Will the first AGI be based on wetware? LLMs have so far fallen well below their promises, so badly they're currently trying to change the definitions of AGI and ASI to pretend they have achieved it. This is contrasted by biological substrates which have demonstrated the ability to produce general intelligence in millions of species throughout history and prehistory. Given our ability to produce nonbiological intelligence is purely theoretical, does it not seem that we should be devoting far more resources and efforts into what we already know can and does work?Or do biocomputers have weaknesses that make them simply not worth the investment yet, if at all?
I've wondered about wetware, I'm not a biologist but it seems to me like one of the great things about digital neural nets is they can be copied and have an unlimited number of copies of itself running in parallel since its software. You can't really do that with a biological brain your kind of just stuck with the one.
One of the upsides is that we can, absolutely, unquestionably, make superintelligences far smarter than us as long as we overcame some engineering issues. We already know brains can get bigger than a humans; some whales have brains thrice as big as ours. And we know it's not just brain size that brings intelligence, with neuron count in the cerebrum/pallium being just as important as total brain size.Going by this logic there seems no reason (outside of ethical ones) to presume we couldn't just grow a wetware structure several times as large as our brain, that is specifically designed so that all it's neurons end up in areas that are important for higher computtation rather than merely controlling and recieving sensation from the body, creating a true superintelligence that is to Einstein what he is to a chimpanzee. Of course, we'd need to figure out ways to grow these neurons in the correct structure, feed them, prevent infections, train it, and potentially look after it's mental health if we give it enough of a consciousness to need it. Not to mention we'd need a greater understanding of brain functioning than we currently posess.But that does not change the fact that it almost certainly COULD be done, no maybes, ifs, or buts, while we're not even sure it is possible to create a hardware-based superintelligence.
>>16953074I mean we could have A superintelligence, but it would just be a really smart singular entity I don't think that has nearly the same practical application as a digital neural net that can service an unlimited number of users or be embedded into mass produced products etc
>>16953077especially considering that we have 8 billion humans in the world and like half of them make 5 dollars a day, biological brains are apparently already cheap and abundant
>>16953077A superintelligence is a superintelligence, no matter how you slice it. It may well be literally infinitely smarter than us in some ways even if it's brain is only say five times as big and structured twice as efficiently. A human being only posesses ten times the cerebellum neuron count of a baboon, but we are capable of entire higher dimensions of thought that even an infinite amount of baboons working together couldn't match. No matter how many baboons you have a human will always outperform them at calculus, quantum physics, mechanical engineering, etc.Even a single superintelligent vat brain may well be similarly beyond us, and it's contributions, wherever they may be, might be utterly invaluable and umatchable. We could use it to fix issues with our current understanding of physics, use it for economic, political, or military advice, have it design utterly revolutionary new technologies, let resolve all sorts of yet unsolved scientific mysteries, have it come up with ideas we may literally have never even been able to concieve on our own, but may be able to with it's guidance, etc. The possibilities are pretty much endless.
>>16953077A neural net would be cheaper but so far the practical abilities of software to do extremely complex tasks that require great flexibility can be called into question. Biocomputers would probably be used for higher level tasks and operations that software just couldn't perform on it's own, and for which humans are inadequate or uneconomical. There's also the possibility of the wetware being outright wired into some nonbiological computers and using them to enhance it's capabilities, fusing the systems.
>>16953067They will keep making up new terms, pushing what they will never achieve down the road with the promise that we will most definitely get there in two more weeks. Like all Marxists.
>>16953067its easier just to use stock humans to do your computation.
>>16953067Itl be analogue
the growth of the neurons is engineered to realiably produce neural activity which, when measured and parsed into ones and zeros, creates a game of pong, but do we have any reason to believe that the neurons are thinking (and activity in human neurons must be some kind of thought) about pong at all?because as long as such a disconnect exists i don't see these computers as having any pretense at approaching human intelligence. for all we know, it's like manipulating the temperatures of different units in a server farm, measuring said temperatures, and trying to run a "computer" off of the measurements. this seems a crude system that involves neurons and does computation, but i do not consider it a neurological computer.
>>16957005It's incredibly simple right now. i think the current computers use only around 200,000 neurons (a fruit fly has 150,000). But given some scaling and time to master it a bit more and something may be there, even if it's literally just growing brains in vats and asking them to think about things.The computers we had in the 1950s could only do 1-2k operations a second. Fast forward to today and we have computers that can do a quintillion. This technology has only been in exeperimentation for a decade, so we'll need to see at least a few more before we can really judge it's feasibility
>>16957008but that big number isn't necessarily impressive unless they have the ability to monitor and interpret what's going on inside those however many neurons at a very deep level, basicly mind reading. otherwise, you're simply manipulating the temperatures of a very large server farm without having anything close to access to the full potential of the tools you're using.
>>16957008>>16957005Additionally, biocomputers don’t really need to be run off of neurons specifically. DNA computing, peptide computing, biochemical computing, and biomechanical computing are all being toyed with
>>16957010Actually they’re made some real advancements in the field of interpreting neuron patterns into information. Even a couple years ago they were able to make someone stare at an image for several minutes while hooked up to some sort of scanner, and computers were able to draw roughly the image he was seeing from just his brain patterns. Pretty simple right now but still proof of concept
>>16957012yeah that makes sense, because these biological components have predictable reactions which is all one really needs for a computer. my thinking is without full on mind reading, the neurons will not outperform these other options you mention, as we will not truly be accessing their thought but observing and manipualting their behavior in much the same way we would those other options. >>16957014that is interesting and i have to admit i have seen this: https://www.youtube.com/watch?v=FLb9EIiSyG8which is quite old. that said, surely things such as these require extensive foreknowledge of the shape of the relevant structures, yes? do you really think cortical labs is able to interface directly with their neurons rather that simply manipulate them superficially?
>>16957021> my thinking is without full on mind reading, the neurons will not outperform these other options you mention, as we will not truly be accessing their thought but observing and manipualting their behavior in much the same way we would those other options.I’d definitely agree with this. I think if development is successful future neuron-based biocomputers will less be computers we can type an answer into and read an output and more literal brain-in-vat AGI (which would have terrible ethical ramifications but also lots of potential benefits…)> that said, surely things such as these require extensive foreknowledge of the shape of the relevant structures, yes? do you really think cortical labs is able to interface directly with their neurons rather that simply manipulate them superficiallyOh yeah it’s definitely mostly superficial right now. They seem to have at least some success at least, enough that it’s commercially available to buy, and enough that they recently made it play doom (https://m.youtube.com/watch?v=yRV8fSw6HaE) but being able to reliably compute with it is a ways off if it’s happening at all.
>>16957024i've seen the doom video. it's amusing to me that it seems to shoot in panic. that's about how i would shoot if my brain were wired into the world of doom.but given the superficiality i think the decision to use human neurons was either pure chauvanism or pure marketing. i have no reason to believe that in this configuration human neurons would outperform dog, ape, crow, pig, whale, elephant, or ant neurons, just to name a give candidates.
>>16957039I think they’d be better off going with bird neurons personally. Consume three times less glucose than mammals, and are more efficient. Look at parrots being much smarter than mammals with equivalent sized brains
>>16957042actually i was just talking about the possibility of parrots being used in this sort of thing with a girl. she pointed out that growing parrots develop language much faster than humans and was very proud of her parrots ability to say "may i breakfast, please".
