A common argument against the current level of AI I often hear is "A child can recognize a dog after seeing it once, whereas it takes a model thousands of images".

This makes sense on the surface, but a child has ~1-1.5 years of "streaming video" before it can do this. Additionally, the first time it sees a dog he/she is probably getting a lot of dopamine due to interaction with dog and positive reinforcement from parent.

So, two questions stemming from thing.

  1. Does the initial argument hold any water?
  2. Is there any research being done to tie reinforcement learning and supervised learning in that trained model acts as reinforcement for a newly learning model? This would seem to model "real" learning a bit more.
  • 1
    What do you mean this is an argument "against the current level of AI"? If you just mean that it's an argument AI is below human levels of intelligence or generality, then it's obviously correct. If it means something else, I'm not sure what it would be.
    – Veedrac
    Feb 24 '18 at 4:51
  • 1
    It's not clear to me that there is a clean way to compare datasets to raw human perception, though you might find that's a reasonable question for the Psychology & Neuroscience Stack Exchange.
    – Veedrac
    Feb 24 '18 at 4:55
  • I was specifically referring to Wozniak's flip-flop... And I've heard it from other folks as well.. cnbc.com/2018/02/23/… It's not that the argument is wrong.. it's that this is an argument used to say AI is no way close to human G.I. ... but I'd say it's a narrow perspective in that it really only focuses on supervised learning.
    – cloud36
    Feb 24 '18 at 5:03
  • Supervised learning you say? How does unsupervised learning strike your fancy? Perhaps something akin to beating top world players and engines in Go, Chess, and Shogi, starting from nothing more than the rules and self play? If that sounds interesting to you, how about a ten year old YouTube video on the tech?
    – H Walters
    Feb 24 '18 at 7:28
  • @cloud36 Wozniak's argument is a poor one since AI risk is not about today's AI.
    – Veedrac
    Feb 24 '18 at 11:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.