Scott is worried about super-intelligent AI and this was my comment.
What if the super-intelligence is already evolving (literally) in front of our eyes? I am talking about the internet.
It already has all the data and plenty of intelligence digesting it.
Already we cannot “turn it off” although it begins to have all kinds of unintended side effects.
It is still in an early stage, like a kid, who needs parents to feed it, and it still needs us to keep it going and learning and grow up.
But from time to time it already throws temper tantrums like a teenager, while we connect more and more of our devices and indeed all our lives to it.
It already recognizes our faces and identifies our voices.
It watches our calls and it begins to bankrupt long established companies, it destroys relationships and it has become an arena for "cyber wars" of all kind, which we can
control less and less.
In a few years it may begin to talk back, make recommendations and give out orders that we better follow or else ...
We still mostly think that we are using the internet and not the other way around, but some governments are beginning to suspect that slowly but surely we are losing control.
added later: some links ...
About the title.
AlphaZero plays Go, chess and shogi.
SILK
AI bots as hackers
singularityNet
How brainless slime molds redefine intelligence.
14 comments:
Indeed.
CIP,
a while ago you asked about Deep Learning and trading.
Meanwhile I can confirm that there is at least one DL trading program active in the US stock market 8-)
what's so special about SILK? Nowadays "AI" companies are mushrooming. Give a child a hammer and everything becomes a nail.
Nothing special, just the first link that came up when I searched for AI and IoT at Google (actually startpage). In other words, the link was selected by an unknown AI/ML algorithm ...
Btw I like this SILK use case: "Detect when your baby is crying to trigger a lullaby."
In the future, AI internet will take care of our babies ...
You might be interested in this one too:
https://www.theverge.com/2015/3/16/8223251/hello-barbie-speech-recognition-privacy
not sure though if it ever got to the market.
"Even if Mattel and ToyTalk aren't using this technology to feed data to advertisers, they're still hoping to manufacture deep, emotional bonds between children and their toys." - not creepy at all.
Wolfgang,
Do you know if the AI experts think super intelligence inevitably leads to consciousness? I don't see any reason why there couldn't be a controlling super intelligence that is completely unaware of it's own existence. Any life is pretty intelligent, but I think most life is unaware of its own existence.
Lee,
I think "consciousness experiences" or the question if an intelligent AI would "really feel" something are topics almost all AI researchers do not discuss (as part of their job). This is certainly true of the AI researchers I talk to; we would not discuss if a trading system actually "feels" something.
>> unaware of it's own existence
I think every AI system, if it is complex enough, needs to some extent "be aware" of it's own existence. E.g. an autonomous car needs to "know" its own geometric properties, its own speed etc. in order to get the job done. But this is not "awareness" that we mean - unfortunately I do not know what we really mean with "awareness".
I suspect that a mathematical description of consciousness does not exist and therefore it is not part of physics and also not part of computer science. Most scientists conclude that therefore consciousness does not exist. I think it means that there are aspects of reality beyond mathematical description - nothing more, nothing less.
But every now and then it is (implicitly) assumed, e.g. in the Copenhagen interpretation.
>> Most scientists conclude that therefore consciousness does not exist.
Possibly, but I remember Scott Aaronson arguing pretty convincingly back in 2014 that phi, as proposed by Tononi, doesn't pass the common sense test of what we mean by consciousness. I think it's hard to argue that consciousness does not exist or is not part of physics if one believes that in principle the human brain should be able to be modeled by a Turing Machine.
I assume that a human brain could in principle be modeled by a Turing machine and more realistically, a human brain could be modeled by a neuronal network (perhaps with more intelligence in the neurons than currently assumed).
But we don't know if and what such a Turing machine would "feel" and I assume that there is no way for us to know, i.e. I think there is no mathematical description, or a description in any language (including python and lisp 8-) and no possible physics experiment that would tell us.
One possible conclusion would be that a Turing machine just does not "feel" anything (and it then follows for many that our own human "feelings" also are not really real).
However, I would argue that there is one important aspect of reality, the "emotional experience" of complicated enough Turing machines and animal brains, which we have no access to and which we cannot describe, but nevertheless it is part of reality.
It follows that mathematics/physics is not a complete description of reality and furthermore I assume this is where the interpretation problem of quantum physics ends ...
Btw if it makes it easier to follow my ideas, I think that "feelings" are somewhat parallel to brain mechanics and/or Turing programs, but do not cause or affect anything in the reality described by physics. Therefore one can leave it out from any calculation in physics.
So in order to finally answer your question: In my opinion one does not need to understand consciousness to develop a super-intelligent being (and indeed we are already developing such a being with the internet ! ) , but if we succeed the super-intelligent being will have a "conscious experience" etc. but we will have no understanding of that part of its reality.
We will never know what it feels like to be the internet, just like we dont know what it feels like to be a bat.
One more comment , Tononi's phi and other debates are about the question what physical systems do actually have a "conscious experience". We would like to have a function phi so that e.g. phi( brick wall ) = 0 and phi( dog brain ) = 0.8 and phi( human brain ) = 1 and we would then use it to calculate phi( AI ) )
My point is that even if we would have such a function phi (currently we do not) , it would not give us any description of what the conscious experience "feels like" and if it is in any way similar to ours ...
>> I think that "feelings" are somewhat parallel to brain mechanics and/or Turing programs, but do not cause or affect anything in the reality described by physics.
Not that it makes any difference, but I agree with everything you said in your two comments above, although I am a little uncertain as to what you mean by what I quoted above.
One last thing. Do you think your ideas are consistent with Deutsch's idea that "everything that is not forbidden by laws of nature is achievable, given the right knowledge?"
Yes, but it is not clear if it is achievable by us right now or in the near future.
Post a Comment