Thoughts on AI After a Month of Stanford

I don’t have many long stretches of time where someone doesn’t mention the existential crisis that many people in intellectual fields are going through: whether AI will take their jobs or not. And what's crazy to me is how dystopian these people make the future out to be. I think that even if we extrapolate timelines based on current rates of progress, there’s still so many interesting problems that will need to be solved post-AI.

At the highest level, I don’t believe AI is the last “big” thing. Problems like inference, robotics, energy, or even world hunger will persist, and will require significant innovations beyond the current state-of-the-art systems. AI is currently, and has been, really really good at solving well-defined problems in the digital world, but as soon a problem does not satisfy these two properties, AI falls flat.

At a slightly more controversial level, the thing that these models are optimizing for does not make them capable of solving problems in the real world. Humans are, pretty uniquely, actually living in the real world. They can interact with other humans to find issues in the world, and then have the agency to chase solutions to them. Yes, AI will make this process a lot faster, however there’s a pretty key distinction. While AI can solve the problems, humans will always need to iterate on the definitions of them.

And at the deepest level, the current models aren’t even all that good about reasoning at macro scale problems. Just empirically, I think they struggle to build scalable systems without expert supervision, and fully AI-built apps have a reputation for not being that stable. At the very least, infra teams will still exist for some time to come. But even beyond that, reasoning about system level interactions to build something new is a significant limitation of AI, even in the digital space.

In short, current models are like really, really good machine guns. They have great aim, fast iteration, and do a lot of damage. But in the hands of the untrained, they are useless.