## Specific Criticism: Intelligence as Prediction

Eliezer has been very excited lately about his idea that intelligence is the ability to make accurate predictions and use those predictions to (partially) determine the course the future will take.

I have an objection to that.

Carl Friedrich Gauss is considered by many to be the greatest mathematical mind in human history. He was brilliant even as a child. One well-known anecdote in particular is often used to demonstrate his mathematical gift, and it is as follows:

Gauss and his classmates were given a tedious assignment in arithmetic that his teacher anticipated would keep them occupied. They were told to add up the integers from one to a hundred and report the sum. While his classmates scratched away, Gauss thought for a moment, then wrote down the correct answer.

He reasoned that adding the lowest number to the highest number, one to one hundred, results in one hundred one. Adding the next lowest to the next highest, two to ninety nine, also results in one hundred one, and so on. Continuing in that manner, there were fifty pairs of numbers that each added to one hundred one. Multiplying fifty by one hundred one, he found the sum of all the numbers: five thousand and fifty.

More importantly, Gauss’ reasoning demonstrated the general method for adding up any continuous series of integers: sum the highest and the lowest, then multiply by the number of integers divided by two.

Now, generating this method let Gauss solve the problem quickly and easily, which we might reasonably assume were goals of his. But when we hear this story, we don’t think of Gauss as brilliant because he recognized the method would solve his goals. If told of the method, presumably all of his classmates would also have so recognized. Any moderate intellect with an understanding of basic reasoning and rudimentary arithmetic could recognize its worth.

What made Gauss a genius was that he was able to analyze the problem and produce a method of solving it that was significantly less taxing and lengthy than the method that immediately occurred. His classmates could recognize that Gauss’ method was superior. But they didn’t think of it.

How is this an example of accurate prediction? How is this an example of directing the future?

I believe I can reasonably say that failing to accept Gauss’ realization as an example of intelligence flies in the face of our intuitive understanding of the concept. Yet his action is not encompassed by Eliezer’s suggested definition at all. Efficient resource use, counting cognitive effort and time as resources to be conserved, is a key aspect of intelligence. But the generation of methods to effectively and efficiently use resources is not predictive.

Thoughts?

November, 2008 at 12:01 am

“But the generation of methods to effectively and efficiently use resources is not predictive.”–but it is efficient cross-domain optimization, which is how Eliezer actually defined intelligence. Eliezer has written a lot about the importance of prediction unto rationalist, but I don’t think he’s ever defined intelligence as prediction, as you seem to mistakenly suggest here.

November, 2008 at 7:06 pm

Eliezer has defined ‘intelligence’ multiple times.

He’s also defined ‘optimization’ in ways other than common usage.

I think you’re going to have to do better than that.

November, 2008 at 7:11 pm

Z.M., please see Building Something Smarter.

He discusses his definition of ‘intelligence’ quite clearly, there. And it involves searching through possibilities to maximize preferred outcomes, not generation of knowledge of efficient possibilities.

November, 2008 at 10:58 pm

I’ve just reviewed “Building Something Smarter,” but I still think my countercriticism stands. The young Gauss was efficiently “steering the future” into an outcome where he had the answer to the problem. Eliezer already knows about efficiency.

*shrugs*

November, 2008 at 6:44 pm

But all of the other students were ‘also steering the future’ into an outcome where they would have the answer to the problem.

The key is ‘efficiently’. But it’s not that the other students were aware of alternative methods to solving the problem and failed to choose the most efficient one. They failed to recognize the existence of another solution method — one that turned out to be more efficient.

Before Gauss could evaluate the two methods and choose one according to some standard of which was ‘best’, he had to generate both methods as viable ways of reaching the answer. He was able to do that. The others were not.

The generation of the method is not ‘steering the future’. It is not predictive superiority. It is

comprehensivesuperiority.