Baxter...
  1700.  
  1701. 360
  1702. 00:20:31,882 --> 00:20:33,449
  1703. Baxter is
  1704. a really good example
  1705.  
  1706. 361
  1707. 00:20:33,493 --> 00:20:36,887
  1708. of the kind of competition
  1709. we face from machines.
  1710.  
  1711. 362
  1712. 00:20:36,931 --> 00:20:42,676
  1713. Baxter can do almost anything
  1714. we can do with our hands.
  1715.  
  1716. 363
  1717. 00:20:42,719 --> 00:20:45,722
  1718. Baxter costs about
  1719. what a minimum-wage worker
  1720.  
  1721. 364
  1722. 00:20:45,766 --> 00:20:47,507
  1723. makes in a year.
  1724.  
  1725. 365
  1726. 00:20:47,550 --> 00:20:50,318
  1727. But Baxter won't be taking the place
  1728. of one minimum-wage worker --
  1729.  
  1730. 366
  1731. 00:20:50,319 --> 00:20:51,930
  1732. He'll be taking
  1733. the place of three,
  1734.  
  1735. 367
  1736. 00:20:51,931 --> 00:20:55,531
  1737. because they never get tired,
  1738. they never take breaks.
  1739.  
  1740. 368
  1741. 00:20:55,558 --> 00:20:57,865
  1742. That's probably the
  1743. first thing we're gonna see --
  1744.  
  1745. 369
  1746. 00:20:57,908 --> 00:20:59,475
  1747. displacement of jobs.
  1748.  
  1749. 370
  1750. 00:20:59,519 --> 00:21:04,088
  1751. They're gonna be done quicker,
  1752. faster, cheaper by machines.
  1753.  
  1754. 371
  1755. 00:21:04,132 --> 00:21:07,657
  1756. Our ability to even stay current
  1757. is so insanely limited
  1758.  
  1759. 372
  1760. 00:21:07,701 --> 00:21:10,138
  1761. compared to
  1762. the machines we build.
  1763.  
  1764. 373
  1765. 00:21:10,181 --> 00:21:13,446
  1766. For example, now we have this
  1767. great movement of Uber and Lyft
  1768.  
  1769. 374
  1770. 00:21:13,489 --> 00:21:16,505
  1771. kind of making transportation cheaper
  1772. and democratizing transportation,
  1773.  
  1774. 375
  1775. 00:21:16,506 --> 00:21:17,768
  1776. which is great.
  1777.  
  1778. 376
  1779. 00:21:17,769 --> 00:21:21,189
  1780. The next step is gonna be that they're
  1781. all gonna be replaced by driverless cars
  1782.  
  1783. 377
  1784. 00:21:21,192 --> 00:21:25,936
  1785. and then all the Uber and Lyft drivers
  1786. have to find something new to do.
  1787.  
  1788. 378
  1789. 00:21:25,980 --> 00:21:29,723
  1790. There are 4 million professional drivers
  1791. in the United States.
  1792.  
  1793. 379
  1794. 00:21:29,766 --> 00:21:31,638
  1795. They're unemployed soon.
  1796.  
  1797. 380
  1798. 00:21:31,681 --> 00:21:34,075
  1799. 7 million people that do data entry.
  1800.  
  1801. 381
  1802. 00:21:34,118 --> 00:21:37,339
  1803. Those people are gonna be jobless.
  1804.  
  1805. 382
  1806. 00:21:37,383 --> 00:21:40,342
  1807. A job isn't just about money, right?
  1808.  
  1809. 383
  1810. 00:21:40,386 --> 00:21:42,605
  1811. On a biological level,
  1812. it serves a purpose.
  1813.  
  1814. 384
  1815. 00:21:42,649 --> 00:21:45,391
  1816. It becomes a defining thing.
  1817.  
  1818. 385
  1819. 00:21:45,434 --> 00:21:48,350
  1820. When the jobs went away
  1821. in any given civilization,
  1822.  
  1823. 386
  1824. 00:21:48,394 --> 00:21:50,987
  1825. it doesn't take long
  1826. until that turns into violence.
  1827.  
  1828. 387
  1829. 00:21:59,622 --> 00:22:02,016
  1830. We face a giant divide
  1831. between rich and poor,
  1832.  
  1833. 388
  1834. 00:22:02,059 --> 00:22:05,019
  1835. because that's what automation
  1836. and AI will provoke --
  1837.  
  1838. 389
  1839. 00:22:05,062 --> 00:22:08,588
  1840. a greater divide between
  1841. the haves and the have-nots.
  1842.  
  1843. 390
  1844. 00:22:08,631 --> 00:22:10,807
  1845. Right now, it's working
  1846. into the middle class,
  1847.  
  1848. 391
  1849. 00:22:10,851 --> 00:22:12,896
  1850. into white-collar jobs.
  1851.  
  1852. 392
  1853. 00:22:12,940 --> 00:22:15,334
  1854. IBM's Watson does
  1855. business analytics
  1856.  
  1857. 393
  1858. 00:22:15,377 --> 00:22:20,600
  1859. that we used to pay a business analyst
  1860. $300 an hour to do.
  1861.  
  1862. 394
  1863. 00:22:20,643 --> 00:22:23,037
  1864. Today, you're going
  1865. to college to be a doctor,
  1866.  
  1867. 395
  1868. 00:22:23,080 --> 00:22:25,082
  1869. to be an accountant,
  1870. to be a journalist.
  1871.  
  1872. 396
  1873. 00:22:25,126 --> 00:22:28,608
  1874. It's unclear that there's
  1875. gonna be jobs there for you.
  1876.  
  1877. 397
  1878. 00:22:28,651 --> 00:22:32,612
  1879. If someone's planning for
  1880. a 40-year career in radiology,
  1881.  
  1882. 398
  1883. 00:22:32,655 --> 00:22:34,222
  1884. just reading images,
  1885.  
  1886. 399
  1887. 00:22:34,265 --> 00:22:37,120
  1888. I think that could be a challenge
  1889. to the new graduates of today.
  1890.  
  1891. 400
  1892. 00:22:58,507 --> 00:23:02,729
  1893. The da Vinci robot is currently utilized
  1894.  
  1895. 401
  1896. 00:23:02,772 --> 00:23:07,516
  1897. by a variety of surgeons
  1898. for its accuracy and its ability
  1899.  
  1900. 402
  1901. 00:23:07,560 --> 00:23:12,303
  1902. to avoid the inevitable
  1903. fluctuations of the human hand.
  1904.  
  1905. 403
  1906. 00:23:23,402 --> 00:23:28,494
  1907. Anybody who watches this
  1908. feels the amazingness of it.
  1909.  
  1910. 404
  1911. 00:23:30,931 --> 00:23:34,674
  1912. You look through the scope,
  1913. and you're seeing the claw hand
  1914.  
  1915. 405
  1916. 00:23:34,717 --> 00:23:36,893
  1917. holding that woman's ovary.
  1918.  
  1919. 406
  1920. 00:23:36,937 --> 00:23:42,638
  1921. Humanity was resting right here
  1922. in the hands of this robot.
  1923.  
  1924. 407
  1925. 00:23:42,682 --> 00:23:46,947
  1926. People say it's the future,
  1927. but it's not the future --
  1928.  
  1929. 408
  1930. 00:23:46,990 --> 00:23:50,516
  1931. It's the present.
  1932.  
  1933. 409
  1934. 00:23:50,559 --> 00:23:52,474
  1935. If you think about
  1936. a surgical robot,
  1937.  
  1938. 410
  1939. 00:23:52,475 --> 00:23:54,894
  1940. there's often not a lot
  1941. of intelligence in these things,
  1942.  
  1943. 411
  1944. 00:23:54,895 --> 00:23:58,567
  1945. but over time, as we put more and more
  1946. intelligence into these systems,
  1947.  
  1948. 412
  1949. 00:23:58,611 --> 00:24:02,281
  1950. the surgical robots can actually
  1951. learn from each robot surgery.
  1952.  
  1953. 413
  1954. 00:24:02,284 --> 00:24:04,581
  1955. They're tracking the movements,
  1956. they're understanding
  1957.  
  1958. 414
  1959. 00:24:04,582 --> 00:24:06,423
  1960. what worked
  1961. and what didn't work.
  1962.  
  1963. 415
  1964. 00:24:06,424 --> 00:24:09,023
  1965. And eventually, the robot
  1966. for routine surgeries
  1967.  
  1968. 416
  1969. 00:24:09,024 --> 00:24:12,362
  1970. is going to be able to perform
  1971. that entirely by itself...
  1972.  
  1973. 417
  1974. 00:24:12,363 --> 00:24:14,056
  1975. or with human supervision.
  1976.  
  1977. 418
  1978. 00:24:35,038 --> 00:24:37,214
  1979. It seems that we're
  1980. feeding it and creating it,
  1981.  
  1982. 419
  1983. 00:24:37,258 --> 00:24:42,785
  1984. but, in a way, we are a slave
  1985. to the technology,
  1986.  
  1987. 420
  1988. 00:24:42,829 --> 00:24:45,701
  1989. because we can't go back.
  1990.  
  1991. 421
  1992. 00:24:50,053 --> 00:24:52,882
  1993. The machines are taking
  1994. bigger and bigger bites
  1995.  
  1996. 422
  1997. 00:24:52,926 --> 00:24:57,147
  1998. out of our skill set
  1999. at an ever-increasing speed.
  2000.  
  2001. 423
  2002. 00:24:57,191 --> 00:24:59,236
  2003. And so we've got to run
  2004. faster and faster
  2005.  
  2006. 424
  2007. 00:24:59,280 --> 00:25:00,890
  2008. to keep ahead of the machines.
  2009.  
  2010. 425
  2011. 00:25:02,675 --> 00:25:04,677
  2012. How do I look?
  2013.  
  2014. 426
  2015. 00:25:04,720 --> 00:25:06,374
  2016. Good.
  2017.  
  2018. 427
  2019. 00:25:10,030 --> 00:25:11,553
  2020. Are you attracted to me?
  2021.  
  2022. 428
  2023. 00:25:11,597 --> 00:25:14,251
  2024. What?
  2025. - Are you attracted to me?
  2026.  
  2027. 429
  2028. 00:25:14,295 --> 00:25:17,777
  2029. You give me indications
  2030. that you are.
  2031.  
  2032. 430
  2033. 00:25:17,820 --> 00:25:20,562
  2034. I do?
  2035. - Yes.
  2036.  
  2037. 431
  2038. 00:25:20,606 --> 00:25:22,608
  2039. This is the future we're headed into.
  2040.  
  2041. 432
  2042. 00:25:22,651 --> 00:25:26,046
  2043. We want to design
  2044. our companions.
  2045.  
  2046. 433
  2047. 00:25:26,089 --> 00:25:29,266
  2048. We're gonna like to see
  2049. a human face on AI.
  2050.  
  2051. 434
  2052. 00:25:29,310 --> 00:25:33,967
  2053. Therefore, gaming our emotions
  2054. will be depressingly easy.
  2055.  
  2056. 435
  2057. 00:25:34,010 --> 00:25:35,272
  2058. We're not that complicated.
  2059.  
  2060. 436
  2061. 00:25:35,316 --> 00:25:38,101
  2062. We're simple.
  2063. Stimulus-response.
  2064.  
  2065. 437
  2066. 00:25:38,145 --> 00:25:42,763
  2067. I can make you like me basically
  2068. by smiling at you a lot.
  2069.  
  2070. 438
  2071. 00:25:43,106 --> 00:25:45,974
  2072. AI's are gonna be fantastic
  2073. at manipulating us.
  2074.  
  2075. 439
  2076. 00:25:54,683 --> 00:25:56,946
  2077. So, you've developed a technology
  2078.  
  2079. 440
  2080. 00:25:56,990 --> 00:26:00,036
  2081. that can sense
  2082. what people are feeling.
  2083.  
  2084. 441
  2085. 00:26:00,080 --> 00:26:03,387
  2086. Right. We've developed technology
  2087. that can read your facial expressions
  2088.  
  2089. 442
  2090. 00:26:03,431 --> 00:26:06,521
  2091. and map that to a number
  2092. of emotional states.
  2093.  
  2094. 443
  2095. 00:26:06,565 --> 00:26:08,697
  2096. 15 years ago,
  2097. I had just finished
  2098.  
  2099. 444
  2100. 00:26:08,741 --> 00:26:11,482
  2101. my undergraduate studies
  2102. in computer science,
  2103.  
  2104. 445
  2105. 00:26:11,526 --> 00:26:15,008
  2106. and it struck me that I was
  2107. spending a lot of time
  2108.  
  2109. 446
  2110. 00:26:15,051 --> 00:26:17,793
  2111. interacting with my laptops
  2112. and my devices,
  2113.  
  2114. 447
  2115. 00:26:17,837 --> 00:26:23,582
  2116. yet these devices had absolutely
  2117. no clue how I was feeling.
  2118.  
  2119. 448
  2120. 00:26:23,625 --> 00:26:26,802
  2121. I started thinking,
  2122. "What if this device could sense
  2123.  
  2124. 449
  2125. 00:26:26,846 --> 00:26:29,326
  2126. that I was stressed
  2127. or I was having a bad day?
  2128.  
  2129. 450
  2130. 00:26:29,370 --> 00:26:31,067
  2131. What would that open up?"
  2132.  
  2133. 451
  2134. 00:26:32,721 --> 00:26:34,418
  2135. Hi, first-graders!
  2136.  
  2137. 452
  2138. 00:26:34,462 --> 00:26:35,855
  2139. How are you?
  2140.  
  2141. 453
  2142. 00:26:35,898 --> 00:26:37,813
  2143. Can I get a hug?
  2144.  
  2145. 454
  2146. 00:26:37,857 --> 00:26:40,773
  2147. We had kids interact
  2148. with the technology.
  2149.  
  2150. 455
  2151. 00:26:40,816 --> 00:26:44,472
  2152. A lot of it is still in development,
  2153. but it was just amazing.
  2154.  
  2155. 456
  2156. 00:26:44,515 --> 00:26:46,648
  2157. Who likes robots?
  2158. - Me!
  2159.  
  2160. 457
  2161. 00:26:46,692 --> 00:26:48,911
  2162. Who wants to have a robot
  2163. in their house?
  2164.  
  2165. 458
  2166. 00:26:48,955 --> 00:26:51,479
  2167. What would you use
  2168. a robot for, Jack?
  2169.  
  2170. 459
  2171. 00:26:51,522 --> 00:26:56,353
  2172. I would use it to ask my mom
  2173. very hard math questions.
  2174.  
  2175. 460
  2176. 00:26:56,397 --> 00:26:58,181
  2177. Okay.
  2178. What about you, Theo?
  2179.  
  2180. 461
  2181. 00:26:58,225 --> 00:27:02,272
  2182. I would use it
  2183. for scaring people.
  2184.  
  2185. 462
  2186. 00:27:02,316 --> 00:27:04,666
  2187. All right.
  2188. So, start by smiling.
  2189.  
  2190. 463
  2191. 00:27:04,710 --> 00:27:06,625
  2192. Nice.
  2193.  
  2194. 464
  2195. 00:27:06,668 --> 00:27:09,018
  2196. Brow furrow.
  2197.  
  2198. 465
  2199. 00:27:09,062 --> 00:27:10,890
  2200. Nice one.
  2201. Eyebrow raise.
  2202.  
  2203. 466
  2204. 00:27:10,933 --> 00:27:15,068
  2205. This generation, technology is just
  2206. surrounding them all the time.
  2207.  
  2208. 467
  2209. 00:27:15,111 --> 00:27:17,853
  2210. It's almost like they expect
  2211. to have robots in their homes,
  2212.  
  2213. 468
  2214. 00:27:17,897 --> 00:27:22,336
  2215. and they expect these robots
  2216. to be socially intelligent.
  2217.  
  2218. 469
  2219. 00:27:22,379 --> 00:27:25,252
  2220. What makes robots smart?
  2221.  
  2222. 470
  2223. 00:27:25,295 --> 00:27:29,648
  2224. Put them in, like, a math
  2225. or biology class.
  2226.  
  2227. 471
  2228. 00:27:29,691 --> 00:27:32,259
  2229. I think you would
  2230. have to train it.
  2231.  
  2232. 472
  2233. 00:27:32,302 --> 00:27:35,218
  2234. All right.
  2235. Let's walk over here.
  2236.  
  2237. 473
  2238. 00:27:35,262 --> 00:27:37,394
  2239. So, if you smile and you
  2240. raise your eyebrows,
  2241.  
  2242. 474
  2243. 00:27:37,438 --> 00:27:39,005
  2244. it's gonna run over to you.
  2245.  
  2246. 475
  2247. 00:27:39,048 --> 00:27:40,833
  2248. It's coming over!
  2249. It's coming over! Look.
  2250.  
  2251. 476
  2252. 00:27:43,183 --> 00:27:45,272
  2253. But if you look angry,
  2254. it's gonna run away.
  2255.  
  2256. 477
  2257. 00:27:46,534 --> 00:27:48,797
  2258. -Awesome!
  2259. -Oh, that was good.
  2260.  
  2261. 478
  2262. 00:27:48,841 --> 00:27:52,366
  2263. We're training computers to read
  2264. and recognize emotions.
  2265.  
  2266. 479
  2267. 00:27:52,409 --> 00:27:53,846
  2268. Ready? Set? Go!
  2269.  
  2270. 480
  2271. 00:27:53,889 --> 00:27:57,414
  2272. And the response so far
  2273. has been really amazing.
  2274.  
  2275. 481
  2276. 00:27:57,458 --> 00:27:59,590
  2277. People are integrating this
  2278. into health apps,
  2279.  
  2280. 482
  2281. 00:27:59,634 --> 00:28:03,865
  2282. meditation apps, robots, cars.
  2283.  
  2284. 483
  2285. 00:28:04,508 --> 00:28:06,728
  2286. We're gonna see
  2287. how this unfolds.
  2288.  
  2289. 484
  2290. 00:28:09,470 --> 00:28:11,602
  2291. Robots can contain AI,
  2292.  
  2293. 485
  2294. 00:28:11,646 --> 00:28:14,388
  2295. but the robot is just
  2296. a physical instantiation,
  2297.  
  2298. 486
  2299. 00:28:14,431 --> 00:28:16,782
  2300. and the artificial intelligence
  2301. is the brain.
  2302.  
  2303. 487
  2304. 00:28:16,825 --> 00:28:19,872
  2305. And so brains can exist purely
  2306. in software-based systems.
  2307.  
  2308. 488
  2309. 00:28:19,915 --> 00:28:22,483
  2310. They don't need to have
  2311. a physical form.
  2312.  
  2313. 489
  2314. 00:28:22,526 --> 00:28:25,094
  2315. Robots can exist without
  2316. any artificial intelligence.
  2317.  
  2318. 490
  2319. 00:28:25,138 --> 00:28:28,097
  2320. We have a lot of
  2321. dumb robots out there.
  2322.  
  2323. 491
  2324. 00:28:28,141 --> 00:28:31,753
  2325. But a dumb robot can be
  2326. a smart robot overnight,
  2327.  
  2328. 492
  2329. 00:28:31,797 --> 00:28:34,103
  2330. given the right software,
  2331. given the right sensors.
  2332.  
  2333. 493
  2334. 00:28:34,147 --> 00:28:38,629
  2335. We can't help but impute
  2336. motive into inanimate objects.
  2337.  
  2338. 494
  2339. 00:28:38,673 --> 00:28:41,502
  2340. We do it with machines.
  2341. We'll treat them like children.
  2342.  
  2343. 495
  2344. 00:28:41,545 --> 00:28:43,330
  2345. We'll treat them like surrogates.
  2346.  
  2347. 496
  2348. 00:28:43,373 --> 00:28:45,027
  2349. Goodbye!
  2350.  
  2351. 497
  2352. 00:28:45,071 --> 00:28:48,204
  2353. And we'll pay the price.
  2354.  
  2355. 498
  2356. 00:29:08,616 --> 00:29:10,792
  2357. Okay, welcome to the ATR.
  2358.  
  2359. 499
  2360. 00:29:18,000 --> 00:29:20,800
  2361. My purpose is to have
  2362. a more human-like robot
  2363.  
  2364. 500
  2365. 00:29:20,801 --> 00:29:24,001
  2366. which has human-like
  2367. intentions and desires.
  2368.  
  2369. 501
  2370. 00:29:36,000 --> 00:29:38,400
  2371. The name of the robot is Erica.
  2372.  
  2373. 502
  2374. 00:29:39,501 --> 00:29:43,901
  2375. Erica is the most advanced
  2376. human-like robot in the world, I think.
  2377.  
  2378. 503
  2379. 00:29:44,202 --> 00:29:47,202
  2380. Erica can gaze at your face.
  2381.  
  2382. 504
  2383. 00:29:51,528 --> 00:29:52,791
  2384. Konnichiwa.
  2385.  
  2386. 505
  2387. 00:29:53,592 --> 00:29:57,092
  2388. Robots can be pretty good
  2389. as conversation partners,
  2390.  
  2391. 506
  2392. 00:29:57,093 --> 00:30:00,493
  2393. especially for the elderly
  2394. and younger children,
  2395.  
  2396. 507
  2397. 00:30:00,494 --> 00:30:02,494
  2398. handicapped people.
  2399.  
  2400. 508
  2401. 00:30:03,094 --> 00:30:06,294
  2402. When we talk to the robot
  2403. we don't fear the social barriers,
  2404.  
  2405. 509
  2406. 00:30:06,295 --> 00:30:08,095
  2407. social pressures.
  2408.  
  2409. 510
  2410. 00:30:08,396 --> 00:30:15,796
  2411. Finally everybody accepts the android
  2412. as just our friend or partner.
  2413.  
  2414. 511
  2415. 00:30:15,997 --> 00:30:18,997
  2416. We have implemented simple desires.
  2417.  
  2418. 512
  2419. 00:30:18,998 --> 00:30:22,798
  2420. She wanted to be well recognized
  2421. and she wanted to go rest.
  2422.  
  2423. 513
  2424. 00:30:29,299 --> 00:30:32,299
  2425. If a robot could have
  2426. intentions and desires,
  2427.  
  2428. 514
  2429. 00:30:32,300 --> 00:30:36,300
  2430. the robot can understand other people's
  2431. intentions and desires.
  2432.  
  2433. 515
  2434. 00:30:44,300 --> 00:30:47,100
  2435. That is tied to
  2436. relationships with people
  2437.  
  2438. 516
  2439. 00:30:47,101 --> 00:30:49,201
  2440. and that means
  2441. they like eachother.
  2442.  
  2443. 517
  2444. 00:30:50,302 --> 00:30:53,302
  2445. That means, well, I'm not sure,
  2446. to rub eachother.
  2447.  
  2448. 518
  2449. 00:30:56,985 --> 00:30:58,682
  2450. We build artificial intelligence,
  2451.  
  2452. 519
  2453. 00:30:58,726 --> 00:31:01,948
  2454. and the very first thing
  2455. we want to do is replicate us.
  2456.  
  2457. 520
  2458. 00:31:02,991 --> 00:31:05,341
  2459. I think the key point will come
  2460.  
  2461. 521
  2462. 00:31:05,385 --> 00:31:08,858
  2463. when all the major senses
  2464. are replicated.
  2465.  
  2466. 522
  2467. 00:31:09,302 --> 00:31:11,130
  2468. Sight...
  2469.  
  2470. 523
  2471. 00:31:11,173 --> 00:31:12,871
  2472. touch...
  2473.  
  2474. 524
  2475. 00:31:12,914 --> 00:31:14,611
  2476. smell.
  2477.  
  2478. 525
  2479. 00:31:14,655 --> 00:31:17,919
  2480. When we replicate our senses,
  2481. is that when it becomes alive?
  2482.  
  2483. 526
  2484. 00:31:27,624 --> 00:31:31,019
  2485. So many of our machines
  2486. are being built to understand us.
  2487.  
  2488. 527
  2489. 00:31:32,847 --> 00:31:35,005
  2490. But what happens when
  2491. an anthropomorphic creature
  2492.  
  2493. 528
  2494. 00:31:35,006 --> 00:31:37,474
  2495. discovers that they can
  2496. adjust their loyalty,
  2497.  
  2498. 529
  2499. 00:31:37,475 --> 00:31:40,043
  2500. adjust their courage,
  2501. adjust their avarice,
  2502.  
  2503. 530
  2504. 00:31:40,072 --> 00:31:42,291
  2505. adjust their cunning?
  2506.  
  2507. 531
  2508. 00:31:44,859 --> 00:31:48,645
  2509. The average person, they don't see
  2510. killer robots going down the streets.
  2511.  
  2512. 532
  2513. 00:31:48,689 --> 00:31:50,996
  2514. They're like,
  2515. "What are you talking about?"
  2516.  
  2517. 533
  2518. 00:31:51,039 --> 00:31:56,245
  2519. Man, we want to make sure that we don't
  2520. have killer robots going down the street.
  2521.  
  2522. 534
  2523. 00:31:57,089 --> 00:31:59,439
  2524. Once they're going down
  2525. the street, it is too late.
  2526.  
  2527. 535
  2528. 00:32:05,053 --> 00:32:08,578
  2529. The thing that worries me right now,
  2530. that keeps me awake,
  2531.  
  2532. 536
  2533. 00:32:08,622 --> 00:32:11,842
  2534. is the development
  2535. of autonomous weapons.
  2536.  
  2537. 537
  2538. 00:32:27,815 --> 00:32:32,733
  2539. Up to now, people have expressed
  2540. unease about drones,
  2541.  
  2542. 538
  2543. 00:32:32,776 --> 00:32:35,127
  2544. which are remotely
  2545. piloted aircraft.
  2546.  
  2547. 539
  2548. 00:32:39,827 --> 00:32:43,309
  2549. If you take a drone's camera
  2550. and feed it into the AI system,
  2551.  
  2552. 540
  2553. 00:32:43,352 --> 00:32:47,443
  2554. it's a very easy step from here
  2555. to fully autonomous weapons
  2556.  
  2557. 541
  2558. 00:32:47,487 --> 00:32:50,881
  2559. that choose their own targets
  2560. and release their own missiles.
  2561.  
  2562. 542
  2563. 00:33:12,729 --> 00:33:15,080
  2564. The expected life-span
  2565. of a human being
  2566.  
  2567. 543
  2568. 00:33:15,123 --> 00:33:19,420
  2569. in that kind of battle environment
  2570. would be measured in seconds.
  2571.  
  2572. 544
  2573. 00:33:20,563 --> 00:33:23,740
  2574. At one point,
  2575. drones were science fiction,
  2576.  
  2577. 545
  2578. 00:33:23,784 --> 00:33:28,832
  2579. and now they've become
  2580. the normal thing in war.
  2581.  
  2582. 546
  2583. 00:33:28,876 --> 00:33:33,402
  2584. There's over 10,000 in
  2585. U.S. military inventory alone.
  2586.  
  2587. 547
  2588. 00:33:33,446 --> 00:33:35,274
  2589. But they're not
  2590. just a U.S. phenomena.
  2591.  
  2592. 548
  2593. 00:33:35,317 --> 00:33:39,060
  2594. There's more than 80 countries
  2595. that operate them.
  2596.  
  2597. 549
  2598. 00:33:39,104 --> 00:33:41,932
  2599. It stands to reason
  2600. that people making some
  2601.  
  2602. 550
  2603. 00:33:41,976 --> 00:33:44,587
  2604. of the most important and
  2605. difficult decisions in the world
  2606.  
  2607. 551
  2608. 00:33:44,631 --> 00:33:46,328
  2609. are gonna start to use
  2610. and implement
  2611.  
  2612. 552
  2613. 00:33:46,372 --> 00:33:48,591
  2614. artificial intelligence.
  2615.  
  2616. 553
  2617. 00:33:50,767 --> 00:33:53,596
  2618. The Air Force just designed
  2619. a $400-billion jet program
  2620.  
  2621. 554
  2622. 00:33:53,640 --> 00:33:55,555
  2623. to put pilots in the sky,
  2624.  
  2625. 555
  2626. 00:33:55,598 --> 00:34:01,300
  2627. and a $500 AI, designed by
  2628. a couple of graduate students,
  2629.  
  2630. 556
  2631. 00:34:01,343 --> 00:34:03,432
  2632. is beating the best human pilots
  2633.  
  2634. 557
  2635. 00:34:03,476 --> 00:34:05,782
  2636. with a relatively
  2637. simple algorithm.
  2638.  
  2639. 558
  2640. 00:34:09,438 --> 00:34:13,399
  2641. AI will have as big an impact
  2642. on the military
  2643.  
  2644. 559
  2645. 00:34:13,442 --> 00:34:17,490
  2646. as the combustion engine
  2647. had at the turn of the century.
  2648.  
  2649. 560
  2650. 00:34:17,533 --> 00:34:21,233
  2651. It will literally touch everything
  2652. that the military does,
  2653.  
  2654. 561
  2655. 00:34:21,276 --> 00:34:25,324
  2656. from driverless convoys
  2657. delivering logistical supplies,
  2658.  
  2659. 562
  2660. 00:34:25,367 --> 00:34:27,021
  2661. to unmanned drones
  2662.  
  2663. 563
  2664. 00:34:27,065 --> 00:34:30,764
  2665. delivering medical aid,
  2666. to computational propaganda,
  2667.  
  2668. 564
  2669. 00:34:30,807 --> 00:34:34,246
  2670. trying to win the hearts
  2671. and minds of a population.
  2672.  
  2673. 565
  2674. 00:34:34,289 --> 00:34:38,337
  2675. And so it stands to reason
  2676. that whoever has the best AI
  2677.  
  2678. 566
  2679. 00:34:38,380 --> 00:34:41,688
  2680. will probably achieve
  2681. dominance on this planet.
  2682.  
  2683. 567
  2684. 00:34:45,561 --> 00:34:47,650
  2685. At some point in
  2686. the early 21st century,
  2687.  
  2688. 568
  2689. 00:34:47,694 --> 00:34:51,219
  2690. all of mankind was
  2691. united in celebration.
  2692.  
  2693. 569
  2694. 00:34:51,263 --> 00:34:53,830
  2695. We marveled
  2696. at our own magnificence
  2697.  
  2698. 570
  2699. 00:34:53,874 --> 00:34:56,833
  2700. as we gave birth to AI.
  2701.  
  2702. 571
  2703. 00:34:56,877 --> 00:34:58,966
  2704. AI?
  2705.  
  2706. 572
  2707. 00:34:59,009 --> 00:35:00,489
  2708. You mean
  2709. artificial intelligence?
  2710.  
  2711. 573
  2712. 00:35:00,533 --> 00:35:01,751
  2713. A singular consciousness
  2714.  
  2715. 574
  2716. 00:35:01,795 --> 00:35:05,886
  2717. that spawned
  2718. an entire race of machines.
  2719.  
  2720. 575
  2721. 00:35:05,929 --> 00:35:09,716
  2722. We don't know
  2723. who struck first -- us or them,
  2724.  
  2725. 576
  2726. 00:35:09,759 --> 00:35:12,980
  2727. but we know that it was us
  2728. that scorched the sky.
  2729.  
  2730. 577
  2731. 00:35:14,677 --> 00:35:16,766
  2732. There's a long history
  2733. of science fiction,
  2734.  
  2735. 578
  2736. 00:35:16,810 --> 00:35:19,987
  2737. not just predicting the future,
  2738. but shaping the future.
  2739.  
  2740. 579
  2741. 00:35:26,863 --> 00:35:30,389
  2742. Arthur Conan Doyle
  2743. writing before World War I
  2744.  
  2745. 580
  2746. 00:35:30,432 --> 00:35:34,393
  2747. on the danger of how
  2748. submarines might be used
  2749.  
  2750. 581
  2751. 00:35:34,436 --> 00:35:38,048
  2752. to carry out civilian blockades.
  2753.  
  2754. 582
  2755. 00:35:38,092 --> 00:35:40,399
  2756. At the time
  2757. he's writing this fiction,
  2758.  
  2759. 583
  2760. 00:35:40,442 --> 00:35:43,402
  2761. the Royal Navy made fun
  2762. of Arthur Conan Doyle
  2763.  
  2764. 584
  2765. 00:35:43,445 --> 00:35:45,230
  2766. for this absurd idea
  2767.  
  2768. 585
  2769. 00:35:45,273 --> 00:35:47,623
  2770. that submarines
  2771. could be useful in war.
  2772.  
  2773. 586
  2774. 00:35:53,455 --> 00:35:55,370
  2775. One of the things
  2776. we've seen in history
  2777.  
  2778. 587
  2779. 00:35:55,414 --> 00:35:58,243
  2780. is that our attitude
  2781. towards technology,
  2782.  
  2783. 588
  2784. 00:35:58,286 --> 00:36:01,942
  2785. but also ethics,
  2786. are very context-dependent.
  2787.  
  2788. 589
  2789. 00:36:01,985 --> 00:36:03,726
  2790. For example, the submarine...
  2791.  
  2792. 590
  2793. 00:36:03,770 --> 00:36:06,468
  2794. nations like Great Britain
  2795. and even the United States
  2796.  
  2797. 591
  2798. 00:36:06,512 --> 00:36:09,863
  2799. found it horrifying
  2800. to use the submarine.
  2801.  
  2802. 592
  2803. 00:36:09,906 --> 00:36:13,214
  2804. In fact, the German use of the
  2805. submarine to carry out attacks
  2806.  
  2807. 593
  2808. 00:36:13,258 --> 00:36:18,480
  2809. was the reason why the United
  2810. States joined World War I.
  2811.  
  2812. 594
  2813. 00:36:18,524 --> 00:36:20,613
  2814. But move the timeline forward.
  2815.  
  2816. 595
  2817. 00:36:20,656 --> 00:36:23,529
  2818. The United States
  2819. of America was suddenly
  2820.  
  2821. 596
  2822. 00:36:23,572 --> 00:36:28,403
  2823. and deliberately attacked
  2824. by the empire of Japan.
  2825.  
  2826. 597
  2827. 00:36:28,447 --> 00:36:32,190
  2828. Five hours after Pearl Harbor,
  2829. the order goes out
  2830.  
  2831. 598
  2832. 00:36:32,233 --> 00:36:36,498
  2833. to commit unrestricted
  2834. submarine warfare against Japan.
  2835.  
  2836. 599
  2837. 00:36:39,936 --> 00:36:43,589
  2838. So Arthur Conan Doyle
  2839. turned out to be right.
  2840.  
  2841. 600
  2842. 00:36:44,332 --> 00:36:46,856
  2843. That's the great old line
  2844. about science fiction --
  2845.  
  2846. 601
  2847. 00:36:46,900 --> 00:36:48,336
  2848. It's a lie that tells the truth.
  2849.  
  2850. 602
  2851. 00:36:48,380 --> 00:36:51,470
  2852. Fellow executives,
  2853. it gives me great pleasure
  2854.  
  2855. 603
  2856. 00:36:51,513 --> 00:36:54,821
  2857. to introduce you to the future
  2858. of law enforcement...
  2859.  
  2860. 604
  2861. 00:36:54,864 --> 00:36:56,562
  2862. ED-209.
  2863.  
  2864. 605
  2865. 00:37:03,656 --> 00:37:05,919
  2866. This isn't just a question
  2867. of science fiction.
  2868.  
  2869. 606
  2870. 00:37:05,962 --> 00:37:09,488
  2871. This is about what's next,
  2872. about what's happening right now.
  2873.  
  2874. 607
  2875. 00:37:13,970 --> 00:37:19,324
  2876. The role of intelligent systems is
  2877. growing very rapidly in warfare.
  2878.  
  2879. 608
  2880. 00:37:19,367 --> 00:37:22,152
  2881. Everyone is pushing
  2882. in the unmanned realm.
  2883.  
  2884. 609
  2885. 00:37:26,418 --> 00:37:28,898
  2886. Today, the Secretary of Defense
  2887. is very, very clear --
  2888.  
  2889. 610
  2890. 00:37:28,942 --> 00:37:32,337
  2891. We will not create fully
  2892. autonomous attacking vehicles.
  2893.  
  2894. 611
  2895. 00:37:32,380 --> 00:37:34,643
  2896. Not everyone
  2897. is gonna hold themselves
  2898.  
  2899. 612
  2900. 00:37:34,687 --> 00:37:36,515
  2901. to that same set of values.
  2902.  
  2903. 613
  2904. 00:37:36,558 --> 00:37:40,693
  2905. And when China and Russia start
  2906. deploying autonomous vehicles
  2907.  
  2908. 614
  2909. 00:37:40,736 --> 00:37:45,611
  2910. that can attack and kill, what's
  2911. the move that we're gonna make?
  2912.  
  2913. 615
  2914. 00:37:50,006 --> 00:37:51,617
  2915. You can't say,
  2916. "Well, we're gonna use
  2917.  
  2918. 616
  2919. 00:37:51,660 --> 00:37:53,967
  2920. autonomous weapons
  2921. for our military dominance,
  2922.  
  2923. 617
  2924. 00:37:54,010 --> 00:37:56,796
  2925. but no one else
  2926. is gonna use them."
  2927.  
  2928. 618
  2929. 00:37:56,839 --> 00:38:00,495
  2930. If you make these weapons,
  2931. they're gonna be used to attack
  2932.  
  2933. 619
  2934. 00:38:00,539 --> 00:38:03,324
  2935. human populations
  2936. in large numbers.
  2937.  
  2938. 620
  2939. 00:38:12,551 --> 00:38:14,596
  2940. Autonomous weapons are,
  2941. by their nature,
  2942.  
  2943. 621
  2944. 00:38:14,640 --> 00:38:16,468
  2945. weapons of mass destruction,
  2946.  
  2947. 622
  2948. 00:38:16,511 --> 00:38:19,862
  2949. because it doesn't need a
  2950. human being to guide it or carry it.
  2951.  
  2952. 623
  2953. 00:38:19,906 --> 00:38:22,517
  2954. You only need one person,
  2955. to, you know,
  2956.  
  2957. 624
  2958. 00:38:22,561 --> 00:38:25,781
  2959. write a little program.
  2960.  
  2961. 625
  2962. 00:38:25,825 --> 00:38:30,220
  2963. It just captures
  2964. the complexity of this field.
  2965.  
  2966. 626
  2967. 00:38:30,264 --> 00:38:32,571
  2968. It is cool.
  2969. It is important.
  2970.  
  2971. 627
  2972. 00:38:32,614 --> 00:38:34,573
  2973. It is amazing.
  2974.  
  2975. 628
  2976. 00:38:34,616 --> 00:38:37,053
  2977. It is also frightening.
  2978.  
  2979. 629
  2980. 00:38:37,097 --> 00:38:38,968
  2981. And it's all about trust.
  2982.  
  2983. 630
  2984. 00:38:42,102 --> 00:38:44,583
  2985. It's an open letter about
  2986. artificial intelligence,
  2987.  
  2988. 631
  2989. 00:38:44,626 --> 00:38:47,063
  2990. signed by some of
  2991. the biggest names in science.
  2992.  
  2993. 632
  2994. 00:38:47,107 --> 00:38:48,413
  2995. What do they want?
  2996.  
  2997. 633
  2998. 00:38:48,456 --> 00:38:50,763
  2999. Ban the use of
  3000. autonomous weapons.
  3001.  
  3002. 634
  3003. 00:38:50,806 --> 00:38:52,373
  3004. The author stated,
  3005.  
  3006. 635
  3007. 00:38:52,417 --> 00:38:54,375
  3008. "Autonomous weapons
  3009. have been described
  3010.  
  3011. 636
  3012. 00:38:54,419 --> 00:38:56,595
  3013. as the third revolution
  3014. in warfare."
  3015.  
  3016. 637
  3017. 00:38:56,638 --> 00:38:58,853
  3018. ...thousand
  3019. artificial-intelligence specialists
  3020.  
  3021. 638
  3022. 00:38:58,855 --> 00:39:01,875
  3023. calling for a global ban
  3024. on killer robots.
  3025.  
  3026. 639
  3027. 00:39:01,876 --> 00:39:04,357
  3028. This open letter basically says
  3029.  
  3030. 640
  3031. 00:39:04,385 --> 00:39:07,954
  3032. that we should redefine the goal
  3033. of the field of artificial intelligence
  3034.  
  3035. 641
  3036. 00:39:07,997 --> 00:39:11,610
  3037. away from just creating pure,
  3038. undirected intelligence,
  3039.  
  3040. 642
  3041. 00:39:11,653 --> 00:39:13,655
  3042. towards creating
  3043. beneficial intelligence.
  3044.  
  3045. 643
  3046. 00:39:13,699 --> 00:39:16,092
  3047. The development of AI
  3048. is not going to stop.
  3049.  
  3050. 644
  3051. 00:39:16,136 --> 00:39:18,094
  3052. It is going to continue
  3053. and get better.
  3054.  
  3055. 645
  3056. 00:39:18,138 --> 00:39:19,835
  3057. If the international community
  3058.  
  3059. 646
  3060. 00:39:19,879 --> 00:39:21,968
  3061. isn't putting
  3062. certain controls on this,
  3063.  
  3064. 647
  3065. 00:39:22,011 --> 00:39:24,666
  3066. people will develop things
  3067. that can do anything.
  3068.  
  3069. 648
  3070. 00:39:24,710 --> 00:39:27,365
  3071. The letter says
  3072. that we are years, not decades,
  3073.  
  3074. 649
  3075. 00:39:27,408 --> 00:39:30,106
  3076. away from these weapons being deployed.
  3077. So first of all...
  3078.  
  3079. 650
  3080. 00:39:30,150 --> 00:39:32,413
  3081. We had 6,000 signatories
  3082. of that letter,
  3083.  
  3084. 651
  3085. 00:39:32,457 --> 00:39:35,155
  3086. including many of
  3087. the major figures in the field.
  3088.  
  3089. 652
  3090. 00:39:37,026 --> 00:39:39,942
  3091. I'm getting a lot of visits
  3092. from high-ranking officials
  3093.  
  3094. 653
  3095. 00:39:39,986 --> 00:39:42,989
  3096. who wish to emphasize that
  3097. American military dominance
  3098.  
  3099. 654
  3100. 00:39:43,032 --> 00:39:45,731
  3101. is very important,
  3102. and autonomous weapons
  3103.  
  3104. 655
  3105. 00:39:45,774 --> 00:39:50,083
  3106. may be part of
  3107. the Defense Department's plan.
  3108.  
  3109. 656
  3110. 00:39:50,126 --> 00:39:52,433
  3111. That's very, very scary,
  3112. because a value system
  3113.  
  3114. 657
  3115. 00:39:52,477 --> 00:39:54,479
  3116. of military developers
  3117. of technology
  3118.  
  3119. 658
  3120. 00:39:54,522 --> 00:39:57,307
  3121. is not the same as a value system
  3122. of the human race.
  3123.  
  3124. 659
  3125. 00:40:00,789 --> 00:40:02,922
  3126. Out of the concerns
  3127. about the possibility
  3128.  
  3129. 660
  3130. 00:40:02,965 --> 00:40:06,665
  3131. that this technology might be
  3132. a threat to human existence,
  3133.  
  3134. 661
  3135. 00:40:06,708 --> 00:40:08,144
  3136. a number of the technologists
  3137.  
  3138. 662
  3139. 00:40:08,188 --> 00:40:09,972
  3140. have funded
  3141. the Future of Life Institute
  3142.  
  3143. 663
  3144. 00:40:10,016 --> 00:40:12,192
  3145. to try to grapple
  3146. with these problems.
  3147.  
  3148. 664
  3149. 00:40:13,193 --> 00:40:14,847
  3150. All of these guys are secretive,
  3151.  
  3152. 665
  3153. 00:40:14,890 --> 00:40:16,805
  3154. and so it's interesting
  3155. to me to see them,
  3156.  
  3157. 666
  3158. 00:40:16,849 --> 00:40:19,735
  3159. you know, all together.
  3160.  
  3161. 667
  3162. 00:40:20,679 --> 00:40:24,030
  3163. Everything we have is a result
  3164. of our intelligence.
  3165.  
  3166. 668
  3167. 00:40:24,073 --> 00:40:26,641
  3168. It's not the result
  3169. of our big, scary teeth
  3170.  
  3171. 669
  3172. 00:40:26,685 --> 00:40:29,470
  3173. or our large claws
  3174. or our enormous muscles.
  3175.  
  3176. 670
  3177. 00:40:29,514 --> 00:40:32,473
  3178. It's because we're actually
  3179. relatively intelligent.
  3180.  
  3181. 671
  3182. 00:40:32,517 --> 00:40:35,520
  3183. And among my generation,
  3184. we're all having
  3185.  
  3186. 672
  3187. 00:40:35,563 --> 00:40:37,086
  3188. what we call "holy cow,"
  3189.  
  3190. 673
  3191. 00:40:37,130 --> 00:40:39,045
  3192. or "holy something else"
  3193. moments,
  3194.  
  3195. 674
  3196. 00:40:39,088 --> 00:40:41,003
  3197. because we see
  3198. that the technology
  3199.  
  3200. 675
  3201. 00:40:41,047 --> 00:40:44,180
  3202. is accelerating faster
  3203. than we expected.
  3204.  
  3205. 676
  3206. 00:40:44,224 --> 00:40:46,705
  3207. I remember sitting
  3208. around the table there
  3209.  
  3210. 677
  3211. 00:40:46,748 --> 00:40:50,099
  3212. with some of the best and
  3213. the smartest minds in the world,
  3214.  
  3215. 678
  3216. 00:40:50,143 --> 00:40:52,058
  3217. and what really
  3218. struck me was,
  3219.  
  3220. 679
  3221. 00:40:52,101 --> 00:40:56,149
  3222. maybe the human brain
  3223. is not able to fully grasp
  3224.  
  3225. 680
  3226. 00:40:56,192 --> 00:40:58,673
  3227. the complexity of the world
  3228. that we're confronted with.
  3229.  
  3230. 681
  3231. 00:40:58,717 --> 00:41:01,415
  3232. As it's currently constructed,
  3233.  
  3234. 682
  3235. 00:41:01,459 --> 00:41:04,766
  3236. the road that AI is following
  3237. heads off a cliff,
  3238.  
  3239. 683
  3240. 00:41:04,810 --> 00:41:07,595
  3241. and we need to change
  3242. the direction that we're going
  3243.  
  3244. 684
  3245. 00:41:07,639 --> 00:41:10,729
  3246. so that we don't take
  3247. the human race off the cliff.
  3248.  
  3249. 685
  3250. 00:41:13,558 --> 00:41:17,126
  3251. Google acquired DeepMind
  3252. several years ago.
  3253.  
  3254. 686
  3255. 00:41:17,170 --> 00:41:22,088
  3256. DeepMind operates as a
  3257. semi-independent subsidiary of Google.
  3258.  
  3259. 687
  3260. 00:41:22,131 --> 00:41:24,960
  3261. The thing that makes
  3262. DeepMind unique
  3263.  
  3264. 688
  3265. 00:41:25,004 --> 00:41:26,919
  3266. is that DeepMind
  3267. is absolutely focused
  3268.  
  3269. 689
  3270. 00:41:26,962 --> 00:41:30,313
  3271. on creating digital
  3272. superintelligence --
  3273.  
  3274. 690
  3275. 00:41:30,357 --> 00:41:34,056
  3276. an AI that is vastly smarter
  3277. than any human on Earth
  3278.  
  3279. 691
  3280. 00:41:34,100 --> 00:41:36,624
  3281. and ultimately smarter than
  3282. all humans on Earth combined.
  3283.  
  3284. 692
  3285. 00:41:36,668 --> 00:41:40,715
  3286. This is from the DeepMind
  3287. reinforcement learning system.
  3288.  
  3289. 693
  3290. 00:41:40,759 --> 00:41:43,544
  3291. Basically wakes up
  3292. like a newborn baby
  3293.  
  3294. 694
  3295. 00:41:43,588 --> 00:41:46,852
  3296. and is shown the screen
  3297. of an Atari video game
  3298.  
  3299. 695
  3300. 00:41:46,895 --> 00:41:50,508
  3301. and then has to learn
  3302. to play the video game.
  3303.  
  3304. 696
  3305. 00:41:50,551 --> 00:41:55,600
  3306. It knows nothing about objects,
  3307. about motion, about time.
  3308.  
  3309. 697
  3310. 00:41:57,602 --> 00:41:59,604
  3311. It only knows that there's
  3312. an image on the screen
  3313.  
  3314. 698
  3315. 00:41:59,647 --> 00:42:02,563
  3316. and there's a score.
  3317.  
  3318. 699
  3319. 00:42:02,607 --> 00:42:06,436
  3320. So, if your baby woke up
  3321. the day it was born
  3322.  
  3323. 700
  3324. 00:42:06,480 --> 00:42:08,090
  3325. and, by late afternoon,
  3326.  
  3327. 701
  3328. 00:42:08,134 --> 00:42:11,093
  3329. was playing
  3330. 40 different Atari video games
  3331.  
  3332. 702
  3333. 00:42:11,137 --> 00:42:15,315
  3334. at a superhuman level,
  3335. you would be terrified.
  3336.  
  3337. 703
  3338. 00:42:15,358 --> 00:42:19,101
  3339. You would say,
  3340. "My baby is possessed. Send it back."
  3341.  
  3342. 704
  3343. 00:42:19,145 --> 00:42:23,584
  3344. The DeepMind system
  3345. can win at any game.
  3346.  
  3347. 705
  3348. 00:42:23,628 --> 00:42:27,588
  3349. It can already beat all
  3350. the original Atari games.
  3351.  
  3352. 706
  3353. 00:42:27,632 --> 00:42:29,155
  3354. It is superhuman.
  3355.  
  3356. 707
  3357. 00:42:29,198 --> 00:42:31,636
  3358. It plays the games at superspeed
  3359. in less than a minute.
  3360.  
  3361. 708
  3362. 00:42:37,076 --> 00:42:38,643
  3363. DeepMind turned
  3364. to another challenge,
  3365.  
  3366. 709
  3367. 00:42:38,686 --> 00:42:40,558
  3368. and the challenge
  3369. was the game of Go,
  3370.  
  3371. 710
  3372. 00:42:40,601 --> 00:42:42,603
  3373. which people
  3374. have generally argued
  3375.  
  3376. 711
  3377. 00:42:42,647 --> 00:42:45,084
  3378. has been beyond
  3379. the power of computers
  3380.  
  3381. 712
  3382. 00:42:45,127 --> 00:42:48,304
  3383. to play with
  3384. the best human Go players.
  3385.  
  3386. 713
  3387. 00:42:48,348 --> 00:42:51,264
  3388. First, they challenged
  3389. a European Go champion.
  3390.  
  3391. 714
  3392. 00:42:53,222 --> 00:42:55,834
  3393. Then they challenged
  3394. a Korean Go champion.
  3395.  
  3396. 715
  3397. 00:42:55,877 --> 00:42:57,836
  3398. Please start the game.
  3399.  
  3400. 716
  3401. 00:42:57,879 --> 00:42:59,838
  3402. And they were able
  3403. to win both times
  3404.  
  3405. 717
  3406. 00:42:59,881 --> 00:43:02,797
  3407. in kind of striking fashion.
  3408.  
  3409. 718
  3410. 00:43:02,841 --> 00:43:05,017
  3411. You were reading articles
  3412. in New York Times years ago
  3413.  
  3414. 719
  3415. 00:43:05,060 --> 00:43:09,761
  3416. talking about how Go would take
  3417. 100 years for us to solve.
  3418.  
  3419. 720
  3420. 00:43:09,804 --> 00:43:13,460
  3421. People said, "Well, you know,
  3422. but that's still just a board.
  3423.  
  3424. 721
  3425. 00:43:13,503 --> 00:43:15,027
  3426. Poker is an art.
  3427.  
  3428. 722
  3429. 00:43:15,070 --> 00:43:16,419
  3430. Poker involves reading people.
  3431.  
  3432. 723
  3433. 00:43:16,463 --> 00:43:18,073
  3434. Poker involves lying
  3435. and bluffing.
  3436.  
  3437. 724
  3438. 00:43:18,117 --> 00:43:19,553
  3439. It's not an exact thing.
  3440.  
  3441. 725
  3442. 00:43:19,597 --> 00:43:21,381
  3443. That will never be,
  3444. you know, a computer.
  3445.  
  3446. 726
  3447. 00:43:21,424 --> 00:43:22,861
  3448. You can't do that."
  3449.  
  3450. 727
  3451. 00:43:22,904 --> 00:43:24,932
  3452. They took the best
  3453. poker players in the world,
  3454.  
  3455. 728
  3456. 00:43:25,176 --> 00:43:30,520
  3457. and it took seven days for the computer
  3458. to start demolishing the humans.
  3459.  
  3460. 729
  3461. 00:43:30,564 --> 00:43:32,461
  3462. So it's the best poker player
  3463. in the world,
  3464.  
  3465. 730
  3466. 00:43:32,462 --> 00:43:35,012
  3467. it's the best Go player in the
  3468. world, and the pattern here
  3469.  
  3470. 731
  3471. 00:43:35,013 --> 00:43:37,454
  3472. is that AI might take a little while
  3473.  
  3474. 732
  3475. 00:43:37,484 --> 00:43:40,443
  3476. to wrap its tentacles
  3477. around a new skill,
  3478.  
  3479. 733
  3480. 00:43:40,487 --> 00:43:44,883
  3481. but when it does, when it
  3482. gets it, it is unstoppable.
  3483.  
  3484. 734
  3485. 00:43:52,020 --> 00:43:55,110
  3486. DeepMind's AI has
  3487. administrator-level access
  3488.  
  3489. 735
  3490. 00:43:55,154 --> 00:43:57,156
  3491. to Google's servers
  3492.  
  3493. 736
  3494. 00:43:57,199 --> 00:44:00,768
  3495. to optimize energy usage
  3496. at the data centers.
  3497.  
  3498. 737
  3499. 00:44:00,812 --> 00:44:04,816
  3500. However, this could be
  3501. an unintentional Trojan horse.
  3502.  
  3503. 738
  3504. 00:44:04,859 --> 00:44:07,253
  3505. DeepMind has to have complete
  3506. control of the data centers,
  3507.  
  3508. 739
  3509. 00:44:07,296 --> 00:44:08,950
  3510. so with a little
  3511. software update,
  3512.  
  3513. 740
  3514. 00:44:08,994 --> 00:44:10,691
  3515. that AI could take
  3516. complete control
  3517.  
  3518. 741
  3519. 00:44:10,735 --> 00:44:12,214
  3520. of the whole Google system,
  3521.  
  3522. 742
  3523. 00:44:12,258 --> 00:44:13,607
  3524. which means
  3525. they can do anything.
  3526.  
  3527. 743
  3528. 00:44:13,651 --> 00:44:16,131
  3529. They could look at all your data.
  3530. They could do anything.
  3531.  
  3532. 744
  3533. 00:44:20,135 --> 00:44:23,051
  3534. We're rapidly heading towards
  3535. digital superintelligence
  3536.  
  3537. 745
  3538. 00:44:23,095 --> 00:44:24,313
  3539. that far exceeds any human.
  3540.  
  3541. 746
  3542. 00:44:24,357 --> 00:44:26,402
  3543. I think it's very obvious.
  3544.  
  3545. 747
  3546. 00:44:26,446 --> 00:44:29,710
  3547. The problem is, we're not gonna
  3548. suddenly hit human-level intelligence
  3549.  
  3550. 748
  3551. 00:44:29,754 --> 00:44:33,105
  3552. and say,
  3553. "Okay, let's stop research."
  3554.  
  3555. 749
  3556. 00:44:33,148 --> 00:44:35,015
  3557. It's gonna go beyond
  3558. human-level intelligence
  3559.  
  3560. 750
  3561. 00:44:35,016 --> 00:44:39,459
  3562. into what's called "superintelligence,"
  3563. and that's anything smarter than us.
  3564.  
  3565. 751
  3566. 00:44:39,502 --> 00:44:42,810
  3567. AI at the superhuman level,
  3568. if we succeed with that, will be
  3569.  
  3570. 752
  3571. 00:44:42,854 --> 00:44:46,553
  3572. by far the most powerful
  3573. invention we've ever made
  3574.  
  3575. 753
  3576. 00:44:46,596 --> 00:44:50,296
  3577. and the last invention
  3578. we ever have to make.
  3579.  
  3580. 754
  3581. 00:44:50,339 --> 00:44:53,168
  3582. And if we create AI
  3583. that's smarter than us,
  3584.  
  3585. 755
  3586. 00:44:53,212 --> 00:44:54,735
  3587. we have to be open
  3588. to the possibility
  3589.  
  3590. 756
  3591. 00:44:54,779 --> 00:44:57,520
  3592. that we might actually
  3593. lose control to them.
  3594.  
  3595. 757
  3596. 00:45:00,785 --> 00:45:02,612
  3597. Let's say
  3598. you give it some objective,
  3599.  
  3600. 758
  3601. 00:45:02,656 --> 00:45:04,745
  3602. like curing cancer,
  3603. and then you discover
  3604.  
  3605. 759
  3606. 00:45:04,789 --> 00:45:06,965
  3607. that the way
  3608. it chooses to go about that
  3609.  
  3610. 760
  3611. 00:45:07,008 --> 00:45:08,444
  3612. is actually in conflict
  3613.  
  3614. 761
  3615. 00:45:08,488 --> 00:45:11,705
  3616. with a lot of other things
  3617. you care about.
  3618.  
  3619. 762
  3620. 00:45:12,448 --> 00:45:16,496
  3621. AI doesn't have to be evil
  3622. to destroy humanity.
  3623.  
  3624. 763
  3625. 00:45:16,539 --> 00:45:20,674
  3626. If AI has a goal, and humanity
  3627. just happens to be in the way,
  3628.  
  3629. 764
  3630. 00:45:20,718 --> 00:45:22,894
  3631. it will destroy humanity
  3632. as a matter of course,
  3633.  
  3634. 765
  3635. 00:45:22,937 --> 00:45:25,113
  3636. without even thinking about it.
  3637. No hard feelings.
  3638.  
  3639. 766
  3640. 00:45:25,157 --> 00:45:27,072
  3641. It's just like
  3642. if we're building a road
  3643.  
  3644. 767
  3645. 00:45:27,115 --> 00:45:29,770
  3646. and an anthill happens
  3647. to be in the way...
  3648.  
  3649. 768
  3650. 00:45:29,814 --> 00:45:31,467
  3651. We don't hate ants.
  3652.  
  3653. 769
  3654. 00:45:31,511 --> 00:45:33,165
  3655. We're just building a road.
  3656.  
  3657. 770
  3658. 00:45:33,208 --> 00:45:34,857
  3659. And so goodbye, anthill.
  3660.  
  3661. 771
  3662. 00:45:37,996 --> 00:45:40,172
  3663. It's tempting
  3664. to dismiss these concerns,
  3665.  
  3666. 772
  3667. 00:45:40,215 --> 00:45:42,783
  3668. 'cause it's, like,
  3669. something that might happen
  3670.  
  3671. 773
  3672. 00:45:42,827 --> 00:45:47,396
  3673. in a few decades or 100 years,
  3674. so why worry?
  3675.  
  3676. 774
  3677. 00:45:47,440 --> 00:45:50,704
  3678. But if you go back
  3679. to September 11, 1933,
  3680.  
  3681. 775
  3682. 00:45:50,748 --> 00:45:54,795
  3683. Ernest Rutherford, who was the most
  3684. well-known nuclear physicist of his time,
  3685.  
  3686. 776
  3687. 00:45:54,839 --> 00:45:58,668
  3688. said that the possibility of ever
  3689. extracting useful amounts of energy
  3690.  
  3691. 777
  3692. 00:45:58,712 --> 00:46:00,801
  3693. from the transmutation
  3694. of atoms, as he called it,
  3695.  
  3696. 778
  3697. 00:46:00,845 --> 00:46:03,151
  3698. was moonshine.
  3699.  
  3700. 779
  3701. 00:46:03,195 --> 00:46:06,502
  3702. The next morning, Leo Szilard,
  3703. who was a much younger physicist,
  3704.  
  3705. 780
  3706. 00:46:06,546 --> 00:46:09,984
  3707. read this and got really annoyed
  3708. and figured out
  3709.  
  3710. 781
  3711. 00:46:10,028 --> 00:46:11,943
  3712. how to make
  3713. a nuclear chain reaction
  3714.  
  3715. 782
  3716. 00:46:11,986 --> 00:46:13,379
  3717. just a few months later.
  3718.  
  3719. 783
  3720. 00:46:20,603 --> 00:46:23,693
  3721. We have spent more
  3722. than $2 billion
  3723.  
  3724. 784
  3725. 00:46:23,737 --> 00:46:27,523
  3726. on the greatest
  3727. scientific gamble in history.
  3728.  
  3729. 785
  3730. 00:46:27,567 --> 00:46:30,222
  3731. So when people say that,
  3732. "Oh, this is so far off
  3733.  
  3734. 786
  3735. 00:46:30,265 --> 00:46:32,528
  3736. in the future, we don't have
  3737. to worry about it,"
  3738.  
  3739. 787
  3740. 00:46:32,572 --> 00:46:36,271
  3741. it might only be three, four
  3742. breakthroughs of that magnitude
  3743.  
  3744. 788
  3745. 00:46:36,315 --> 00:46:40,275
  3746. that will get us from here
  3747. to superintelligent machines.
  3748.  
  3749. 789
  3750. 00:46:40,319 --> 00:46:42,974
  3751. If it's gonna take
  3752. 20 years to figure out
  3753.  
  3754. 790
  3755. 00:46:43,017 --> 00:46:45,237
  3756. how to keep AI beneficial,
  3757.  
  3758. 791
  3759. 00:46:45,280 --> 00:46:48,849
  3760. then we should start today,
  3761. not at the last second
  3762.  
  3763. 792
  3764. 00:46:48,893 --> 00:46:51,460
  3765. when some dudes
  3766. drinking Red Bull
  3767.  
  3768. 793
  3769. 00:46:51,504 --> 00:46:53,832
  3770. decide to flip the switch
  3771. and test the thing.
  3772.  
  3773. 794
  3774. 00:46:56,814 --> 00:46:58,859
  3775. We have five years.
  3776.  
  3777. 795
  3778. 00:46:58,903 --> 00:47:03,764
  3779. I think digital superintelligence
  3780. will happen in my lifetime.
  3781.  
  3782. 796
  3783. 00:47:03,908 --> 00:47:05,735
  3784. 100%.
  3785.  
  3786. 797
  3787. 00:47:05,779 --> 00:47:07,215
  3788. When this happens,
  3789.  
  3790. 798
  3791. 00:47:07,259 --> 00:47:09,696
  3792. it will be surrounded
  3793. by a bunch of people
  3794.  
  3795. 799
  3796. 00:47:09,739 --> 00:47:13,091
  3797. who are really just excited
  3798. about the technology.
  3799.  
  3800. 800
  3801. 00:47:13,134 --> 00:47:15,571
  3802. They want to see it succeed,
  3803. but they're not anticipating
  3804.  
  3805. 801
  3806. 00:47:15,615 --> 00:47:16,964
  3807. that it can get out of control.
  3808.  
  3809. 802
  3810. 00:47:25,494 --> 00:47:28,584
  3811. Oh, my God, I trust
  3812. my computer so much.
  3813.  
  3814. 803
  3815. 00:47:28,628 --> 00:47:30,195
  3816. That's an amazing question.
  3817.  
  3818. 804
  3819. 00:47:30,238 --> 00:47:31,457
  3820. I don't trust
  3821. my computer.
  3822.  
  3823. 805
  3824. 00:47:31,500 --> 00:47:32,937
  3825. If it's on,
  3826. I take it off.
  3827.  
  3828. 806
  3829. 00:47:32,980 --> 00:47:34,242
  3830. Like, even when it's off,
  3831.  
  3832. 807
  3833. 00:47:34,286 --> 00:47:35,896
  3834. I still think it's on.
  3835. Like, you know?
  3836.  
  3837. 808
  3838. 00:47:35,897 --> 00:47:37,694
  3839. Like, you really cannot tru--
  3840. Like, the webcams,
  3841.  
  3842. 809
  3843. 00:47:37,695 --> 00:47:39,625
  3844. you don't know if, like,
  3845. someone might turn it...
  3846.  
  3847. 810
  3848. 00:47:39,639 --> 00:47:41,249
  3849. You don't know, like.
  3850.  
  3851. 811
  3852. 00:47:41,293 --> 00:47:42,903
  3853. I don't trust my computer.
  3854.  
  3855. 812
  3856. 00:47:42,947 --> 00:47:46,907
  3857. Like, in my phone,
  3858. every time they ask me
  3859.  
  3860. 813
  3861. 00:47:46,951 --> 00:47:49,475
  3862. "Can we send your
  3863. information to Apple?"
  3864.  
  3865. 814
  3866. 00:47:49,518 --> 00:47:50,998
  3867. every time, I...
  3868.  
  3869. 815
  3870. 00:47:51,042 --> 00:47:53,087
  3871. So, I don't trust my phone.
  3872.  
  3873. 816
  3874. 00:47:53,131 --> 00:47:56,743
  3875. Okay. So, part of it is,
  3876. yes, I do trust it,
  3877.  
  3878. 817
  3879. 00:47:56,786 --> 00:48:00,660
  3880. because it would be really
  3881. hard to get through the day
  3882.  
  3883. 818
  3884. 00:48:00,703 --> 00:48:04,011
  3885. in the way our world is
  3886. set up without computers.
  3887.  
  3888. 819
  3889. 00:48:10,975 --> 00:48:13,368
  3890. Trust is such a human experience.
  3891.  
  3892. 820
  3893. 00:48:21,289 --> 00:48:25,119
  3894. I have a patient coming in
  3895. with an intracranial aneurysm.
  3896.  
  3897. 821
  3898. 00:48:30,037 --> 00:48:31,691
  3899. They want to look
  3900. in my eyes and know
  3901.  
  3902. 822
  3903. 00:48:31,734 --> 00:48:34,955
  3904. that they can trust
  3905. this person with their life.
  3906.  
  3907. 823
  3908. 00:48:34,999 --> 00:48:39,129
  3909. I'm not horribly concerned
  3910. about anything.
  3911.  
  3912. 824
  3913. 00:48:39,138 --> 00:48:40,204
  3914. Good.
  3915.  
  3916. 825
  3917. 00:48:40,206 --> 00:48:42,920
  3918. Part of that is because I
  3919. have confidence in you.
  3920.  
  3921. 826
  3922. 00:48:50,753 --> 00:48:57,151
  3923. This procedure we're doing today,
  3924. 20 years ago was essentially impossible.
  3925.  
  3926. 827
  3927. 00:48:57,195 --> 00:49:00,328
  3928. We just didn't have the
  3929. materials and the technologies.
  3930.  
  3931. 828
  3932. 00:49:22,698 --> 00:49:26,485
  3933. So, the coil is barely
  3934. in there right now.
  3935.  
  3936. 829
  3937. 00:49:26,528 --> 00:49:29,923
  3938. It's just a feather
  3939. holding it in.
  3940.  
  3941. 830
  3942. 00:49:29,967 --> 00:49:32,012
  3943. It's nervous time.
  3944.  
  3945. 831
  3946. 00:49:36,190 --> 00:49:40,673
  3947. We're just in purgatory,
  3948. intellectual, humanistic purgatory,
  3949.  
  3950. 832
  3951. 00:49:40,716 --> 00:49:43,632
  3952. and AI might know
  3953. exactly what to do here.
  3954.  
  3955. 833
  3956. 00:49:50,639 --> 00:49:52,554
  3957. We've got the coil
  3958. into the aneurysm.
  3959.  
  3960. 834
  3961. 00:49:52,598 --> 00:49:54,556
  3962. But it wasn't in
  3963. tremendously well
  3964.  
  3965. 835
  3966. 00:49:54,600 --> 00:49:56,428
  3967. that I knew that it would stay,
  3968.  
  3969. 836
  3970. 00:49:56,471 --> 00:50:01,041
  3971. so with a maybe 20% risk
  3972. of a very bad situation,
  3973.  
  3974. 837
  3975. 00:50:01,085 --> 00:50:04,436
  3976. I elected
  3977. to just bring her back.
  3978.  
  3979. 838
  3980. 00:50:04,479 --> 00:50:05,959
  3981. Because of my relationship
  3982. with her
  3983.  
  3984. 839
  3985. 00:50:06,003 --> 00:50:08,222
  3986. and knowing the difficulties
  3987. of coming in
  3988.  
  3989. 840
  3990. 00:50:08,266 --> 00:50:11,051
  3991. and having the procedure,
  3992. I consider things,
  3993.  
  3994. 841
  3995. 00:50:11,095 --> 00:50:14,272
  3996. when I should only consider
  3997. the safest possible route
  3998.  
  3999. 842
  4000. 00:50:14,315 --> 00:50:16,361
  4001. to achieve success.
  4002.  
  4003. 843
  4004. 00:50:16,404 --> 00:50:19,755
  4005. But I had to stand there for
  4006. 10 minutes agonizing about it.
  4007.  
  4008. 844
  4009. 00:50:19,799 --> 00:50:21,757
  4010. The computer feels nothing.
  4011.  
  4012. 845
  4013. 00:50:21,801 --> 00:50:24,760
  4014. The computer just does
  4015. what it's supposed to do,
  4016.  
  4017. 846
  4018. 00:50:24,804 --> 00:50:26,284
  4019. better and better.
  4020.  
  4021. 847
  4022. 00:50:30,331 --> 00:50:32,551
  4023. I want to be AI in this case.
  4024.  
  4025. 848
  4026. 00:50:35,945 --> 00:50:38,861
  4027. But can AI be compassionate?
  4028.  
  4029. 849
  4030. 00:50:43,083 --> 00:50:47,827
  4031. I mean, it's everybody's
  4032. question about AI.
  4033.  
  4034. 850
  4035. 00:50:47,870 --> 00:50:51,961
  4036. We are the sole
  4037. embodiment of humanity,
  4038.  
  4039. 851
  4040. 00:50:52,005 --> 00:50:55,269
  4041. and it's a stretch for us
  4042. to accept that a machine
  4043.  
  4044. 852
  4045. 00:50:55,313 --> 00:50:58,794
  4046. can be compassionate
  4047. and loving in that way.
  4048.  
  4049. 853
  4050. 00:51:05,149 --> 00:51:07,281
  4051. Part of me doesn't believe in magic,
  4052.  
  4053. 854
  4054. 00:51:07,325 --> 00:51:09,805
  4055. but part of me has faith
  4056. that there is something
  4057.  
  4058. 855
  4059. 00:51:09,849 --> 00:51:11,546
  4060. beyond the sum of the parts,
  4061.  
  4062. 856
  4063. 00:51:11,590 --> 00:51:15,637
  4064. that there is at least a oneness
  4065. in our shared ancestry,
  4066.  
  4067. 857
  4068. 00:51:15,681 --> 00:51:19,738
  4069. our shared biology,
  4070. our shared history.
  4071.  
  4072. 858
  4073. 00:51:20,381 --> 00:51:23,210
  4074. Some connection there
  4075. beyond machine.
  4076.  
  4077. 859
  4078. 00:51:30,348 --> 00:51:32,567
  4079. So, then, you have
  4080. the other side of that, is,
  4081.  
  4082. 860
  4083. 00:51:32,611 --> 00:51:37,137
  4084. does the computer know it's conscious,
  4085. or can it be conscious, or does it care?
  4086.  
  4087. 861
  4088. 00:51:37,181 --> 00:51:40,009
  4089. Does it need to be conscious?
  4090.  
  4091. 862
  4092. 00:51:40,053 --> 00:51:42,011
  4093. Does it need to be aware?
  4094.  
  4095. 863
  4096. 00:51:52,892 --> 00:51:56,417
  4097. I do not think that a robot
  4098. could ever be conscious.
  4099.  
  4100. 864
  4101. 00:51:56,461 --> 00:51:58,376
  4102. Unless they programmed it that way.
  4103.  
  4104. 865
  4105. 00:51:58,419 --> 00:52:00,639
  4106. Conscious? No.
  4107.  
  4108. 866
  4109. 00:52:00,682 --> 00:52:03,163
  4110. No.
  4111.  
  4112. 867
  4113. 00:52:03,207 --> 00:52:06,035
  4114. I mean, think a robot could be
  4115. programmed to be conscious.
  4116.  
  4117. 868
  4118. 00:52:06,079 --> 00:52:09,648
  4119. How are they programmed
  4120. to do everything else?
  4121.  
  4122. 869
  4123. 00:52:09,691 --> 00:52:12,390
  4124. That's another big part
  4125. of artificial intelligence,
  4126.  
  4127. 870
  4128. 00:52:12,433 --> 00:52:15,741
  4129. is to make them conscious
  4130. and make them feel.
  4131.  
  4132. 871
  4133. 00:52:22,443 --> 00:52:27,709
  4134. Back in 2005, we started trying to
  4135. build machines with self-awareness.
  4136.  
  4137. 872
  4138. 00:52:33,106 --> 00:52:37,284
  4139. This robot, to begin with,
  4140. didn't know what it was.
  4141.  
  4142. 873
  4143. 00:52:37,328 --> 00:52:40,244
  4144. All it knew was that it needed
  4145. to do something like walk.
  4146.  
  4147. 874
  4148. 00:52:44,117 --> 00:52:45,597
  4149. Through trial and error,
  4150.  
  4151. 875
  4152. 00:52:45,640 --> 00:52:49,731
  4153. it figured out how to walk
  4154. using its imagination,
  4155.  
  4156. 876
  4157. 00:52:49,775 --> 00:52:54,040
  4158. and then it walked away.
  4159.  
  4160. 877
  4161. 00:52:54,083 --> 00:52:56,390
  4162. And then we did
  4163. something very cruel.
  4164.  
  4165. 878
  4166. 00:52:56,434 --> 00:52:58,653
  4167. We chopped off a leg
  4168. and watched what happened.
  4169.  
  4170. 879
  4171. 00:53:03,049 --> 00:53:07,749
  4172. At the beginning, it didn't
  4173. quite know what had happened.
  4174.  
  4175. 880
  4176. 00:53:07,793 --> 00:53:13,233
  4177. But over about a period
  4178. of a day, it then began to limp.
  4179.  
  4180. 881
  4181. 00:53:13,277 --> 00:53:16,845
  4182. And then, a year ago,
  4183. we were training an AI system
  4184.  
  4185. 882
  4186. 00:53:16,889 --> 00:53:20,240
  4187. for a live demonstration.
  4188.  
  4189. 883
  4190. 00:53:20,284 --> 00:53:24,113
  4191. We wanted to show how we wave
  4192. all these objects in front of the camera
  4193.  
  4194. 884
  4195. 00:53:24,157 --> 00:53:27,334
  4196. and the AI could
  4197. recognize the objects.
  4198.  
  4199. 885
  4200. 00:53:27,378 --> 00:53:29,031
  4201. And so, we're preparing
  4202. this demo,
  4203.  
  4204. 886
  4205. 00:53:29,075 --> 00:53:31,251
  4206. and we had on a side screen
  4207. this ability
  4208.  
  4209. 887
  4210. 00:53:31,295 --> 00:53:36,778
  4211. to watch what certain
  4212. neurons were responding to.
  4213.  
  4214. 888
  4215. 00:53:36,822 --> 00:53:41,087
  4216. And suddenly we noticed that one
  4217. of the neurons was tracking faces.
  4218.  
  4219. 889
  4220. 00:53:41,130 --> 00:53:45,483
  4221. It was tracking our faces
  4222. as we were moving around.
  4223.  
  4224. 890
  4225. 00:53:45,526 --> 00:53:48,616
  4226. Now, the spooky thing about this
  4227. is that we never trained
  4228.  
  4229. 891
  4230. 00:53:48,660 --> 00:53:52,490
  4231. the system
  4232. to recognize human faces,
  4233.  
  4234. 892
  4235. 00:53:52,533 --> 00:53:55,710
  4236. and yet, somehow,
  4237. it learned to do that.
  4238.  
  4239. 893
  4240. 00:53:57,973 --> 00:53:59,784
  4241. Even though these robots
  4242. are very simple,
  4243.  
  4244. 894
  4245. 00:53:59,785 --> 00:54:02,658
  4246. we can see there's
  4247. something else going on there.
  4248.  
  4249. 895
  4250. 00:54:02,659 --> 00:54:05,867
  4251. It's not just programming.
  4252.  
  4253. 896
  4254. 00:54:05,894 --> 00:54:08,462
  4255. So, this is just the beginning.
  4256.  
  4257. 897
  4258. 00:54:10,377 --> 00:54:14,294
  4259. I often think about
  4260. that beach in Kitty Hawk.
  4261.  
  4262. 898
  4263. 00:54:14,338 --> 00:54:18,255
  4264. The 1903 flight
  4265. by Orville and Wilbur Wright.
  4266.  
  4267. 899
  4268. 00:54:21,214 --> 00:54:24,289
  4269. It was kind of a canvas plane,
  4270. and it's wood and iron,
  4271.  
  4272. 900
  4273. 00:54:24,291 --> 00:54:26,928
  4274. and it gets off the ground for,
  4275. what, a minute and 20 seconds,
  4276.  
  4277. 901
  4278. 00:54:26,929 --> 00:54:31,006
  4279. on this windy day
  4280. before touching back down again.
  4281.  
  4282. 902
  4283. 00:54:33,270 --> 00:54:37,143
  4284. And it was
  4285. just around 65 summers or so
  4286.  
  4287. 903
  4288. 00:54:37,186 --> 00:54:43,149
  4289. after that moment that you have
  4290. a 747 taking off from JFK...
  4291.  
  4292. 904
  4293. 00:54:50,099 --> 00:54:52,184
  4294. ...where a major concern
  4295. of someone on the airplane
  4296.  
  4297. 905
  4298. 00:54:52,185 --> 00:54:55,380
  4299. might be whether or not
  4300. their salt-free diet meal
  4301.  
  4302. 906
  4303. 00:54:55,381 --> 00:54:56,917
  4304. is gonna be coming to them or not.
  4305.  
  4306. 907
  4307. 00:54:56,945 --> 00:55:01,385
  4308. We have a whole infrastructure,
  4309. with travel agents and tower control,
  4310.  
  4311. 908
  4312. 00:55:01,428 --> 00:55:03,778
  4313. and it's all casual,
  4314. and it's all part of the world.
  4315.  
  4316. 909
  4317. 00:55:07,086 --> 00:55:09,523
  4318. Right now, as far
  4319. as we've come with machines
  4320.  
  4321. 910
  4322. 00:55:09,567 --> 00:55:12,134
  4323. that think and solve problems,
  4324. we're at Kitty Hawk now.
  4325.  
  4326. 911
  4327. 00:55:12,178 --> 00:55:13,745
  4328. We're in the wind.
  4329.  
  4330. 912
  4331. 00:55:13,788 --> 00:55:17,052
  4332. We have our tattered-canvas
  4333. planes up in the air.
  4334.  
  4335. 913
  4336. 00:55:20,926 --> 00:55:23,885
  4337. But what happens
  4338. in 65 summers or so?
  4339.  
  4340. 914
  4341. 00:55:23,929 --> 00:55:27,889
  4342. We will have machines
  4343. that are beyond human control.
  4344.  
  4345. 915
  4346. 00:55:27,933 --> 00:55:30,457
  4347. Should we worry about that?
  4348.  
  4349. 916
  4350. 00:55:32,633 --> 00:55:34,853
  4351. I'm not sure it's going to help.
  4352.  
  4353. 917
  4354. 00:55:40,337 --> 00:55:44,036
  4355. Nobody has any idea today
  4356. what it means for a robot
  4357.  
  4358. 918
  4359. 00:55:44,079 --> 00:55:46,430
  4360. to be conscious.
  4361.  
  4362. 919
  4363. 00:55:46,473 --> 00:55:48,649
  4364. There is no such thing.
  4365.  
  4366. 920
  4367. 00:55:48,693 --> 00:55:50,172
  4368. There are a lot of smart people,
  4369.  
  4370. 921
  4371. 00:55:50,216 --> 00:55:53,088
  4372. and I have a great deal
  4373. of respect for them,
  4374.  
  4375. 922
  4376. 00:55:53,132 --> 00:55:57,528
  4377. but the truth is,
  4378. machines are natural psychopaths.
  4379.  
  4380. 923
  4381. 00:55:57,571 --> 00:55:59,225
  4382. Fear came back into the market.
  4383.  
  4384. 924
  4385. 00:55:59,268 --> 00:56:01,706
  4386. Went down 800,
  4387. nearly 1,000, in a heartbeat.
  4388.  
  4389. 925
  4390. 00:56:01,749 --> 00:56:03,360
  4391. I mean,
  4392. it is classic capitulation.
  4393.  
  4394. 926
  4395. 00:56:03,403 --> 00:56:07,146
  4396. There are some people who are proposing
  4397. it was some kind of fat-finger error.
  4398.  
  4399. 927
  4400. 00:56:07,189 --> 00:56:09,583
  4401. Take the Flash Crash of 2010.
  4402.  
  4403. 928
  4404. 00:56:09,627 --> 00:56:13,413
  4405. In a matter of minutes,
  4406. $1 trillion in value
  4407.  
  4408. 929
  4409. 00:56:13,457 --> 00:56:15,415
  4410. was lost in the stock market.
  4411.  
  4412. 930
  4413. 00:56:15,459 --> 00:56:18,984
  4414. The Dow dropped nearly
  4415. 1,000 points in a half-hour.
  4416.  
  4417. 931
  4418. 00:56:19,027 --> 00:56:22,553
  4419. So, what went wrong?
  4420.  
  4421. 932
  4422. 00:56:22,596 --> 00:56:26,644
  4423. By that point in time,
  4424. more than 60% of all the trades
  4425.  
  4426. 933
  4427. 00:56:26,687 --> 00:56:29,124
  4428. that took place
  4429. on the stock exchange
  4430.  
  4431. 934
  4432. 00:56:29,168 --> 00:56:32,693
  4433. were actually being
  4434. initiated by computers.
  4435.  
  4436. 935
  4437. 00:56:32,737 --> 00:56:35,783
  4438. Panic selling on the way down, and all
  4439. of a sudden it stopped on a dime.
  4440.  
  4441. 936
  4442. 00:56:35,827 --> 00:56:37,611
  4443. This is all happening
  4444. in real time, folks.
  4445.  
  4446. 937
  4447. 00:56:37,612 --> 00:56:39,883
  4448. The short story of what
  4449. happened in the Flash Crash
  4450.  
  4451. 938
  4452. 00:56:39,884 --> 00:56:42,513
  4453. is that algorithms
  4454. responded to algorithms,
  4455.  
  4456. 939
  4457. 00:56:42,514 --> 00:56:45,430
  4458. and it compounded upon itself
  4459. over and over and over again
  4460.  
  4461. 940
  4462. 00:56:45,431 --> 00:56:47,041
  4463. in a matter of minutes.
  4464.  
  4465. 941
  4466. 00:56:47,055 --> 00:56:50,972
  4467. At one point, the market
  4468. fell as if down a well.
  4469.  
  4470. 942
  4471. 00:56:51,016 --> 00:56:54,323
  4472. There is no regulatory body
  4473. that can adapt quickly enough
  4474.  
  4475. 943
  4476. 00:56:54,367 --> 00:56:57,979
  4477. to prevent potentially
  4478. disastrous consequences
  4479.  
  4480. 944
  4481. 00:56:58,023 --> 00:57:01,243
  4482. of AI operating
  4483. in our financial systems.
  4484.  
  4485. 945
  4486. 00:57:01,287 --> 00:57:03,898
  4487. They are so prime
  4488. for manipulation.
  4489.  
  4490. 946
  4491. 00:57:03,942 --> 00:57:05,639
  4492. Let's talk about the speed
  4493. with which
  4494.  
  4495. 947
  4496. 00:57:05,683 --> 00:57:08,076
  4497. we are watching
  4498. this market deteriorate.
  4499.  
  4500. 948
  4501. 00:57:08,120 --> 00:57:11,602
  4502. That's the type of AI-run-amuck
  4503. that scares people.
  4504.  
  4505. 949
  4506. 00:57:11,645 --> 00:57:13,560
  4507. When you give them a goal,
  4508.  
  4509. 950
  4510. 00:57:13,604 --> 00:57:17,225
  4511. they will relentlessly
  4512. pursue that goal.
  4513.  
  4514. 951
  4515. 00:57:17,869 --> 00:57:20,393
  4516. How many computer programs
  4517. are there like this?
  4518.  
  4519. 952
  4520. 00:57:20,437 --> 00:57:22,683
  4521. Nobody knows.
  4522.  
  4523. 953
  4524. 00:57:23,527 --> 00:57:27,444
  4525. One of the fascinating aspects
  4526. about AI in general
  4527.  
  4528. 954
  4529. 00:57:27,487 --> 00:57:31,970
  4530. is that no one really
  4531. understands how it works.
  4532.  
  4533. 955
  4534. 00:57:32,013 --> 00:57:36,975
  4535. Even the people who create AI
  4536. don't really fully understand.
  4537.  
  4538. 956
  4539. 00:57:37,018 --> 00:57:41,675
  4540. Because it has millions of elements,
  4541. it becomes completely impossible
  4542.  
  4543. 957
  4544. 00:57:41,719 --> 00:57:45,113
  4545. for a human being
  4546. to understand what's going on.
  4547.  
  4548. 958
  4549. 00:57:52,556 --> 00:57:56,037
  4550. Microsoft had set up
  4551. this artificial intelligence
  4552.  
  4553. 959
  4554. 00:57:56,081 --> 00:57:59,127
  4555. called Tay on Twitter,
  4556. which was a chatbot.
  4557.  
  4558. 960
  4559. 00:58:00,912 --> 00:58:02,696
  4560. They started out in the morning,
  4561.  
  4562. 961
  4563. 00:58:02,740 --> 00:58:06,526
  4564. and Tay was starting to tweet
  4565. and learning from stuff
  4566.  
  4567. 962
  4568. 00:58:06,570 --> 00:58:10,835
  4569. that was being sent to him
  4570. from other Twitter people.
  4571.  
  4572. 963
  4573. 00:58:10,878 --> 00:58:13,272
  4574. Because some people,
  4575. like trolls, attacked him,
  4576.  
  4577. 964
  4578. 00:58:13,315 --> 00:58:18,582
  4579. within 24 hours, the Microsoft bot
  4580. became a terrible person.
  4581.  
  4582. 965
  4583. 00:58:18,625 --> 00:58:21,367
  4584. They had to literally
  4585. pull Tay off the Net
  4586.  
  4587. 966
  4588. 00:58:21,410 --> 00:58:24,718
  4589. because he had turned
  4590. into a monster.
  4591.  
  4592. 967
  4593. 00:58:24,762 --> 00:58:30,550
  4594. A misanthropic, racist, horrible person
  4595. you'd never want to meet.
  4596.  
  4597. 968
  4598. 00:58:30,594 --> 00:58:32,857
  4599. And nobody had foreseen this.
  4600.  
  4601. 969
  4602. 00:58:35,337 --> 00:58:38,602
  4603. The whole idea of AI is that
  4604. we are not telling it exactly
  4605.  
  4606. 970
  4607. 00:58:38,645 --> 00:58:42,780
  4608. how to achieve a given
  4609. outcome or a goal.
  4610.  
  4611. 971
  4612. 00:58:42,823 --> 00:58:46,435
  4613. AI develops on its own.
  4614.  
  4615. 972
  4616. 00:58:46,479 --> 00:58:48,829
  4617. We're worried about
  4618. superintelligent AI,
  4619.  
  4620. 973
  4621. 00:58:48,873 --> 00:58:52,790
  4622. the master chess player
  4623. that will outmaneuver us,
  4624.  
  4625. 974
  4626. 00:58:52,833 --> 00:58:55,923
  4627. but AI won't have to
  4628. actually be that smart
  4629.  
  4630. 975
  4631. 00:58:55,967 --> 00:59:00,145
  4632. to have massively disruptive
  4633. effects on human civilization.
  4634.  
  4635. 976
  4636. 00:59:00,188 --> 00:59:01,886
  4637. We've seen over the last century
  4638.  
  4639. 977
  4640. 00:59:01,929 --> 00:59:05,150
  4641. it doesn't necessarily take
  4642. a genius to knock history off
  4643.  
  4644. 978
  4645. 00:59:05,193 --> 00:59:06,804
  4646. in a particular direction,
  4647.  
  4648. 979
  4649. 00:59:06,847 --> 00:59:09,589
  4650. and it won't take a genius AI
  4651. to do the same thing.
  4652.  
  4653. 980
  4654. 00:59:09,633 --> 00:59:13,158
  4655. Bogus election news stories
  4656. generated more engagement
  4657.  
  4658. 981
  4659. 00:59:13,201 --> 00:59:17,075
  4660. on Facebook
  4661. than top real stories.
  4662.  
  4663. 982
  4664. 00:59:17,118 --> 00:59:21,079
  4665. Facebook really is
  4666. the elephant in the room.
  4667.  
  4668. 983
  4669. 00:59:21,122 --> 00:59:23,777
  4670. AI running Facebook news feed --
  4671.  
  4672. 984
  4673. 00:59:23,821 --> 00:59:28,347
  4674. The task for AI
  4675. is keeping users engaged,
  4676.  
  4677. 985
  4678. 00:59:28,390 --> 00:59:29,827
  4679. but no one really understands
  4680.  
  4681. 986
  4682. 00:59:29,870 --> 00:59:34,832
  4683. exactly how this AI
  4684. is achieving this goal.
  4685.  
  4686. 987
  4687. 00:59:34,875 --> 00:59:38,792
  4688. Facebook is building an
  4689. elegant mirrored wall around us.
  4690.  
  4691. 988
  4692. 00:59:38,836 --> 00:59:41,665
  4693. A mirror that we can ask,
  4694. "Who's the fairest of them all?"
  4695.  
  4696. 989
  4697. 00:59:41,708 --> 00:59:45,016
  4698. and it will answer, "You, you,"
  4699. time and again
  4700.  
  4701. 990
  4702. 00:59:45,059 --> 00:59:48,193
  4703. and slowly begin
  4704. to warp our sense of reality,
  4705.  
  4706. 991
  4707. 00:59:48,236 --> 00:59:53,502
  4708. warp our sense of politics,
  4709. history, global events,
  4710.  
  4711. 992
  4712. 00:59:53,546 --> 00:59:57,028
  4713. until determining what's true
  4714. and what's not true,
  4715.  
  4716. 993
  4717. 00:59:57,071 --> 00:59:58,943
  4718. is virtually impossible.
  4719.  
  4720. 994
  4721. 01:00:01,032 --> 01:00:03,861
  4722. The problem is that AI
  4723. doesn't understand that.
  4724.  
  4725. 995
  4726. 01:00:03,904 --> 01:00:08,039
  4727. AI just had a mission --
  4728. maximize user engagement,
  4729.  
  4730. 996
  4731. 01:00:08,082 --> 01:00:10,041
  4732. and it achieved that.
  4733.  
  4734. 997
  4735. 01:00:10,084 --> 01:00:13,653
  4736. Nearly 2 billion people
  4737. spend nearly one hour
  4738.  
  4739. 998
  4740. 01:00:13,697 --> 01:00:17,831
  4741. on average a day
  4742. basically interacting with AI
  4743.  
  4744. 999
  4745. 01:00:17,875 --> 01:00:21,530
  4746. that is shaping
  4747. their experience.
  4748.  
  4749. 1000
  4750. 01:00:21,574 --> 01:00:24,664
  4751. Even Facebook engineers,
  4752. they don't like fake news.
  4753.  
  4754. 1001
  4755. 01:00:24,708 --> 01:00:28,015
  4756. It's very bad business.
  4757. They want to get rid of fake news.
  4758.  
  4759. 1002
  4760. 01:00:28,059 --> 01:00:32,324
  4761. It's just very difficult to do because,
  4762. how do you recognize news as fake
  4763.  
  4764. 1003
  4765. 01:00:32,367 --> 01:00:34,456
  4766. if you cannot read
  4767. all of those news personally?
  4768.  
  4769. 1004
  4770. 01:00:34,500 --> 01:00:39,418
  4771. There's so much
  4772. active misinformation
  4773.  
  4774. 1005
  4775. 01:00:39,461 --> 01:00:41,115
  4776. and it's packaged very well,
  4777.  
  4778. 1006
  4779. 01:00:41,159 --> 01:00:44,553
  4780. and it looks the same when
  4781. you see it on a Facebook page
  4782.  
  4783. 1007
  4784. 01:00:44,597 --> 01:00:47,426
  4785. or you turn on your television.
  4786.  
  4787. 1008
  4788. 01:00:47,469 --> 01:00:51,691
  4789. It's not terribly sophisticated,
  4790. but it is terribly powerful.
  4791.  
  4792. 1009
  4793. 01:00:51,735 --> 01:00:54,346
  4794. And what it means is
  4795. that your view of the world,
  4796.  
  4797. 1010
  4798. 01:00:54,389 --> 01:00:56,435
  4799. which, 20 years ago,
  4800. was determined,
  4801.  
  4802. 1011
  4803. 01:00:56,478 --> 01:01:00,004
  4804. if you watched the nightly news,
  4805. by three different networks,
  4806.  
  4807. 1012
  4808. 01:01:00,047 --> 01:01:02,528
  4809. the three anchors who endeavored
  4810. to try to get it right.
  4811.  
  4812. 1013
  4813. 01:01:02,529 --> 01:01:04,583
  4814. Might have had a little bias
  4815. one way or the other,
  4816.  
  4817. 1014
  4818. 01:01:04,584 --> 01:01:08,273
  4819. but, largely speaking, we could all
  4820. agree on an objective reality.
  4821.  
  4822. 1015
  4823. 01:01:08,316 --> 01:01:10,754
  4824. Well, that objectivity is gone,
  4825.  
  4826. 1016
  4827. 01:01:10,797 --> 01:01:13,757
  4828. and Facebook has
  4829. completely annihilated it.
  4830.  
  4831. 1017
  4832. 01:01:17,108 --> 01:01:20,807
  4833. If most of your understanding of how
  4834. the world works is derived from Facebook,
  4835.  
  4836. 1018
  4837. 01:01:20,851 --> 01:01:23,418
  4838. facilitated by algorithmic software
  4839.  
  4840. 1019
  4841. 01:01:23,462 --> 01:01:27,118
  4842. that tries to show you
  4843. the news you want to see,
  4844.  
  4845. 1020
  4846. 01:01:27,161 --> 01:01:28,815
  4847. that's a terribly
  4848. dangerous thing.
  4849.  
  4850. 1021
  4851. 01:01:28,859 --> 01:01:33,080
  4852. And the idea that we have not
  4853. only set that in motion,
  4854.  
  4855. 1022
  4856. 01:01:33,124 --> 01:01:37,258
  4857. but allowed bad-faith actors
  4858. access to that information...
  4859.  
  4860. 1023
  4861. 01:01:37,302 --> 01:01:39,565
  4862. I mean, this is a recipe
  4863. for disaster.
  4864.  
  4865. 1024
  4866. 01:01:43,177 --> 01:01:45,876
  4867. I think that there will definitely
  4868. be lots of bad actors
  4869.  
  4870. 1025
  4871. 01:01:45,919 --> 01:01:48,922
  4872. trying to manipulate the world with AI.
  4873.  
  4874. 1026
  4875. 01:01:48,966 --> 01:01:52,143
  4876. 2016 was a perfect example
  4877. of an election
  4878.  
  4879. 1027
  4880. 01:01:52,186 --> 01:01:55,015
  4881. where there was lots of AI
  4882. producing lots of fake news
  4883.  
  4884. 1028
  4885. 01:01:55,059 --> 01:01:58,323
  4886. and distributing it
  4887. for a purpose, for a result.
  4888.  
  4889. 1029
  4890. 01:01:59,890 --> 01:02:02,283
  4891. Ladies and gentlemen,
  4892. honorable colleagues...
  4893.  
  4894. 1030
  4895. 01:02:02,327 --> 01:02:04,546
  4896. it's my privilege
  4897. to speak to you today
  4898.  
  4899. 1031
  4900. 01:02:04,590 --> 01:02:07,985
  4901. about the power of big data
  4902. and psychographics
  4903.  
  4904. 1032
  4905. 01:02:08,028 --> 01:02:09,682
  4906. in the electoral process
  4907.  
  4908. 1033
  4909. 01:02:09,726 --> 01:02:12,206
  4910. and, specifically,
  4911. to talk about the work
  4912.  
  4913. 1034
  4914. 01:02:12,250 --> 01:02:14,513
  4915. that we contributed
  4916. to Senator Cruz's
  4917.  
  4918. 1035
  4919. 01:02:14,556 --> 01:02:16,558
  4920. presidential primary campaign.
  4921.  
  4922. 1036
  4923. 01:02:16,602 --> 01:02:19,910
  4924. Cambridge Analytica
  4925. emerged quietly as a company
  4926.  
  4927. 1037
  4928. 01:02:19,953 --> 01:02:21,563
  4929. that, according to its own hype,
  4930.  
  4931. 1038
  4932. 01:02:21,607 --> 01:02:26,307
  4933. has the ability to use
  4934. this tremendous amount of data
  4935.  
  4936. 1039
  4937. 01:02:26,351 --> 01:02:30,137
  4938. in order
  4939. to effect societal change.
  4940.  
  4941. 1040
  4942. 01:02:30,181 --> 01:02:33,358
  4943. In 2016, they had
  4944. three major clients.
  4945.  
  4946. 1041
  4947. 01:02:33,401 --> 01:02:34,794
  4948. Ted Cruz was one of them.
  4949.  
  4950. 1042
  4951. 01:02:34,838 --> 01:02:37,884
  4952. It's easy to forget that,
  4953. only 18 months ago,
  4954.  
  4955. 1043
  4956. 01:02:37,928 --> 01:02:42,846
  4957. Senator Cruz was one of the less
  4958. popular candidates seeking nomination.
  4959.  
  4960. 1044
  4961. 01:02:42,889 --> 01:02:47,241
  4962. So, what was not possible maybe,
  4963. like, 10 or 15 years ago,
  4964.  
  4965. 1045
  4966. 01:02:47,285 --> 01:02:49,374
  4967. was that you can send fake news
  4968.  
  4969. 1046
  4970. 01:02:49,417 --> 01:02:52,420
  4971. to exactly the people
  4972. that you want to send it to.
  4973.  
  4974. 1047
  4975. 01:02:52,464 --> 01:02:56,685
  4976. And then you could actually see
  4977. how he or she reacts on Facebook
  4978.  
  4979. 1048
  4980. 01:02:56,729 --> 01:02:58,905
  4981. and then adjust that information
  4982.  
  4983. 1049
  4984. 01:02:58,949 --> 01:03:01,778
  4985. according to the feedback
  4986. that you got.
  4987.  
  4988. 1050
  4989. 01:03:01,821 --> 01:03:03,257
  4990. So you can start developing
  4991.  
  4992. 1051
  4993. 01:03:03,301 --> 01:03:06,130
  4994. kind of a real-time management
  4995. of a population.
  4996.  
  4997. 1052
  4998. 01:03:06,173 --> 01:03:10,699
  4999. In this case, we've zoned in on
  5000. a group we've called "Persuasion."
  5001.  
  5002. 1053
  5003. 01:03:10,743 --> 01:03:13,746
  5004. These are people who are
  5005. definitely going to vote,
  5006.  
  5007. 1054
  5008. 01:03:13,790 --> 01:03:16,705
  5009. to caucus, but they need
  5010. moving from the center
  5011.  
  5012. 1055
  5013. 01:03:16,749 --> 01:03:18,490
  5014. a little bit more
  5015. towards the right.
  5016.  
  5017. 1056
  5018. 01:03:18,533 --> 01:03:19,708
  5019. in order to support Cruz.
  5020.  
  5021. 1057
  5022. 01:03:19,752 --> 01:03:22,059
  5023. They need a persuasion message.
  5024.  
  5025. 1058
  5026. 01:03:22,102 --> 01:03:23,800
  5027. "Gun rights," I've selected.
  5028.  
  5029. 1059
  5030. 01:03:23,843 --> 01:03:25,802
  5031. That narrows the field
  5032. slightly more.
  5033.  
  5034. 1060
  5035. 01:03:25,845 --> 01:03:29,066
  5036. And now we know that we need
  5037. a message on gun rights,
  5038.  
  5039. 1061
  5040. 01:03:29,109 --> 01:03:31,111
  5041. it needs to be
  5042. a persuasion message,
  5043.  
  5044. 1062
  5045. 01:03:31,155 --> 01:03:34,201
  5046. and it needs to be nuanced
  5047. according to the certain personality
  5048.  
  5049. 1063
  5050. 01:03:34,245 --> 01:03:36,029
  5051. that we're interested in.
  5052.  
  5053. 1064
  5054. 01:03:36,073 --> 01:03:39,946
  5055. Through social media, there's an
  5056. infinite amount of information
  5057.  
  5058. 1065
  5059. 01:03:39,990 --> 01:03:42,514
  5060. that you can gather
  5061. about a person.
  5062.  
  5063. 1066
  5064. 01:03:42,557 --> 01:03:45,734
  5065. We have somewhere close
  5066. to 4,000 or 5,000 data points
  5067.  
  5068. 1067
  5069. 01:03:45,778 --> 01:03:48,563
  5070. on every adult
  5071. in the United States.
  5072.  
  5073. 1068
  5074. 01:03:48,607 --> 01:03:51,915
  5075. It's about targeting
  5076. the individual.
  5077.  
  5078. 1069
  5079. 01:03:51,958 --> 01:03:55,962
  5080. It's like a weapon, which can be used
  5081. in the totally wrong direction.
  5082.  
  5083. 1070
  5084. 01:03:56,006 --> 01:03:58,051
  5085. That's the problem
  5086. with all of this data.
  5087.  
  5088. 1071
  5089. 01:03:58,095 --> 01:04:02,229
  5090. It's almost as if we built the bullet
  5091. before we built the gun.
  5092.  
  5093. 1072
  5094. 01:04:02,273 --> 01:04:06,407
  5095. Ted Cruz employed our data,
  5096. our behavioral insights.
  5097.  
  5098. 1073
  5099. 01:04:06,451 --> 01:04:09,541
  5100. He started from a base
  5101. of less than 5%
  5102.  
  5103. 1074
  5104. 01:04:09,584 --> 01:04:15,590
  5105. and had a very slow-and-steady-
  5106. but-firm rise to above 35%,
  5107.  
  5108. 1075
  5109. 01:04:15,634 --> 01:04:17,157
  5110. making him, obviously,
  5111.  
  5112. 1076
  5113. 01:04:17,201 --> 01:04:20,465
  5114. the second most threatening
  5115. contender in the race.
  5116.  
  5117. 1077
  5118. 01:04:20,508 --> 01:04:23,120
  5119. Now, clearly, the Cruz campaign
  5120. is over now,
  5121.  
  5122. 1078
  5123. 01:04:23,163 --> 01:04:28,168
  5124. but what I can tell you is that of the
  5125. two candidates left in this election,
  5126.  
  5127. 1079
  5128. 01:04:28,212 --> 01:04:30,867
  5129. one of them is using
  5130. these technologies.
  5131.  
  5132. 1080
  5133. 01:04:32,564 --> 01:04:35,959
  5134. I, Donald John Trump,
  5135. do solemnly swear
  5136.  
  5137. 1081
  5138. 01:04:36,002 --> 01:04:38,222
  5139. that I will faithfully execute
  5140.  
  5141. 1082
  5142. 01:04:38,265 --> 01:04:42,226
  5143. the office of President
  5144. of the United States.
  5145.  
  5146. 1083
  5147. 01:04:48,275 --> 01:04:50,234
  5148. Elections are
  5149. a marginal exercise.
  5150.  
  5151. 1084
  5152. 01:04:50,277 --> 01:04:53,237
  5153. It doesn't take
  5154. a very sophisticated AI
  5155.  
  5156. 1085
  5157. 01:04:53,280 --> 01:04:57,719
  5158. in order to have
  5159. a disproportionate impact.
  5160.  
  5161. 1086
  5162. 01:04:57,763 --> 01:05:02,550
  5163. Before Trump, Brexit was
  5164. another supposed client.
  5165.  
  5166. 1087
  5167. 01:05:02,594 --> 01:05:04,726
  5168. Well, at 20 minutes to 5:00,
  5169.  
  5170. 1088
  5171. 01:05:04,770 --> 01:05:08,730
  5172. we can now say
  5173. the decision taken in 1975
  5174.  
  5175. 1089
  5176. 01:05:08,774 --> 01:05:10,950
  5177. by this country to join
  5178. the common market
  5179.  
  5180. 1090
  5181. 01:05:10,994 --> 01:05:15,999
  5182. has been reversed by
  5183. this referendum to leave the EU.
  5184.  
  5185. 1091
  5186. 01:05:16,042 --> 01:05:19,828
  5187. Cambridge Analytica
  5188. allegedly uses AI
  5189.  
  5190. 1092
  5191. 01:05:19,872 --> 01:05:23,267
  5192. to push through two of
  5193. the most ground-shaking pieces
  5194.  
  5195. 1093
  5196. 01:05:23,310 --> 01:05:27,967
  5197. of political change
  5198. in the last 50 years.
  5199.  
  5200. 1094
  5201. 01:05:28,011 --> 01:05:30,709
  5202. These are epochal events,
  5203. and if we believe the hype,
  5204.  
  5205. 1095
  5206. 01:05:30,752 --> 01:05:33,755
  5207. they are connected directly
  5208. to a piece of software,
  5209.  
  5210. 1096
  5211. 01:05:33,799 --> 01:05:37,194
  5212. essentially, created
  5213. by a professor at Stanford.
  5214.  
  5215. 1097
  5216. 01:05:41,459 --> 01:05:45,593
  5217. Back in 2013, I described that
  5218. what they are doing is possible
  5219.  
  5220. 1098
  5221. 01:05:45,637 --> 01:05:49,293
  5222. and warned against this
  5223. happening in the future.
  5224.  
  5225. 1099
  5226. 01:05:49,336 --> 01:05:51,382
  5227. At the time, Michal Kosinski
  5228.  
  5229. 1100
  5230. 01:05:51,425 --> 01:05:54,994
  5231. was a young Polish researcher
  5232. working at the Psychometrics Centre.
  5233.  
  5234. 1101
  5235. 01:05:55,038 --> 01:06:00,217
  5236. So, what Michal had done was to
  5237. gather the largest-ever data set
  5238.  
  5239. 1102
  5240. 01:06:00,260 --> 01:06:03,481
  5241. of how people
  5242. behave on Facebook.
  5243.  
  5244. 1103
  5245. 01:06:03,524 --> 01:06:07,789
  5246. Psychometrics is trying
  5247. to measure psychological traits,
  5248.  
  5249. 1104
  5250. 01:06:07,833 --> 01:06:09,922
  5251. such as personality,
  5252. intelligence,
  5253.  
  5254. 1105
  5255. 01:06:09,966 --> 01:06:11,880
  5256. political views, and so on.
  5257.  
  5258. 1106
  5259. 01:06:11,924 --> 01:06:15,058
  5260. Now, traditionally,
  5261. those traits were measured
  5262.  
  5263. 1107
  5264. 01:06:15,101 --> 01:06:17,712
  5265. using tests and questions.
  5266.  
  5267. 1108
  5268. 01:06:17,756 --> 01:06:20,715
  5269. Personality test, the most benign thing
  5270. you could possibly think of.
  5271.  
  5272. 1109
  5273. 01:06:20,759 --> 01:06:24,197
  5274. Something that doesn't necessarily
  5275. have a lot of utility, right?
  5276.  
  5277. 1110
  5278. 01:06:24,241 --> 01:06:27,331
  5279. Our idea was that
  5280. instead of tests and questions,
  5281.  
  5282. 1111
  5283. 01:06:27,374 --> 01:06:30,029
  5284. we could simply look at the
  5285. digital footprints of behaviors
  5286.  
  5287. 1112
  5288. 01:06:30,073 --> 01:06:32,553
  5289. that we are all leaving behind
  5290.  
  5291. 1113
  5292. 01:06:32,597 --> 01:06:34,903
  5293. to understand openness,
  5294.  
  5295. 1114
  5296. 01:06:34,947 --> 01:06:37,732
  5297. conscientiousness,
  5298. neuroticism.
  5299.  
  5300. 1115
  5301. 01:06:37,776 --> 01:06:39,560
  5302. You can easily buy
  5303. personal data,
  5304.  
  5305. 1116
  5306. 01:06:39,604 --> 01:06:43,129
  5307. such as where you live,
  5308. what club memberships you've tried,
  5309.  
  5310. 1117
  5311. 01:06:43,173 --> 01:06:45,044
  5312. which gym you go to.
  5313.  
  5314. 1118
  5315. 01:06:45,088 --> 01:06:47,873
  5316. There are actually marketplaces
  5317. for personal data.
  5318.  
  5319. 1119
  5320. 01:06:47,916 --> 01:06:51,442
  5321. It turns out, we can discover
  5322. an awful lot about what you're gonna do
  5323.  
  5324. 1120
  5325. 01:06:51,485 --> 01:06:55,750
  5326. based on a very, very tiny
  5327. set of information.
  5328.  
  5329. 1121
  5330. 01:06:55,794 --> 01:06:58,275
  5331. We are training
  5332. deep-learning networks
  5333.  
  5334. 1122
  5335. 01:06:58,318 --> 01:07:01,278
  5336. to infer intimate traits,
  5337.  
  5338. 1123
  5339. 01:07:01,321 --> 01:07:04,759
  5340. people's political views,
  5341. personality,
  5342.  
  5343. 1124
  5344. 01:07:04,803 --> 01:07:07,806
  5345. intelligence,
  5346. sexual orientation
  5347.  
  5348. 1125
  5349. 01:07:07,849 --> 01:07:10,504
  5350. just from an image
  5351. from someone's face.
  5352.  
  5353. 1126
  5354. 01:07:17,076 --> 01:07:20,645
  5355. Now think about countries which
  5356. are not so free and open-minded.
  5357.  
  5358. 1127
  5359. 01:07:20,688 --> 01:07:23,300
  5360. If you can reveal people's
  5361. religious views
  5362.  
  5363. 1128
  5364. 01:07:23,343 --> 01:07:25,954
  5365. or political views
  5366. or sexual orientation
  5367.  
  5368. 1129
  5369. 01:07:25,998 --> 01:07:28,740
  5370. based on only profile pictures,
  5371.  
  5372. 1130
  5373. 01:07:28,783 --> 01:07:33,310
  5374. this could be literally
  5375. an issue of life and death.
  5376.  
  5377. 1131
  5378. 01:07:37,009 --> 01:07:39,751
  5379. I think there's no going back.
  5380.  
  5381. 1132
  5382. 01:07:42,145 --> 01:07:44,321
  5383. Do you know what
  5384. the Turing test is?
  5385.  
  5386. 1133
  5387. 01:07:44,364 --> 01:07:48,977
  5388. It's when a human interacts
  5389. with a computer,
  5390.  
  5391. 1134
  5392. 01:07:49,021 --> 01:07:52,546
  5393. and if the human doesn't know they're
  5394. interacting with a computer,
  5395.  
  5396. 1135
  5397. 01:07:52,590 --> 01:07:54,126
  5398. the test is passed.
  5399.  
  5400. 1136
  5401. 01:07:54,170 --> 01:07:57,247
  5402. And over the next few days,
  5403.  
  5404. 1137
  5405. 01:07:57,290 --> 01:07:59,684
  5406. you're gonna be the human
  5407. component in a Turing test.
  5408.  
  5409. 1138
  5410. 01:07:59,727 --> 01:08:02,295
  5411. Holy shit.
  5412. - That's right, Caleb.
  5413.  
  5414. 1139
  5415. 01:08:02,339 --> 01:08:04,080
  5416. You got it.
  5417.  
  5418. 1140
  5419. 01:08:04,123 --> 01:08:06,865
  5420. 'Cause if that test is passed,
  5421.  
  5422. 1141
  5423. 01:08:06,908 --> 01:08:10,825
  5424. you are dead center
  5425. of the greatest scientific event
  5426.  
  5427. 1142
  5428. 01:08:10,869 --> 01:08:12,958
  5429. in the history of man.
  5430.  
  5431. 1143
  5432. 01:08:13,001 --> 01:08:17,615
  5433. If you've created a conscious machine,
  5434. it's not the history of man.
  5435.  
  5436. 1144
  5437. 01:08:17,658 --> 01:08:19,356
  5438. That's the history of gods.
  5439.  
  5440. 1145
  5441. 01:08:26,841 --> 01:08:29,975
  5442. It's almost like technology
  5443. is a god in and of itself.
  5444.  
  5445. 1146
  5446. 01:08:33,196 --> 01:08:35,241
  5447. Like the weather.
  5448. We can't impact it.
  5449.  
  5450. 1147
  5451. 01:08:35,285 --> 01:08:39,293
  5452. We can't slow it down.
  5453. We can't stop it.
  5454.  
  5455. 1148
  5456. 01:08:39,637 --> 01:08:42,849
  5457. We feel powerless.
  5458.  
  5459. 1149
  5460. 01:08:43,293 --> 01:08:46,687
  5461. If we think of God as an unlimited
  5462. amount of intelligence,
  5463.  
  5464. 1150
  5465. 01:08:46,731 --> 01:08:50,474
  5466. the closest we can get to that is
  5467. by evolving our own intelligence
  5468.  
  5469. 1151
  5470. 01:08:50,517 --> 01:08:55,566
  5471. by merging with the artificial
  5472. intelligence we're creating.
  5473.  
  5474. 1152
  5475. 01:08:55,609 --> 01:08:58,003
  5476. Today, our computers, phones,
  5477.  
  5478. 1153
  5479. 01:08:58,046 --> 01:09:01,615
  5480. applications give us
  5481. superhuman capability.
  5482.  
  5483. 1154
  5484. 01:09:01,659 --> 01:09:04,662
  5485. So, as the old maxim says,
  5486. if you can't beat 'em, join 'em.
  5487.  
  5488. 1155
  5489. 01:09:06,968 --> 01:09:09,971
  5490. It's about
  5491. a human-machine partnership.
  5492.  
  5493. 1156
  5494. 01:09:10,015 --> 01:09:11,669
  5495. I mean, we already see how, you know,
  5496.  
  5497. 1157
  5498. 01:09:11,712 --> 01:09:14,933
  5499. our phones, for example,
  5500. act as memory prosthesis, right?
  5501.  
  5502. 1158
  5503. 01:09:14,976 --> 01:09:17,196
  5504. I don't have to remember
  5505. your phone number anymore
  5506.  
  5507. 1159
  5508. 01:09:17,240 --> 01:09:19,198
  5509. 'cause it's on my phone.
  5510.  
  5511. 1160
  5512. 01:09:19,242 --> 01:09:22,070
  5513. It's about machines
  5514. augmenting our human abilities,
  5515.  
  5516. 1161
  5517. 01:09:22,114 --> 01:09:25,248
  5518. as opposed to, like,
  5519. completely displacing them.
  5520.  
  5521. 1162
  5522. 01:09:25,249 --> 01:09:27,538
  5523. If you look at all the objects
  5524. that have made the leap
  5525.  
  5526. 1163
  5527. 01:09:27,539 --> 01:09:30,237
  5528. from analog to digital
  5529. over the last 20 years...
  5530.  
  5531. 1164
  5532. 01:09:30,238 --> 01:09:32,123
  5533. it's a lot.
  5534.  
  5535. 1165
  5536. 01:09:32,124 --> 01:09:35,388
  5537. We're the last analog object
  5538. in a digital universe.
  5539.  
  5540. 1166
  5541. 01:09:35,389 --> 01:09:37,068
  5542. And the problem with that,
  5543. of course,
  5544.  
  5545. 1167
  5546. 01:09:37,069 --> 01:09:40,609
  5547. is that the data input/output
  5548. is very limited.
  5549.  
  5550. 1168
  5551. 01:09:40,611 --> 01:09:42,613
  5552. It's this.
  5553. It's these.
  5554.  
  5555. 1169
  5556. 01:09:42,856 --> 01:09:45,355
  5557. Our eyes are pretty good.
  5558.  
  5559. 1170
  5560. 01:09:45,398 --> 01:09:48,445
  5561. We're able to take in a lot
  5562. of visual information.
  5563.  
  5564. 1171
  5565. 01:09:48,488 --> 01:09:52,536
  5566. But our information output
  5567. is very, very, very low.
  5568.  
  5569. 1172
  5570. 01:09:52,579 --> 01:09:55,669
  5571. The reason this is important --
  5572. If we envision a scenario
  5573.  
  5574. 1173
  5575. 01:09:55,713 --> 01:09:59,543
  5576. where AI's playing a more
  5577. prominent role in societies,
  5578.  
  5579. 1174
  5580. 01:09:59,586 --> 01:10:02,023
  5581. we want good ways to interact
  5582. with this technology
  5583.  
  5584. 1175
  5585. 01:10:02,067 --> 01:10:04,983
  5586. so that it ends up
  5587. augmenting us.
  5588.  
  5589. 1176
  5590. 01:10:07,855 --> 01:10:12,295
  5591. I think it's incredibly important
  5592. that AI not be "other."
  5593.  
  5594. 1177
  5595. 01:10:12,338 --> 01:10:14,562
  5596. It must be us.
  5597.  
  5598. 1178
  5599. 01:10:14,906 --> 01:10:18,605
  5600. And I could be wrong
  5601. about what I'm saying.
  5602.  
  5603. 1179
  5604. 01:10:18,649 --> 01:10:23,915
  5605. I'm certainly open to ideas if anybody
  5606. can suggest a path that's better.
  5607.  
  5608. 1180
  5609. 01:10:23,958 --> 01:10:27,266
  5610. But I think we're gonna really
  5611. have to either merge with AI
  5612.  
  5613. 1181
  5614. 01:10:27,310 --> 01:10:29,063
  5615. or be left behind.
  5616.  
  5617. 1182
  5618. 01:10:36,406 --> 01:10:38,756
  5619. It's hard to kind of
  5620. think of unplugging a system
  5621.  
  5622. 1183
  5623. 01:10:38,799 --> 01:10:41,802
  5624. that's distributed
  5625. everywhere on the planet,
  5626.  
  5627. 1184
  5628. 01:10:41,846 --> 01:10:45,806
  5629. that's distributed now
  5630. across the solar system.
  5631.  
  5632. 1185
  5633. 01:10:45,850 --> 01:10:49,375
  5634. You can't just, you know,
  5635. shut that off.
  5636.  
  5637. 1186
  5638. 01:10:49,419 --> 01:10:51,290
  5639. We've opened Pandora's box.
  5640.  
  5641. 1187
  5642. 01:10:51,334 --> 01:10:55,642
  5643. We've unleashed forces that
  5644. we can't control, we can't stop.
  5645.  
  5646. 1188
  5647. 01:10:55,686 --> 01:10:59,516
  5648. We're in the midst of essentially
  5649. creating a new life-form on Earth.
  5650.  
  5651. 1189
  5652. 01:11:05,870 --> 01:11:07,611
  5653. We don't know what happens next.
  5654.  
  5655. 1190
  5656. 01:11:07,654 --> 01:11:10,353
  5657. We don't know what shape
  5658. the intellect of a machine
  5659.  
  5660. 1191
  5661. 01:11:10,396 --> 01:11:14,531
  5662. will be when that intellect is
  5663. far beyond human capabilities.
  5664.  
  5665. 1192
  5666. 01:11:14,574 --> 01:11:17,360
  5667. It's just not something
  5668. that's possible.
  5669.  
  5670. 1193
  5671. 01:11:24,758 --> 01:11:26,976
  5672. The least scary future
  5673. I can think of is one
  5674.  
  5675. 1194
  5676. 01:11:26,978 --> 01:11:29,633
  5677. where we have at least
  5678. democratized AI.
  5679.  
  5680. 1195
  5681. 01:11:31,548 --> 01:11:34,159
  5682. Because if one company
  5683. or small group of people
  5684.  
  5685. 1196
  5686. 01:11:34,202 --> 01:11:37,031
  5687. manages to develop godlike
  5688. digital superintelligence,
  5689.  
  5690. 1197
  5691. 01:11:37,075 --> 01:11:39,739
  5692. they can take over the world.
  5693.  
  5694. 1198
  5695. 01:11:40,383 --> 01:11:44,343
  5696. At least when there's an evil dictator,
  5697. that human is going to die,
  5698.  
  5699. 1199
  5700. 01:11:44,387 --> 01:11:46,998
  5701. but, for an AI,
  5702. there would be no death.
  5703.  
  5704. 1200
  5705. 01:11:47,041 --> 01:11:49,392
  5706. It would live forever.
  5707.  
  5708. 1201
  5709. 01:11:49,435 --> 01:11:53,670
  5710. And then you have an immortal dictator
  5711. from which we can never escape.
  5712.  
  5713. 1202
  5714. 01:13:33,000 --> 01:13:36,500
  5715. Correction and synchronisation:
  5716. Mazrim Taim
RAW Paste Data
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy. OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!
 
Top