Squasher

Squashing squashed biweekly
Thu Mar 26

We’re Bad at Math

squashed:

In the smoking aftermath of the Robot Wars, we’ll look back and wonder why we couldn’t get along with our creations. Didn’t we program them to love us?

I think it will be because we’re bad at math. At least, we’re bad at the kind of math computers are good at. While we may not be irrational, we don’t always think in terms of precise numbers. If I set my coffee maker to make coffee at 8:00, it makes coffee at exactly 8:00. If I make coffee at 8:00 in the morning, I make it at approximately 8:00. If I start brewing the coffee at 8:02, it’s essentially 8:00 to me because the two times are about the same. To the coffee maker, on the other hand, 8:02 is a different time than 8:00—and no more like 8:00 than it is like 12:37. We’re much better at analogy than we are with manipulating numbers.

It wouldn’t be too hard to program a computer to ace the math portion of the SAT. The biggest difficulty would be input. But could the computer manage the analogies on the verbal section? Tree is to timber as sheep is to a) mailbox b) mutton c) stable d) goat. It’s easy for us—but programming a computer to manage that is incredibly difficult. We are much better at thinking by  analogies than we are mathematically.

This is one of the difficulties neoclassical economics runs into. Value is caluculated numerically. While we’re capable of doing this—we usually don’t. If we buy a $5 drink and absolutely hate it, we might insist on drinking it because we paid for it. If $5 falls out of our pocket, we might spend the entire day being upset about it.

Actually, as a a computer scientist, I’d like to refine that: Natural language (As in English instead of C++) is extremely hard. Logic is easy, but only if you know the rules in advance. The math puzzles are given in natural language, and the format is different between problems—some are algebra (easy) some are geometry (harder) and some are word problems (practically impossible). The analogy sections though have consistent format and consistent rules. All you need is a list of the top 20 analogies (opposites, part to whole, etc) and an enormous dictionary with the rules between each word. Very time intensive to enter, but possible. It’s the reading analysis that’s impossible.