"I walk slowly on the race track, but I can climb a ladder"

Collin Lysford, in a recent issue of his Desytemize newsletter, eases into a thoughtful discussion about intelligence (artificial and otherwise) through a baking analogy.

Here's what I highlighted:

AI does a lot of things that we call “thinking” when we do them slower and worse. The fact it can also do those things should make us humble and curious, not proud and dismissive. But I think it’s equally silly to lump all these capacities together into “intelligence” and say “Intelligence is going up, so soon it will do everything intelligence can do.”

The way through this dilemma is to notice that the thing you’re looking for is not a skill that’s literally impossible for machines. You’re looking for knowledge that needs to be gathered the slow, iterative human way and not the fast, recursive, copied with perfect fidelity machine way. Allow your humility to say that a machine living in the world and subject to the same selection pressures as you could eventually reach a better place than you. The material question is how well the selection pressures can be simulated by sufficient compute and compressed into a much smaller time frame than the millions of years it took us.

But are intelligences developed by other intelligences, or are they developed by environments? We haven’t had to think about this distinction because we’re stuck in the environment and everything we do is mediated through it. All of our intellectual endeavors have a sense of place because: we’re here. But the “A” in “AI” stands for “artificial”, ie, not developed by an environment. Now that some intelligences are non-environmental, we have to consciously understand the intelligence/environment dynamics in ways we previously could lump together and call it a day.

I don’t believe that human intelligence is “superior” to machine intelligence per se. I think that it’s a category error to think in those terms. We tend to do much worse at things we’re both trying to do under the same conditions (like games), but we’re operating under much stricter constraints instead of playing in the consequence-free sandbox. It’s like asking whether walking is faster than driving a car. Not on a well-groomed road, but clearly there’s lots of terrain where a car simply doesn’t work at all. I walk slowly on the race track, but I can climb a ladder.