John robbed a bank. Now he’s serving ten years. Writing in the New Yorker, Malcolm Gladwell points out that in the previous sentences, our brain fills in the information “John was apprehended, tried, and convicted.” Computers don’t do this. In fact, duplicating this ability to sift through huge stores of background knowledge and seamlessly plug story-line gaps is one of the major challenges of artificial intelligence.
But it also means that computers can sometimes think outside the box created by human expectations.
This was the case, Gladwell writes, when Stanford computer scientist Doug Lenat unleashed his program Eurisko on the Traveller Trillion Credit Squadron competition. In this competition, players are given hypothetical trillion dollar budgets, which they use to design navies. Then the navies battle mano a mano, with the winner advancing to the next round. Being human, we know that the best navies include a mix of cruisers, destroyers, carriers, etc., all strategically deployed and utilized. But Lenat’s program Eurisko didn’t know this. Instead, it read the rules of the game and designed the best strategy: a swarm of miniscule PT boats with big guns, no movement, and no defense whatsoever.
Eurisko cleaned up.
The next year, the competition changed the requirements, adding a minimum movement requirement. Eurisko used the same strategy, with exactly the minimum required movement. When a ship was damaged so that it brought Eurisko’s navy below the required movement minimum, Eurisko sank its own slowed ship.
Math Outsmarts Brains
The Bay of Biscay is off the west coast of France, south of the English Channel. During World War II, Royal Air Force patrols frequently spotted enemy submarines there. Certainly it was best to heavily patrol this area where the most submarines were found. Right?
It turned out that submarines were found here due to patrol density, not submarine density, and that patrol efficiency was higher in other areas. When RAF commanders sent patrols elsewhere, efficiency was increased and fewer planes were able to find more submarines. The field of Operations Research was born.
Eye Hack: AI Necker
The Necker cube (left) has been used to test visual recognition software: Can a computer figure out which face points out? Once a computer arrives at the conclusion, can it switch its opinion like the human mind? The impossible cube on the right just hurts our minds.
Eurisko cleaned up.
The competition organizers asked that Lenat refrain from entering the competition a third time.
It’s certainly cool to think about how computers show us the shackles of our conventional thinking. But we needn’t even look as far as the microchip for this. In fact, there are many human brains that don’t play by the rules. We label these brains ADHD or dyslexic or autistic. Should we medicate these brains back onto the bell curve, or does humanity need this novel thinking to help us innovate? Does society’s need for innovative thinkers outweigh the needs and rights of individuals living with challenging brains, who could live easier lives with medication? This certainly isn’t the first book to ask these questions.
