Search interesting materials

Saturday, November 21, 2015

There will be no pigeon singularity

Pigeons were trained to behave like skilled doctors. When a computer is trained to behave like a skilled doctor, we think this is quite portentous. But when a pigeon does this, we never fret about the coming pigeon singularity.

The computational hardware in the pigeon's head is some bunch of neurons. Using rewards and punishment, we are training this neural network. Training neural networks gives the ability to do specific tasks. However, no amount of hardware for special purpose tasks (pigeons or dogs or Watson) closes the gap which separates the machine learning revolution from general intelligence, creativity and consciousness. It is genuine progress in engineering when dogs can smell abnormal blood sugar or when computers can drive cars. These do not, however, add up to strong AI.

4 comments:

  1. Our brain too is a neural network.... a skilled doctor became so via similar learning in the doctor's brain (neural network). In the hardware collection mentioned above, one should consider and add the human brain: ie; (pigeons or dogs or Watson or the human brain).

    ReplyDelete
  2. Secondly, general intelligence can be considered as nothing more than a collection of specific purpose tasks. If we take a cluster of machines and expose them to all the stimuli and rewards that a baby experiences till the time of becoming a teenager. we could well end up with general intelligence. Perhaps we would need to add some general goals like the survival instinct (which machines and specific purpose neural nets do not have) to end up with general intelligence. Perhaps we would need to delve deeper beyond the survival instinct to see what secondary goals a human being aspires too if survival is not threatened (like arts, sports, etc) and why.

    ReplyDelete
    Replies
    1. Actually, we would need more than the survival instinct as pigeons/rats have it too. I guess we would need to mimic evolution. So, when we have the capacity and capability to train a neural network with millions of years of evolution data (or a pruned/accelerated data set that encapsulates the same) then perhaps we will get general intelligence. Anyway, my point is, its a bit early to say that the same concepts with the same or similar hardware cannot result in general intelligence. It could be solved with advances in the software domain with more training data.

      Delete
  3. AI would form a symbiotic relationship with humans- man machine interface. The Innovators by Walter Isaacson provides this perspective, albeit briefly. No question of singularity there!

    ReplyDelete

Please note: Comments are moderated. Only civilised conversation is permitted on this blog. Criticism is perfectly okay; uncivilised language is not. We delete any comment which is spam, has personal attacks against anyone, or uses foul language. We delete any comment which does not contribute to the intellectual discussion about the blog article in question.

LaTeX mathematics works. This means that if you want to say $10 you have to say \$10.