Philosophy Discussion Forums | A Humans-Only Club for Open-Minded Discussion & Debate

[The Beast]

Beetle in the box (thought experiment)
If the computer is a tool, the human attached to the computer is doing the computing. Humans compute. To describe how computing is done, we describe the system (box) and software (beetle). In the case of super-computing, Humans create futuristic scenarios like the emergence of autonomous intelligence from unprogrammed functionality… There are scenarios of few variables but, super-computing done today goes to “understand” major societal problems. So, humans do supercomputing (directing the computer) towards understanding and, in some cases program agency for the sake of efficiency. A below average scientist could in theory improve to the top of his field because his understanding is done with supercomputing.

[Pattern-chaser]

Here's an AI-related article I came across today. Here're the first couple of paragraphs:

The rigid structures of language we once clung to with certainty are cracking. Take gender, nationality or religion: these concepts no longer sit comfortably in the stiff linguistic boxes of the last century. Simultaneously, the rise of AI presses upon us the need to understand how words relate to meaning and reasoning.

A global group of philosophers, mathematicians and computer scientists have come up with a new understanding of logic that addresses these concerns, dubbed “inferentialism”.

You can find the whole article here. TheConversation is a site like Medium or Aeon, and I have found it to be trustworthy, and not to carry evil links into the 'dark web'. <shudder>

[Count Lucanor]

I am unable to post links here but these provide a taste of what I have been reading lately:

A Thousand Brains: A New Theory of Intelligence, Jeff Hawkins, March 2, 2021, Basic Books
“These Living Computers Are Made from Human Neurons”, Scientific American, 8 August, 2024
“How (and why) to think that the brain is literally a computer”, Front. Comput. Sci., 09 September 2022
“Neural tuning instantiates prior expectations in the human visual system”, Nature Communications, 1 Sept, 2023
“The computational power of the human brain”, Frontiers in Cellular Neuroscience, 7 August 2023

Maybe if you broadened your conception of "computation", which you seem to associate only with present day computers, you would not be so dogmatically impossiblist. What computers currently do is a very limited form of computation which I agree is never likely to achieve intelligence or consciousness. Neuroscientist and computer scientist, Jeff Hawkins, explains that what is needed is a better understanding of the brain and the processes which occur therein so that those processes can be emulated.

Thanks for the references. At least a couple of them we can discard right away, because they obviously subscribe to the computational theory of mind, so they can hardly represent a new research frontier.

About Jeff Hawkins's book, I've been going through some pages and I find it quite interesting, so I'll "eat" the entire book for sure and will propose to my book club. Perhaps because of a confirmation bias, since in some parts it looks as if he was repeating my stances on this thread. From Dawkins' foreword:

"Among the more important of the brain’s models are models of the body itself, coping, as they must, with how the body’s own movement changes our perspective on the world outside the prison wall of the skull. And this is relevant to the major preoccupation of the middle section of the book, the intelligence of machines."

"It is not that Hawkins underestimates the power of artificial intelligence and the robots of the future. On the contrary. But he thinks most present-day research is going about it the wrong way"

And from Hawkins himself:

"You might be surprised by my claim that the human brain remains a mystery. Every year, new brain-related discoveries are announced, new brain books are published [...] But if you ask neuroscientists, almost all of them would admit that we are still in the dark. We have learned a tremendous amount of knowledge and facts about the brain, but we have little understanding of how the whole thing works."

"In the forty years since Crick wrote his essay there have been many significant discoveries about the brain, several of which I will talk about later, but overall his observation is still true. How intelligence arises from cells in your head is still a profound mystery."

"In my interviews with MIT faculty, my proposal to create intelligent machines based on brain theory was rejected. I was told that the brain was just a messy computer and there was no point in studying it."

"The long-term goal of AI research is to create machines that exhibit human-like intelligence [...]The essential question today’s AI industry faces is: Are we currently on a path to creating truly intelligent AGI machines, or will we once again get stuck and enter another AI winter? The current wave of AI has attracted thousands of researchers and billions of dollars of investment. [...] When you are in the middle of a bubble, it is easy to get swept up in the enthusiasm and believe it will go on forever. History suggests we should be cautious.

I don’t know how long the current wave of AI will continue to grow. But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach."

"Nothing we call AI today is intelligent."

So, in general, from what I read so far, I welcome Hawkins' book, even though I disagree with some other things he says. But in the context of our current discussion, a few things must be pointed out:

First, this book does not represent current state of research in AI, nor a new research program that is making its way. It's a proposal to change the current path, which I solidly agree with, but it does not support the predictions that the current wave of AI research will eventually catch up and produce real intelligence.

Hawkins does not seem to support the computational theory of mind and he is firmly opposed to trying to get intelligence from modern computers (Universal Turing Machines). In that sense, Hawkins is no less an "impossibilist" and a "dogmatic" anticomputationalist than I supposedly am. He agrees with me in that AI currently dismisses the importance of the actual physics involved in intelligence. That's because they are only interested in the flow chart algorithms, in a hierarchical process, which does not actually exist in brains.

Secondly, it clearly states that it's proposing a theoretical framework, which is fine and should be welcomed, but that does not mean we have already started to understand the physics behind intelligence. Hawkins is quite open about this:

"I believe we discovered the framework that Crick wrote about, a framework that not only explains the basics of how the neocortex works but also gives rise to a new way to think about intelligence. We do not yet have a complete theory of the brain—far from it. Scientific fields typically start with a theoretical framework and only later do the details get worked out.

In the first part, I describe our theory of reference frames, which we call the Thousand Brains Theory. The theory is partly based on logical deduction."

[Lagayascienza]

I hope you enjoy Hawkins' book. It's a bit repetitive in parts but is an easy read for lay-people. At the end of the book he provides a list of references to papers which go into more technical detail than he provides in his book. Are there any references you can point me to dealing with work being done in neuroscience and AI?

[Sy Borg]

But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach.

Why would anyone assume that all AI researchers will follow exactly the same path without deviation for tens, hundreds or even thousands of years?

There will be many different approaches.

[Lagayascienza]

Right. Hawkins talks about getting machines to do what brains do rather than making exact replicas of organic brains. That would entail just building another organic brain which may be neither possible or necessary. Over millions of years mindless evolution lands on solutions that get various jobs done. But these are not necessarily the solutions that would work best - they just need to work well enough. Take the male urogenital tract in humans for example. From a design point of view it's a dogs breakfast that ends up causing lots of problems for men as they get older. If we were designing it from scratch we wouldn't route the urethra through the prostate. There are better ways to do it.

Perhaps mindful solutions to getting artificial brains to do what organic brains do will also able to take advantage of different ways of designing things. Just as airplanes didn't need wings that flap in order to fly, there will likely be ways to get artificial brains to work as well, if not better in some ways, than organic brains. For example, our brains rely on electro-chemical processes which are very slow compared with the way things happen in silicon. That's partly why today's AIs can crunch numbers much faster than our organic brains. Hawkins is right to say we will be able to understand the brain better and, once we understand it well enough, we will be able to emulate, rather than copy, what it does. It's not impossible. It's just going to take time.

[Pattern-chaser]

But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach.

Why would anyone assume that all AI researchers will follow exactly the same path without deviation for tens, hundreds or even thousands of years?

Because that's what they've done in the recent past? Because that's what all inventors and applied scientists have done for ... ages. In the case of AI, we haven't yet *had* "hundreds or even thousands of years" to work, but so far, we're following our usual track of human 'progress', I think?

[Lagayascienza]

I’d like to come back to a point I made earlier. Those who say that AGI is impossible seem always to resort to the fact that it has not yet been achieved, and that it will be difficult to achieve. But they never enlarge on that and explain why they think it is unachievable in principle.

I think that AGI IS possible, and I think it is possible because intelligence and consciousness emerge from physical processes in physical brains and because those processes, once adequately understood, can eventually be emulated. We are capable of understanding those processes and, when we do, the rest will follow. Eventually. At the moment AGI is science-fiction in the process of becoming science fact. However, it’s not going to become science-fact tomorrow, or even in a decade as some enthusiasts argue.

I agree that what is currently called AI is not true intelligence, and certainly not consciousness, but I disagree that these are impossible to achieve in an artificial substrate in which enough of the processes that occur in organic brains are emulated to the requisite degree.

Whilst my mind is not closed on the matter, I don’t see how I could agree with the impossibilists unless they can provide a reasoned account of why AGI is, in principle, impossible. Saying that it has never been done is no argument. Heavier-than-air flight and staying under water for an hour had once never been done and were once impossible for us. So we invented airplanes and SCUBA which made them possible, along with inventions that made many other things possible which once seemed impossible. There is no in-principle reason that I know of to think that we cannot eventually do the same with AGI. If anyone knows of such a reason, then I am open to it and would like to hear it.

Some things may well be impossible for us. Travelling at light speed may be one such. The laws of physics don't seem to allow it. There may be ways we could get around this limit via hypothetical worm-holes or whatever, but that really is just science-fiction and there is no current way forward. We don't have a clue about how to establish whether worm-holes exist or whether they are possible.

I don’t think AGI is in the same league as worm holes. We know how to proceed - there is a clear path forward for AGI. We just need to understand in more detail how biological brains physically do what they do, and then come up with the requisite technology. Such understanding is possible for us. We are smart. We'll figure it out.

[Count Lucanor]

I hope you enjoy Hawkins' book. It's a bit repetitive in parts but is an easy read for lay-people. At the end of the book he provides a list of references to papers which go into more technical detail than he provides in his book. Are there any references you can point me to dealing with work being done in neuroscience and AI?

As Hawkins and almost any other neuroscientist say, there's a mountain of research on natural intelligence, yet most if not all of the physical processes involved remain a mystery. All we can do is to continue, as Hawkins points out, focusing on the theoretical frameworks. The computational theory of mind was at some time that promising theoretical framework, the field even became the reigning paradigm, but we can now categorically put it to rest to seek for new approaches, even though this represents a big challenge, considering the pernicious influence of the tech industry, driving market forces to the research programs focused on computationalism.

So, we know which theoretical framework of human cognition doesn't work. And which one is the right theoretical framework? I cannot tell you which one is the right one, but I can tell you where I'm looking for it. I'm quite interested in the concept of embodied cognition, the path originally signaled by Merleau-Ponty that has gained some track in the work of ecological psychologists. There is an article on embodied cognition in the Stanford Encyclopedia page that sums up well what is all about and how it opposes computationalism. Interestingly, there are some hints of embodied cognition in some chapters of Hawkins's book focusing on the patterns produced by the organism in motion. This is right now for me the most interesting research program. Surely, there will not much hype around it, not much sci-fi literature and movies, nor enthusiastic tech lords promoting it while making big business, but that's how it is, at least for now.

Right. Hawkins talks about getting machines to do what brains do rather than making exact replicas of organic brains. That would entail just building another organic brain which may be neither possible or necessary. Over millions of years mindless evolution lands on solutions that get various jobs done. But these are not necessarily the solutions that would work best - they just need to work well enough. Take the male urogenital tract in humans for example. From a design point of view it's a dogs breakfast that ends up causing lots of problems for men as they get older. If we were designing it from scratch we wouldn't route the urethra through the prostate. There are better ways to do it.

Perhaps mindful solutions to getting artificial brains to do what organic brains do will also able to take advantage of different ways of designing things. Just as airplanes didn't need wings that flap in order to fly, there will likely be ways to get artificial brains to work as well, if not better in some ways, than organic brains. For example, our brains rely on electro-chemical processes which are very slow compared with the way things happen in silicon. That's partly why today's AIs can crunch numbers much faster than our organic brains. Hawkins is right to say we will be able to understand the brain better and, once we understand it well enough, we will be able to emulate, rather than copy, what it does. It's not impossible. It's just going to take time.

One thing I'll keep highlighting is the need to overthrow the "brain-as-the-center-of-the-self" paradigm, which is a legacy of computationalism and its implicit dualism. That means the actual task of AI research should not be replicating or emulating brains, or building devices that work as more efficient brains, but to build artificial organisms, whatever that actually means, as long as it implies learning from how nature does it. Is it possible? Theoretically, anything is, but being a practical realist, we don't know yet, and we cannot know until actual technical developments are achieved.

[Count Lucanor]

But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach.

Why would anyone assume that all AI researchers will follow exactly the same path without deviation for tens, hundreds or even thousands of years?

Because that's what they've done in the recent past? Because that's what all inventors and applied scientists have done for ... ages. In the case of AI, we haven't yet *had* "hundreds or even thousands of years" to work, but so far, we're following our usual track of human 'progress', I think?

For sure. The future is contingent and in the case of human technology, determined by actual social and historical forces, that is, by human actions, which we can hardly predict. It will not depend on a supposedly continuous march of progress, understood as an ethereal but natural force beyond history. We can say that, theoretically, AGI is achievable, but that doesn't give us a clue of whether it will be actually achieved or not.

[Sy Borg]

But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach.

Why would anyone assume that all AI researchers will follow exactly the same path without deviation for tens, hundreds or even thousands of years?

Because that's what they've done in the recent past? Because that's what all inventors and applied scientists have done for ... ages. In the case of AI, we haven't yet *had* "hundreds or even thousands of years" to work, but so far, we're following our usual track of human 'progress', I think? 🤔

Wrong. If what you said was true, then the only technology humans would have are increasingly complex and refined stone axes.

[Pattern-chaser]

But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach.

Why would anyone assume that all AI researchers will follow exactly the same path without deviation for tens, hundreds or even thousands of years?

Because that's what they've done in the recent past? Because that's what all inventors and applied scientists have done for ... ages. In the case of AI, we haven't yet *had* "hundreds or even thousands of years" to work, but so far, we're following our usual track of human 'progress', I think?

Wrong. If what you said was true, then the only technology humans would have are increasingly complex and refined stone axes.

I think we sometimes can find that rushing toward knee-jerk dismissal can cause us to miss the contributory aspects of what we discard.

In general, most discoveries and inventions come from doing what we've always done. That's why the majority of such breakthroughs are evolutionary, not revolutionary. I think it's also why we remember the ones that deviate from the 'norm'. Like carbon rings, double helices, or quantum entanglement...

In this case, it seems likely that the progress of AI research and development could perhaps do with something revolutionary; something that isn't what we've always done. We need a new and different insight, yes?

[Sy Borg]

But I do know that deep learning does not put us on the path to creating truly intelligent machines. We can’t get to artificial general intelligence by doing more of what we are currently doing. We have to take a different approach.

Why would anyone assume that all AI researchers will follow exactly the same path without deviation for tens, hundreds or even thousands of years?

Because that's what they've done in the recent past? Because that's what all inventors and applied scientists have done for ... ages. In the case of AI, we haven't yet *had* "hundreds or even thousands of years" to work, but so far, we're following our usual track of human 'progress', I think? 🤔

Wrong. If what you said was true, then the only technology humans would have are increasingly complex and refined stone axes.

I think we sometimes can find that rushing toward knee-jerk dismissal can cause us to miss the contributory aspects of what we discard.

In general, most discoveries and inventions come from doing what we've always done. That's why the majority of such breakthroughs are evolutionary, not revolutionary. I think it's also why we remember the ones that deviate from the 'norm'. Like carbon rings, double helices, or quantum entanglement...

In this case, it seems likely that the progress of AI research and development could perhaps do with something revolutionary; something that isn't what we've always done. We need a new and different insight, yes?

This isn't about "most discoveries", it's about discoveries that change the playing field. Any suggestion that there will be no more game-changing discoveries in AI - not in decades, even centuries - is clearly misguided.

[Lagayascienza]

AGI will become a reality eventually, not only because it is possible, but also because it will generate a lot of wealth for those who own it. That is why the already-wealthy are throwing billions at it - even if most of that money is barking up the wrong research tree by focusing on "computation" in the narrow sense of the term. That will just produce better AI and not AGI.

People like Hawkins will be proven right IMO - we will have to first better understand how brains do what they do and then emulate that in an artificial substrate, in a substrate with the ability to sense the world around it and make mental models of it. That is, it will need to be embodied in some way. But that, too, is possible and will come down to technology.

[Pattern-chaser]

In general, most discoveries and inventions come from doing what we've always done. That's why the majority of such breakthroughs are evolutionary, not revolutionary. I think it's also why we remember the ones that deviate from the 'norm'. Like carbon rings, double helices, or quantum entanglement...

In this case, it seems likely that the progress of AI research and development could perhaps do with something revolutionary; something that isn't what we've always done. We need a new and different insight, yes?

This isn't about "most discoveries", it's about discoveries that change the playing field. Any suggestion that there will be no more game-changing discoveries in AI - not in decades, even centuries - is clearly misguided.

Yes, it's about what I called "revolutionary" discoveries. And I'm pretty sure that no-one has suggested that there can or will be no such breakthroughs, but only that the field of AI could do with one or more such ground-breaking insights.