Moravec's Paradox

If you haven’t read Eric Jang’s new book “AI Is Good For You” yet, you should.

I already did a short review of the book and initially planned to do a long, in-depth review, but I kept coming back to this:

Brief summary of Moravec’s Paradox: walking is easy for humans but hard for robots. Chess is hard for humans but easy for robots.

We don’t know why this happens.

A lot of people think it’s because humans evolved to be good at walking and other tasks over very long time-frames, while AI hasn’t had that much time to practice.

This is mostly the level of discourse about Moravec’s paradox that I see.

It’s fine, but (in my opinion) nobody has proposed anything that feels like a real approach to this other than “physical stuff is hard :( anyways here’s my new NeurIPS paper.”

Eric’s quote at least gives us something of substance to debate.

Finer-grain simulations, which are directly related to more computing power, might be all we need to have AI go from cute chatbot to genuinely useful robot.

Personally, I believe Moore’s law will eventually cause Moravec’s Paradox to give out.

Pixar had to play the waiting game in its earlier years to make Toy Story because the graphics weren’t good enough, and right now it seems like we just need a couple more years for the hardware to catch up to roboticists ambitions.

Perhaps Ray Kurzweil was right about his 2029 date after all…

Regardless, long NVIDIA.