Add Q-Learning to the list of activities where AIs are now at parity with the average human.
The very best humans are still currently better than the very best AIs. But I think it’s reasonable to expect that the best AIs will be much better than the best humans in only a matter of years.
This could be a problem though since Q-learning is a fully general way to deduce what rewards are in any environment and start winning them more.
Once that happens, the only thing keeping AIs from outcompeting humans in any given intellectual domain is simply better algorithms for pre-processing or feature detection. Or waiting for 4-5 doublings of Moore’s law so that today’s feature detection algorithms begin reliably scaling to real-world problems.
See on link.springer.com
One Response to “Comparing Humans and AI Agents”
November 27
Joshua FoxThe link here is paywalled. This seems to be a reporting on the same research http://users.dsic.upv.es/proy/anynt/paper1-comparing.pdf