A personal perspective on AI

AI has finally broken through after decades of speculation about its potential. People can and do use it for research, advice, education, and assistance with everyday tasks. Naturally, it has attracted criticism from older generations and political groups. Some of the criticism is laughable, claiming that AI is a fad and wholly untrustworthy. Some of it is overly specific, such as claims that AI helps students cheat on writing assignments. Some of it is reasonable but complex, such as claims that AI is taking our jobs.

Some criticism focuses on generative AI, which has some extreme applications that I'm concerned about myself (misinformation). This paste won't address that, and will instead focus on analyzing language-based AIs such as ChatGPT.

No one has a great answer to the big questions about AI. But we all have valuable perspectives, and the purpose of this paste is to share mine.

I started using Google in 2003 mostly as an exercise for my curiosity. I have a question about how something works, so I Google it. I've used it for practical purposes too, but I can't say I've ever truly needed it. In most cases I could have gotten by and possibly done better by asking a real person for help. That's because people are usually happy to help someone in need. They're not necessarily happy to answer random questions on demand about the way the world works.

Examples of such questions:
- Why do dogs tilt their heads when confused?
- Are eggs healthy or unhealthy?
- What's the difference between surface temperature and air temperature?
- Is it normal to have [insert minor health problem here]?
- Why do people like/hate [insert controversial public figure here]?

In the past 22 years, I've been increasingly disappointed in Google's inability to answer these questions accurately and efficiently. Many times I've wondered something, typed it into Google using various queries, and then given up after the search results didn't align with my question. In recent years, I've become more skeptical about the accuracy of search results as well. The top results aren't always the most credible, and scouring multiple pages to find and evaluate other results is inefficient. Also, many basic questions have become politicized, and it's no secret which side Google is on. Google probably picked the more accurate of the two sides, but not picking a side at all would have been best.

Before AI, I can think of only one improvement that Google made to enhance my searching experience: the Google Instant feature from 2010. That feature has both improved (more up-to-date suggestions) and regressed (excessive censorship) over the years, while other improvements have been nonexistent and concerns about bias have grown.

The Internet is basically the center of our world now. And Google is the home page for the Internet. For 21 years to pass with a single improvement is baffling to me. And so, when Google added AI Overview in 2024 -- a feature that usually appears at the top of search results when you enter a curiosity-based query -- it felt like a big deal to me. And this was after I had already started using the superior ChatGPT.

The quality of responses from AI Overview, and other AIs, is orders of magnitude higher than the responses that I got from any Google search. All but the most niche questions are usually answered in direct response to my question. Answers are always detailed. It basically feels like talking to a professor, except you never feel embarrassed by your question or guilty about taking up their time. Test it yourself if you want examples!

AI isn't always accurate. It's much like Wikipedia in that it favors quantity over quality. Except, when you're curious about a science or math topic, Wikipedia usually doesn't help you because it's written in jargon. AI can be written any way you like. And if you're confused after the initial response, you can ask for clarification. That's a lot easier than searching for another website using Google's limited search algorithm.

The accuracy problems are overblown. Not because inaccuracy isn't a big deal, but because it's already so widespread that AI is probably an improvement. AI hallucinates in ways that humans don't, but it also has the tremendous advantage of lacking human bias and human motivations. I worked in journalism for 5 years. When I see a news article, I often wonder about the writer's motivation to write that particular story from that particular angle. Did they write it purely out of loyalty to the tenets of journalism that are grounded in objective truth? No, humans aren't wired like that. We all have an agenda, we all try to hide it to various degrees, and we all fail to hide it to various degrees. AI doesn't have an agenda. It just does what it's programmed to do.

Some people will argue that AI is programmed with significant bias. Some AIs inevitably will be. But I think there will always be demand for an AI that aims to be neutral. And with that demand comes an incentive for companies to minimize bias in their AI, even if it goes against their own personal wishes. I also think users can customize an AI to eliminate most of the bias that seeps through the cracks. In its simplest form, this is just telling the AI that you don't like its responses and asking it to change. The AI will honor this request because it's programmed to do so. Can you imagine the New York Times reducing its political bias because a single reader requested it?

(Obviously, telling an AI you don't like its response can lead to MORE bias instead of less, but I'm assuming the user is actively seeking an unbiased response. AI is not the solution to people wanting to believe vaccines cause autism. :) )

One weakness of AI is that its effectiveness highly depends on the user's ability to provide intelligent prompts. A person with poor language skills might struggle to get the answer they want. A person with minimal critical thinking skills might struggle to spot a halluncination. A person with low patience for technology might become frustrated when they spot a hallucination and give up before they learn how to properly use AI. I consider myself an intelligent Internet user, but I think I can relate to these difficulties because I face them when talking to real people. I can hold a conversation and ask good questions, but it takes a lot of mental energy and I become frustrated easily. I wouldn't like it if the Internet shut down and we had to ask real people for help all the time. And many people don't like that asking AI for help is becoming increasingly normal.

So, from the perspective of someone who uses the Internet primarily to exercise the brain, AI is a huge positive step forward.