Moravec's paradox claims tasks hard for humans are easy for AI and vice versa, but this has never been empirically tested. The paradox reflects which problems AI researchers find interesting rather than having predictive power. Its evolutionary explanation is dubious, and it has led to both alarmism about superintelligent reasoning and false comfort about robotics progress. Computer vision, once considered hard for AI, saw breakthroughs with deep learning around 2012. Rather than trying to predict AI capability breakthroughs, we should focus on adapting to technologies we know are coming, as deployment takes considerable time even after technical breakthroughs.
Table of contents
The evidence behind the paradox is flakyA flawed evolutionary argumentHow simplistic models have misled AI researchers and tech leadersConclusion: if not Moravec’s paradox, then what?Other videosSort: