> The program must fit in a genome-sized instruction set of about 1 gigabyte
True but this can be a lot larger, maybe even a 1 exabyte instruction size since it depends on he programming language used to express the program. And even accounting for the invariance of runtimes at scale*, the constant factor might be gigantic since we have to build up a lot of cellular machinery first. Reminds me of the Carl Sagan quote: To make a sandwich,you must first construct the universe!
* For example, to translate a program from language A to language B, you can dedicate a constant size to write a language A to B translator. Thus at large program sizes , the kolmogorov complexity (i.e instruction set size) is fairly similar between programs since the size of the program dominates the size of the translator program which is constant. But the constant factor for cellular machinery might be gigantic**
I have this science fiction story in my head that consciousness can evolve from organic to mechanical and back to organic as it is perhaps the most energy efficient method to encode complexity.
This isn't quite the question the article is asking, but if you expand it just a bit to a brain rather than a simulation of a biological brain specifically, the specification complexity of a modern AI is not that large. You can get a good idea of the foundation of the system with some 3blue1brown videos. While just watching those and deeply understanding isn't enough to score a job at a frontier AI lab, I'd expect that a lot of what the labs do is similarly not that complicated to specify... it's coming up with the ideas, tuning them, and then trying to test them at the now-staggering scale it takes to prove them out for commercial use that is most of the issue.
The complexity of the AI itself, the final weights, is vastly higher, because that incorporates all the data that was flowed through the "brain".
In a decent encoding you could probably still fit the initial state of the frontier models comfortably in just a few kilobytes of code-golfed code, including the update functions and every necessary for the actual training. Recall the specification complexity is the smallest program that can create the initial state, not the initial state itself; the specification complexity of "give me a trillion 64-bit numbers generated from this psuedo-random number generator" is on the order of that very English sentence in size, not 8 tebibytes of specification complexity.
Is the result a biological brain? Obviously not. Biological brains do not do what the LLMs are doing. But unlike someone asking this question 20 or 30 years ago, where "yeah, but what if some other architecture could work too?" was still largely hypothetical, and neural nets were still largely toys, now it isn't. We may argue if an LLM is as smart as a human but I would say that at least on its playing field it is quite clearly smarter than most biological brains in existence on most measures we care about. I have to qualify "on its playing field" because it is fundamentally a text completion engine, so for instance even very very tiny biological systems still beat it on things like "ability to drive an ant body around to do useful tasks". That's a separate field of AI right now. Stay tuned on that matter, but it's not how things work today for sure.
There is still something to what biological brains do that we are not matching; as I like to say, humans do what they do without the entire contents of the internet being poured through their head multiple times over. But the idea that maybe we don't have to exactly match biology to get something useful is no longer just a theory.
Everything is subject to quantum mechanics. If you meant something like a non-trivial quantum interaction that is nor modeled in the chemical and electrical theories of the brain presently, still, as far as I am aware quantum effects don't need anything beyond Turing to simulate. If I am wrong, someone let me know.
> LLMs suggest that human-like "cognition" can be somewhat mimicked by purely mechanical means.
This is not a strike against that hypothesis. Quantum systems can be simulated using non-quantum computers but that doesn't make non-quantum computers into quantum computers. Similarly, mimicking some aspects of human cognition doesn't mean you can replicate all of human cognition with the same method.
when mice were exposed to cosmic ray bursts (to test for effects on past and future astronauts traveling to moon and mars)
the mice learned much much more slowly and forgot things much much faster
how's that for "cognitive corruption" ?
(btw this is also why I laugh when Musk thinks he's going to Mars ever, it's not just a one-way trip, he wouldn't even survive the journey itself with current technology and unlike earth there's no magnetic field or enough atmosphere on Mars to deflect them)
> unlike earth there's no magnetic field or enough atmosphere on Mars to deflect them
These are solvable with known methods.
They make the whole thing rather less romantic.
If you're going to the trouble of e.g. burying your base under regolith or wrapping a planet in a big coil and lots of PV to make an artificial magnetic field, it very quickly becomes apparent that for essentially all possible disasters short of "oops we made a black hole and it caused a magnitude 15 earthquake and all the crust everywhere liquified and mixed with all the water so even ocean bases weren't safe", it's much, much easier to survive disasters on Earth without actually leaving Earth, than to go to Mars and have a normal day.
If someone made a robot factory that went mad and converted all the nitrogen in our atmosphere into ammonia (killing all the plants that no longer had access to the atmospheric nitrogen), while also producing enough smog to halve how much sunlight we got, and dumped a lot of endocrine disruptors into our water cycle, the top of Mount Everest and the Sahara desert would each still be much more hospitable than Mars is today.
LLM emulate cognition, there is a huge difference, it's a mimick process not genuine
if you read about psychopaths/sociopaths, they do not comprehend emotions but they can learn to emulate them by watching others and mimicking them
that's exactly what LLM is doing, it is taught to assemble words in a way that mimics all the humans in its huge datapool, so it comes off as "cognition" because that's what the humans it's emulating were doing
True but this can be a lot larger, maybe even a 1 exabyte instruction size since it depends on he programming language used to express the program. And even accounting for the invariance of runtimes at scale*, the constant factor might be gigantic since we have to build up a lot of cellular machinery first. Reminds me of the Carl Sagan quote: To make a sandwich,you must first construct the universe!
* For example, to translate a program from language A to language B, you can dedicate a constant size to write a language A to B translator. Thus at large program sizes , the kolmogorov complexity (i.e instruction set size) is fairly similar between programs since the size of the program dominates the size of the translator program which is constant. But the constant factor for cellular machinery might be gigantic**
** Or not if we only need a rough simulation
You can make a sandwich with two slices of bread and whatever filling.
The complexity of the AI itself, the final weights, is vastly higher, because that incorporates all the data that was flowed through the "brain".
In a decent encoding you could probably still fit the initial state of the frontier models comfortably in just a few kilobytes of code-golfed code, including the update functions and every necessary for the actual training. Recall the specification complexity is the smallest program that can create the initial state, not the initial state itself; the specification complexity of "give me a trillion 64-bit numbers generated from this psuedo-random number generator" is on the order of that very English sentence in size, not 8 tebibytes of specification complexity.
Is the result a biological brain? Obviously not. Biological brains do not do what the LLMs are doing. But unlike someone asking this question 20 or 30 years ago, where "yeah, but what if some other architecture could work too?" was still largely hypothetical, and neural nets were still largely toys, now it isn't. We may argue if an LLM is as smart as a human but I would say that at least on its playing field it is quite clearly smarter than most biological brains in existence on most measures we care about. I have to qualify "on its playing field" because it is fundamentally a text completion engine, so for instance even very very tiny biological systems still beat it on things like "ability to drive an ant body around to do useful tasks". That's a separate field of AI right now. Stay tuned on that matter, but it's not how things work today for sure.
There is still something to what biological brains do that we are not matching; as I like to say, humans do what they do without the entire contents of the internet being poured through their head multiple times over. But the idea that maybe we don't have to exactly match biology to get something useful is no longer just a theory.
which would make it impossible to emulate via current technology
excellent PBS Space Time on that
* https://www.youtube.com/watch?v=xa2Kpkksf3k
- LLMs suggest that human-like "cognition" can be somewhat mimicked by purely mechanical means.
- Exposing the brain to an MRI would affect cognition if there was a quantum element to it.
This is not a strike against that hypothesis. Quantum systems can be simulated using non-quantum computers but that doesn't make non-quantum computers into quantum computers. Similarly, mimicking some aspects of human cognition doesn't mean you can replicate all of human cognition with the same method.
when mice were exposed to cosmic ray bursts (to test for effects on past and future astronauts traveling to moon and mars)
the mice learned much much more slowly and forgot things much much faster
how's that for "cognitive corruption" ?
(btw this is also why I laugh when Musk thinks he's going to Mars ever, it's not just a one-way trip, he wouldn't even survive the journey itself with current technology and unlike earth there's no magnetic field or enough atmosphere on Mars to deflect them)
These are solvable with known methods.
They make the whole thing rather less romantic.
If you're going to the trouble of e.g. burying your base under regolith or wrapping a planet in a big coil and lots of PV to make an artificial magnetic field, it very quickly becomes apparent that for essentially all possible disasters short of "oops we made a black hole and it caused a magnitude 15 earthquake and all the crust everywhere liquified and mixed with all the water so even ocean bases weren't safe", it's much, much easier to survive disasters on Earth without actually leaving Earth, than to go to Mars and have a normal day.
If someone made a robot factory that went mad and converted all the nitrogen in our atmosphere into ammonia (killing all the plants that no longer had access to the atmospheric nitrogen), while also producing enough smog to halve how much sunlight we got, and dumped a lot of endocrine disruptors into our water cycle, the top of Mount Everest and the Sahara desert would each still be much more hospitable than Mars is today.
LLM emulate cognition, there is a huge difference, it's a mimick process not genuine
if you read about psychopaths/sociopaths, they do not comprehend emotions but they can learn to emulate them by watching others and mimicking them
that's exactly what LLM is doing, it is taught to assemble words in a way that mimics all the humans in its huge datapool, so it comes off as "cognition" because that's what the humans it's emulating were doing
LLMs don't have conscious processes that perform that kind of reasoning.
If anything the fact that they are blackboxes to their own behavior and are able to non-verbally empathise, is the opposite of what you describe.
Your requirment for something supernatural/quantum to decide between emulation and real thing is self referential.
If only we could tell either way, we'd have made a huge breakthrough in the philosophy of consciousness.