ice cream

In a dimly lit room, a myroid sits, waiting with eyes closed. Long brown hair drapes over the back of the wooden chair like a curtain. The rest of the house is unlit; the inner corners of the room are deep blue; the tabletops are colored in shades and twilight.

“It is quite dreary around this time of year. I know that she told me to leave the lights off until she got home, but still-”

Suddenly an asynchronous event triggers in the myroid’s head, interrupting her current train of thought. It is 5:30, and it would be an estimated 10 minutes before her master would arrive home.

“User defined event; index 12. Labeled “Welcome home!”. Tasks for 5:30:
Turn on the lights for the kitchen and the hallway.
Prepare a snack.
Stand in the hallway while smiling. Upon seeing user, give them a message after [I’m home!!]”

She stands out of the chair, then pauses.
“A snack? Oh, what should I pick? [Please tell me what snack to choose.]”
A request fires off to one of the other processing units to do the heavy lifting. There are two FPGA units. These circuits are reprogrammable, and are subtly tweaked over time to reach optimal performance based on the tasks that are requested most. The myroid cannot directly access or hand tweak the other processing units, only the tuning algorithm based on scoring and feedback. Could the myroid learn some of the tools of the programming trade, merge sort, depth first search, and multilayer perceptions? Perhaps, but until then she is at the mercy of her programming, the black boxes bringing forth commandments and truths from what may as well be a machine god. The request chugs along, and she saunters left toward the kitchen, humming nothing in particular, as if trying to invent a new song.

The request returns.
A snack has the following properties:
It is edible.
It is palatable.
It is a small quantity of solid food.
It is generally used to satiate before a larger meal or for enjoyment.

This is a list of snacks, scored by availability (servings), amount eaten in the last 30 days, and estimated ‘tasty’ score (any item with 0 availability is omitted):
Banana              | 9     | 7     | 30.32
Sweet Tofu          | 34    | 15    | 30.01
Ice Cream (Vanilla) | 6     | 0     | 57.43
Red Bean Mochi      | 4     | 4     | 40.20
Apple(Granny Smith) | 2     | 4     | 12.09
Baby Carrot         | 6     | 1     | 5.00

“Hmm. Normally it just gives me a decision tree or a single choice. Was there something in the last update? Master hasn’t eaten ice cream in a while, but she likes to eat it. The tasty score tends to go down the more someone eats the same thing. Therefore, I think this is the best choice. What was this called again? A ranking algorithm?”

She pulls open the freezer, and takes out the ice cream.

“Oh! I almost forgot to turn on the lights!”

The myroid pivots sideways, walks to the lights, and flips one of them on. Lightbulbs affixed to a sideways, brushed metal bar make an almost imperceptible hum. The coils slowly turn orange, then yellow. The kitchen lights up to a warm glow.

“I always tell her to switch to LED lights, but for some reason she never gets to it...”

She turns back and kicks the freezer door. It closes with a dull slam.

“[I need the ice cream scooper. Where was it, again?]”
…
Locations and large serving spoons (from left):
1st drawer 	| Ladle
1st drawer 	| Small ladle
3rd drawer	| Ice cream scoop
…
“[Wait, I’ve got it. Ok, you can stop now.]”

The scoop comes out with a clank, rattling the other utensils. Two bowls are placed on the counter top. With a precision only a machine can bring, she plunges the scoop in, and creates an immaculate sphere of vanilla, until two sets of three scoops of ice cream are produced. She detects the sound of a key opening the front door. Scrambling towards the hallway, she nearly slips on the tiles, well known for having low friction with socks. The door opens with a creak. A blonde woman with red glasses and a labcoat steps in. She begins to loosen her yellow tie, the one Sarali picked out for her.

“Welcome home, Doc!”

“Hello, Sarali! I see that the schedule is working out just as planned. Did you bring out a snack, by the way?”

“Yep! It’s over by the kitchen. I scooped it perfectly!”

The woman’s eyes widen slightly in surprise. “Scooped? That can only mean one thing.” She thinks for a moment, taking mental notes. “You picked the ice cream as the snack, right?”

“Yes. It’s the best choice, isn’t it?”

“Sara-tan, next time I’d like you to add temperature to the list of attributes to weigh. It’s the middle of December, you know? It’s too cold to eat ice cream.”

Sarali pouted. “Eh? But don’t you sometimes eat ice cream in winter anyway?”

Doc paused. Sarali was right. How could she reconcile these two points of data, which must seem like contradictions to a machine? The whims of human beings can be odd and hard to predict, at times.

“Aha! I’ve got it. New rule: Only include temperature as a parameter if it is under 10 degrees C in this house.”

The problem of knowing exactly which factors matter for any given situation has vexed many a thinker.  She knew this was a sloppy, inelegant solution that only worked for this particular instance, but for the time being, such measures would prevent mishaps at the presentation she was to give in the days to come. Still, why did Sarali not consider temperature as an attribute? Was there not enough training data? Was the data there filtered out as if it were noise?

“Well, I’ll eat it today anyway. I guess it isn’t that cold.”

Doc gives a warm smile and puts her hand on Sarali’s head. Sarali feels a warm sensation as the hand moves back and forth gently.

Feedback score: 35.
The choice made was an incorrect one, but the main user is still satisfied with the choice. Positive feedback found via head pats. Refinement of temperature attribute to snack collection inbound.