Moravec was actually horribly wrong in his reasoning why,
As reasoning turns out to require incredible compute, as modern LLM approaches are making clear
(He was referring to symbolic methods, which are incredibly compute efficient, but those were ultimately not the right path, at least not by themselves)
Still, the high-level idea remains:
AI advances at a rate of moore’s law, doubling every ~18 months
Robotics advances at a rate that is much, much slower.
Robotics started with a head start,
But that head start eventually gets crushed by the faster doubling rate
We’ll see AI that crushes humans in all thinking activities before we see robotics that crush humans in all motor activities.
As reasoning turns out to require incredible compute, as modern LLM approaches are making clear
(He was referring to symbolic methods, which are incredibly compute efficient, but those were ultimately not the right path, at least not by themselves)
Still, the high-level idea remains:
AI advances at a rate of moore’s law, doubling every ~18 months
Robotics advances at a rate that is much, much slower.
Robotics started with a head start,
But that head start eventually gets crushed by the faster doubling rate
We’ll see AI that crushes humans in all thinking activities before we see robotics that crush humans in all motor activities.
✍2