AI, to the level that we attempt to create it these days, involves creating neural networks that learn from stimulus (experience, data) and reinforcement. A higher score such as in a Go game, or a higher % of matched words, faces or signs seems to equate to positive reinforcement within classical conditioning. Given that behaviourism is mostly seen as somewhat outdated, if the above makes sense, could this model of learning be limiting progress in AI?

  • $\begingroup$ While I agree with this general premise as I have been expressing a similar idea myself for some time -- that current AI research is too much like mimicry or cargo-cult thinking -- whether or not this approach is limiting research is probably difficult to answer since it implies knowledge of what is currently not known, at least not widely. Specifically, depending on how difficult it is to understand and reverse-engineer human thinking, it may actually be easier (and hence more productive) to mimic the external behaviour of the mind. $\endgroup$
    – Michael
    Feb 8 '16 at 9:06
  • $\begingroup$ Nevertheless, if I were to begin researching AI myself, I think I would begin more from the theoretical side and less from the mimicry side. The problem is that the public would rather have a prototype that appears human (in language and or body), so money is more likely to be invested into mimicry than into other approaches. $\endgroup$
    – Michael
    Feb 8 '16 at 9:09
  • 1
    $\begingroup$ could you give a definition/background of what behaviorism is, why it is seen as outdated, and why you think AI research is participating in it? $\endgroup$
    – honi
    Feb 18 '16 at 17:49

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.