Re:Fake Dog for Home Security
by mlcrypto on Wednesday July 27, 2022 @05:48PM
(#32257478)
That reminds me of the argument about sentience lately. Why can't a neural network be sentient if all our actions have been trained by interacting with the world since birth?
[ Parent
| on HN ]
The operative word in your comment being "actions". What "actions" can (current generation) neural networks undertake? Can they initiate anything? Can they choose their own learning material? Can they even choose when or whether to repeat a certain training set?
There was a recent HN comment about this that I think illustrates the point well [0]:
experimentation is an act on the world to set its state and then measure it. That's what learning involves. These machines do not act on the world, they just capture correlations.
Wouldn't you say that a neural network that learns to play a game by playing the game does so via experimentation?
A lot of human "choice" is actually just a neuron firing or chemical in the brain going off
Re:Fake Dog for Home Security
by uj8efdjkfdshf on Wednesday July 27, 2022 @07:02PM
(#32258220)
Because most people define sentience as 'a black box that reacts meaningfully to inputs in a sufficiently unpredictable and inexplicable manner', and for most people (especially researchers in these fields), current neural networks do not meet this bar.