Re:Can AI Become Conscious?
by proverbialbunny on Tuesday May 12, 2020 @09:08PM
(#23162253)
>But are your features really sentient self-symbols, or are they just mechanically derived by-products - like reflection in a computer language, taken up a level?
Why couldn't we apply the concept to our self?
Self awareness is one commonly believed description for consciousness. Going off of that, we become self aware in three primary ways: 1) We're told about us by our parents at a young age, before the age of one. 2) We observe our selves in the environment often influencing the environment, from looking in a mirror, to social interaction. 3) We get meta about a series of things (body+mind) and self awareness arises from that generalized intelligence.
Though, you could argue that #3 may not be enough alone to create self awareness and ultimately #2 is required.
But it's also quite hand-wavey, because RNNs have no real hooks for self-reflection.
You can strap a meta-RNN across a working RNN, and if you're lucky it may start to recognise features in the operation of the working RNN. But are your features really sentient self-symbols, or are they just mechanically derived by-products - like reflection in a computer language, taken up a level?
You need at least one extra level of feedback to do anything interesting with this, and it's really not obvious how that feedback would work.
It's even less obvious if this will give you "genuine" experiential self-awareness, or just a poor and dumbed-down emulation of it.