When will there be AI superintelligence?
-
Other humans at least have the same kinds of bodies as I do, so by the principle of mediocrity (I'm probably not an exception), if I have a mind other humans probably do too.
We can't assume that with machine intelligence. It is by definition very different from me. Everything I know about AI suggests that what's going on inside the machine is very different from what goes on inside me.
Ultimately, I think it kind of doesn't matter. If machine intelligence can communicate and behave as if it "really" understands its percepts, I don't know if I care whether it truly is thinking or not.
-
I'm pretty dubious about claims that superintelligence is so immanent, and so dangerous, that LLM technology should be restricted to only a few trusted companies and the US government. That sounds like some self-serving bullshit by companies that want to consolidate economic power by partnering with a protectionist government.
@evan Definitely an attempted ladder-pull, before they IPO.
-
I'm not convinced that they will keep growing at that rate - I think we are fairly well along the road of diminishing returns for dollars-to-model-quality.
Yes, models are getting better. I agree.
But I don't think that it's currently true that if you spend 10x as much, you train a model that is 10x better. I'm not going to pretend that I know what the constants of this function are, but I think they are sub-linear. I think that a scenario of ever-increasing amounts of money spent for ever-decreasing marginal improvement is going to make further linear increase in models unlikely.
I'm also not convinced that number of neurons are a great proxy for intelligence - both in terms of comparison to synapses, and in terms of linear scaling. At some point, it used to be easy to convert more transistors on a chip into a commensurate amount of computing but now ... it's not so linear. I don't know if there is such a point of diminishing return for neural nets, and if there is, what it would be but - my hunch is that it likely exists. There are not a lot of things that scale linearly forever.
Thanks for this thought-provoking thread!
@ricci thanks for the thoughtful reply!
-
@evan "they serve about the same function"
Citation? Naming them neural nodes doesn't count
@davep I'm sorry I had an angry response to this earlier. I shouldn't have flown off the handle. I read your post as much more confrontational than it is on second reading.
-
Dude is pretty open about their methodologies and the danger.
I don't see how their monitoring can be quick and comprehensive enough. They trained Astra in a data center with 100K GPUs. A half million tflops/s.
-
@davep I'm sorry I had an angry response to this earlier. I shouldn't have flown off the handle. I read your post as much more confrontational than it is on second reading.
@davep I don't know if I can provide a citation that params play the same role in LLMs that synapses play in biological brains. I only used it by very bare analogy; they are the "connections" between parts of the network. And from the little I know, I think they're much less complex than biological synapses.
-
A lot of the replies say that it will never happen. My guess is that some significant portion of the majority of people who answered 2060 or later mean "never" or "in the far future".
@evan I might have voted if there was a "stop drinking the poo-flavour koolaid" option
-
@davep I don't know if I can provide a citation that params play the same role in LLMs that synapses play in biological brains. I only used it by very bare analogy; they are the "connections" between parts of the network. And from the little I know, I think they're much less complex than biological synapses.
@davep it's perfectly ok with me if this back of the envelope calculation isn't convincing to you. I'm sharing my reasoning for coming to the answer I would give to this question, not trying to convince you.
-
@evan I might have voted if there was a "stop drinking the poo-flavour koolaid" option
@jpaskaruk why's that?
-
Ultimately, I think it kind of doesn't matter. If machine intelligence can communicate and behave as if it "really" understands its percepts, I don't know if I care whether it truly is thinking or not.
@evan I think, and hope, the biggest impediment to that exponential growth of “intelligence” will be raw resources. Power, material, and water.
These data centres are already hoovering up precious minerals and complex components so much that it is causing serious shortages. They are already consuming magnitudes of electricity to require their own generation, or sucking power from communities. They are already consuming so much water that concerns are rising for capacity and environmental impact.
And then there is the financial capital.
We have never seen a technology this *hungry* before. Short of turning humanity itself into batteries, I don’t think there will be true intelligence before these issues are solved. So my answer would be beyond 2050.
-
@sandriver do you think intelligence can only arise in brain-like systems? Or is that just our easiest path to getting there, since we have an existence proof of it?
@evan@cosocial.ca just as an aside, I think it's important to distinguish cognition and intelligence, since obviously even trivial neural nets have some level of cognition, as do life forms that don't even have a nervous system, or only exist as single cells.
As to whether hardware needs to be brainlike, I lean towards the existence proof, but also with the proviso that neural net software is too abstract and deliberately elides potentially important physical properties of a brain. Some examples:
* neurons have chemical signalling that propagates at the speed of sound, in a volume, beyond the synaptic transmission.
* neurons have internal state due to their epigenome.
* neurons have complex, mesh-like connections between anatomical regions.
* neural information is inherently sensitive to time in a variety of ways, including those mesh-like connections.
* non-neural tissue in the brain also affects neuronal behaviour.
* there are unknown nonclassical properties of neurons.
Maybe you could throw enough matrix munching transistors at a neural net and get a life-like mind that is superior to life in all domains, but I doubt it, not with current software models of neural nets. They are only able to reproduce particular kinds of cognitions, and certainly nothing like a mind. -
@evan@cosocial.ca just as an aside, I think it's important to distinguish cognition and intelligence, since obviously even trivial neural nets have some level of cognition, as do life forms that don't even have a nervous system, or only exist as single cells.
As to whether hardware needs to be brainlike, I lean towards the existence proof, but also with the proviso that neural net software is too abstract and deliberately elides potentially important physical properties of a brain. Some examples:
* neurons have chemical signalling that propagates at the speed of sound, in a volume, beyond the synaptic transmission.
* neurons have internal state due to their epigenome.
* neurons have complex, mesh-like connections between anatomical regions.
* neural information is inherently sensitive to time in a variety of ways, including those mesh-like connections.
* non-neural tissue in the brain also affects neuronal behaviour.
* there are unknown nonclassical properties of neurons.
Maybe you could throw enough matrix munching transistors at a neural net and get a life-like mind that is superior to life in all domains, but I doubt it, not with current software models of neural nets. They are only able to reproduce particular kinds of cognitions, and certainly nothing like a mind.@sandriver thanks for this!
-
@bignose why are you confident in that? LLMs are great at a lot of things that computers haven't done well previously.
@evan
> LLMs are great at a lot of things that computers haven't done well previously.I hear this claim a lot, but it's never substantiated verifiably.
In fields where I understand the LLM output, it is unreliable plagiarism or unreliable garbage.
In fields where I need to rely on others more knowledgeable, those who *aren't incentivised to hype the LLM* report LLMs are unreliable plagiarism or unreliable garbage.
They're okay at language patterns. What are you saying they're great at?
-
I don't think we know for sure a lot about intelligence. We don't know that it can be implemented on silicon chips, but we also don't know if it can't. We don't know if it has to work like brains do, or if it could work in entirely different ways. We don't even know exactly how our own minds arise from our brains. We don't know if we can replicate intelligence without understanding our own first.
@evan I agree we don't know much about intelligence - in the way that we tend to confuse it with other potentially very different phenomena. One thing that often gets confused with it is self-consciousness. Another thing that is not synonymous with either of those is intention. Psychologists working on those concepts have interesting things to say about all three, but it doesn't get through in the public yet.
