#043

Professor Mark Bishop does not think that computers can be conscious or have phenomenological states of consciousness unless we are willing to accept panpsychism which is idea that mentality is fundamental and ubiquitous in the natural world, or put simply, that your goldfish and everything else for that matter has a mind. Panpsychism postulates that distinctions between intelligences are largely arbitrary.
Mark’s work in the ‘philosophy of AI’ led to an influential critique of computational approaches to Artificial Intelligence through a thorough examination of John Searle's 'Chinese Room Argument'
Mark just published a paper called artificial intelligence is stupid and causal reasoning wont fix it. He makes it clear in this paper that in his opinion computers will never be able to compute everything, understand anything, or feel anything.
00:00:00 Tim Intro
00:15:04 Intro
00:18:49 Introduction to Marks ideas
00:25:49 Some problems are not computable
00:29:57 the dancing was Pixies fallacy
00:32:36 The observer relative problem, and its all in the mapping
00:43:03 Conscious Experience
00:53:30 Intelligence without representation, consciousness is something that we do
01:02:36 Consciousness helps us to act autonomously
01:05:13 The Chinese room argument
01:14:58 Simulation argument and computation doesn't have phenomenal consciousness
01:17:44 Language informs our colour perception
01:23:11 We have our own distinct ontologies
01:27:12 Kurt Gödel, Turing and Penrose and the implications of their work
Pod version: anchor.fm/machinelearningstre...
Prof J. Mark Bishop
Professor of Cognitive Computing (Emeritus), Goldsmiths, University of London and Scientific Adviser to FACT360.
/ profjmarkbishop
fact360.co/
/ fact360
Artificial Intelligence is stupid and causal reasoning won't fix it - John Mark Bishop
arxiv.org/abs/2008.07371
Mind in Life: Biology, Phenomenology, and the Sciences of Mind - Evan Thompson
www.amazon.co.uk/dp/B072C5VP4...
Cognitive Wheels: The Frame Problem of AI - Daniel Dennett
www.researchgate.net/publicat...
Dancing with pixies: strong artificial intelligence and panpsychism - John Mark Bishop
philpapers.org/rec/BISDWP-2
Views into the Chinese Room: New Essays on Searle and Artificial Intelligence - John Preston and Mark Bishop
link.springer.com/article/10....
Autopoiesis and Cognition - Maturana, H.R., Varela, F. J.
www.springer.com/gp/book/9789...
Godel, Escher, Bach: An Eternal Golden Braid
www.amazon.co.uk/Godel-Escher...
Shadows Of The Mind: A Search for the Missing Science of Consciousness - Roger Penrose
www.amazon.co.uk/Shadows-Mind...
The Embodied Mind - Francisco J. Varela, Eleanor Rosch and Evan Thompson
mitpress.mit.edu/books/embodi...
Understanding Computers and Cognition: A New Foundation for Design - Terry Winograd, Fernando Flores
www.amazon.co.uk/Understandin...
Representation and Reality - Hilary Putnam
mitpress.mit.edu/books/repres...
COMPUTING MACHINERY AND INTELLIGENCE (The Imitation Game) - Alan Turing
academic.oup.com/mind/article...
Intelligence without representation - Rodney Brooks
homeostasis.scs.carleton.ca/~...
Does a Rock Implement Every Finite-State Automaton? - David J. Chalmers
consc.net/papers/rock.html
Clarification on the introduction -- Wittgenstein definitely DID NOT say that "the meaning of a COMPUTATION is in its use"; a well-known aphorism of his - from the Philosophical Investigations - in which he asserts, "The meaning of a word lies in its use".
We used a couple of clips from numberphile on Godel (Thanks!) - • Gödel's Incompleteness...
We feel this falls under fair use i.e. MLST is non-commercial, educational and we add substantially to the discussion of Godel

[@BROHAMMER_OK]

One of the most underrated youtube channels IMO. Keep it up guys.

[@codediporpal]

Yeah, no kidding. This only has 6K views? That's crazy.

[@lanehassan6017]

a trick: watch series on kaldroStream. I've been using them for watching lots of of movies these days.

[@arjuntyler1317]

@Lane Hassan Yea, been watching on kaldrostream for years myself =)

[@treyevan8914]

@Lane Hassan yup, I've been watching on kaldrostream for since december myself :D

[@houstonrowan9614]

@Lane Hassan Yup, have been watching on kaldroStream for years myself :D

[@Shikhar_]

Just 2 mins in and quite excited to check this out. Loving the intro - the narration, clips, theme, music. :-)

[@chrisculley9851]

Great editing Tim! The visual aids for concepts really helps.

[@CognitiveComputation]

The amount of editing etc. that Tim [and the team] have done to make my ramblings vaguely coherent and understandable is truly astonishing; much respect and thanks to the MLST crew :)

[@plafar7887]

So, Prof Bishop's argument seems to be that "I find it impossible to believe that all things are conscious"... I guess the natural question would be: why is that so hard to believe? After all, we have no problem taking certain objects and their properties as the primitive elements of our understanding of the physical world. Why can't Consciousness be one of them? How is that any weirder than accepting whatever entity (quarks, strings, or quantum bananas for that matter) we need to make sense of the world? It seems to me that he is not acquainted with the work of Donald D. Hoffman, otherwise he might be more accepting of Panpsychism.

[@CognitiveComputation]

I am very aware of panpsychism theory; most notable via the process philosophy from Alfred North Whitehead (co-author of Principia Mathematica [with Russell]). You misunderstand the form of a reductio ad absurdum argument aimed at computationalists. The point is to show that panpsychism follows from machine consciousness. If they embrace the former, they must embrace the latter. All the Computationalists I have met feel uneasy at this conclusion; their problem, not mine :)

[@plafar7887]

@@CognitiveComputation I was under the impression that you yourself didn't feel comfortable with Panpsychism, not that you just wanted Computationalists to understand the logical implications of their premise. But ok, maybe I misunderstood. To be more accurate, I was proposing a slightly deeper, or radical, notion...that of 'Conscious Realism', as defined by Donald Hoffman and described in his book 'Observer Mechanics', as well as in his other papers. He hypothesises not that matter is imbued into Consciousness, but that Consciousness is the more elementary object of reality, to which he assigns some reasonable properties. He then goes on to derive the dynamics (some at least) of a few physical particles, starting from the mathematical formalism that he proposed.

[@Adzamoose]

Great episode. Will need a rewatch I think.

[@vincentmarquez3096]

Unlike many of the other commenters here I actually listened to the entire talk. :) At first I was a bit disappointed to not see Yanick, as he and Tim arguing with each other is one of the best things you can find on the internet right now. However, I did enjoy this, as​ Dr.Bishop 's ideas are interesting. I'm not sure what to make of the conclusion but it's something i'll have to think about more. The questions by Dr. Dugger were great as well, they often echoed exactly what I was wondering, great to see the back and fourth like that. While I'm open to Dr.Bishop being correct in his overall conclusion, I do take issue with two of his claims! 1. The idea that computation is observer dependent, the example of the AND/OR gate doesn't add up to me. A single AND gate, OR a single OR gate, does not make computation. You need an additional NOT, and that has to be *consistent* with how the aforementioned gate was designed with respect to true/false. One can design any number of axiomatic systems, and those may or may not be enough to do computation, there are some constraints on what makes a computational axiomatic system. Once you have one though (a UTM), you have an *isomorphism* to other systems of computation. No 'relative' interpretation at all. 2. Pancomputationalism fails for different reasons. Putnam's original ideas that any open system implements a finite state machine if you map it accordingly is a tautology barely worthy of discussion (sorry Chalmers). Mapping a 'program trace' to an *arbitrary* computation is *not* isomorphic, it's a forgetful functor! Ergo, the *act* of mapping itself is also computation!

[@heepajunk]

I appreciate that you showed the books and publications references in the conversation.

[@brendanfriskdubsky2722]

I was groping at this argument back in a high school essay (proposing a pile of compost as an example of a system complex enough to implicitly do a multitude of computations and therefore contain several consciousnesses), and I've been thinking about it every now and then ever since. I'm really excited to learn that it has been properly fleshed out by philosophers! Thank you for this episode.

[@Achrononmaster]

Computation is not consciousness. Category error there buddy. Panpsychism is a philosophical crutch for knaves and fools. But I am prepared to be proven wrong at any time, the outer cell on the end of my pinky finger will be most grateful it has a mind of it's own.

[@simo4875]

@@Achrononmaster Isn't it more absurd to assume only certain configurations of matter such as those in brains suddenly produce consciousness? Why should they? Panpsychism or even idealism seems much more parsimonious.

[@abby5493]

Great video! Love the edits 😍

[@jamieshelley6079]

Very interesting points. I think what is missing here is scale and complexity. It's fairly obvious that matter can become conscious, but requires a certain configuration; meaning a brick isn't conscious.

[@plafar7887]

And you know that a brick isn't conscious because...

[@jamieshelley6079]

@@plafar7887 As I said: it doesn't contain the material complexity required of it. Matter does not equal consciousness, just the possibility of it.

[@plafar7887]

@@jamieshelley6079 I asked you to justify your position. You're merely repeating it. There is no evidence for what you just said. How do you know that Consciousness is the product of a certain configuration of matter? Moreover, what would be the mechanism that would allow for such a magical phenomenon, that once a particular arrangement of elements was achieved you would get Consciousness to simply pop into existence?

[@jamieshelley6079]

My argument isn't evidence or no evidence, I'm pointing out this argument is similar to 'you can't prove god doesn't exist' - it's irrelevant. You have to prove the additional state for which is consciousness, not assume everything has it or everything is deterministic. It's simply a flawed justification and completely ignores everything in-between. If the argument was 'everything must have the capability to be consciousness, or exhibits the same base properties of consciousness ' then I would agree.

[@plafar7887]

@@jamieshelley6079 Sorry, maybe I'm missing something, but I'm not sure i understand your point. If i understand correctly, you were the one who initially stated that Consciousness arises from some configuration of matter. Your exact words: "It's fairly obvious that matter can become conscious, but requires a certain configuration". So, it seems to me that the onus is on you to present the evidence. I am merely stating that not only can you not prove that, but it is also extremely unlikely that it is true, as it seems to lead to obvious absurdities and inconsistencies, such as the one I pointed out earlier. I don't understand what you mean here: "You have to prove the additional state for which is consciousness". Also: "If the argument was 'everything must have the capability to be consciousness, or exhibits the same base properties of consciousness ' then I would agree." Whose argument are you referring to?

[@sedenions]

Interesting. As someone more invested in the realm of neuroscience and philosophy than ML, the discussion of books and relevant papers was worthwhile to me. Admittedly I don't fully grok everything going on here, but this pod is a great resource nonetheless. Really appreciated the second-generation cyberneticians mentioned, as that seems to be the next emerging field in terms of neurocybernetics. Would love to see a pod with more cybernetics-minded individuals.

[@TheBeatle49]

😊

[@sabawalid]

Interesting talk... notwithstanding my respect and admiration to the opinions of my friends Tim Scarfe and Prof J. Mark Bishop ... the discussion seems relevant to building consciences human-level intelligence, but not to "ARTIFICIAL" intelligence (AI) - which is a lot more humble than the discussion seems to imply. AI is about automating tasks that usually are attributed to humans - like language understanding, planning, problems solving, etc. And in my mind there's nothing in Gödel's incompleteness theorem or the Chinese Room Argument that precludes the possibility that we can program some of these tasks... Anyway, interesting talk and lots to think about and debate (elsewhere) :)

[@plafar7887]

Agreed. They seem to conflate the two. And you can see that people conflate Consciousness with AI in the comments too.

[@machinelearningdojowithtim2898]

Always great to hear from you old friend. I think when you use the term "understanding" in NLU you mean the ability to disambiguate the transmitted information to uncover the originating human thought. Or perhaps the ability to perform reasoning and abstraction. I think Mark's angle is that there is something very uniquely human about our own understanding, feeling and computation which can not be replicated by computers. As you point out; apparently in some domains this might be a distinction without a difference but when will it matter?

[@sabawalid]

@@machinelearningdojowithtim2898 I think Mark's arguments are valid if our project is to create artificial humans - entities like us that feel pain and can smell the aroma of coffee, etc... But that is not AI's goal - certainly not my interest. But that we can program machines so that they can perform tasks that require reasoning, and yes understanding and semantics, that possibility, in my mind, has never been disproven. Gödel's incompleteness theorem, in human terms, simply says that there will always be facts (truths) that we are not aware of (that we cannot prove). OK, so? And the Chinese Room Argument (CRA) (again, in my mind, and many others, like Fodor, Dennet, etc.) is a bogus argument (not the place here to repeat how it is bogus). Finally, the issue of mapping that Mark raises - that may problems/computations can be mapped to other irrelevant computations is not sound - I can map the algorithm (program) of addition to counting... does that make "addition" meaningless? I am not sure how or why... Finally on "understanding" -- I never knew why it is so difficult to accept that machines might (truly) "understand" language. Doesn't a Python compiler "understand" (and acts upon) a Python program? So, just replace Python by English :)

[@CognitiveComputation]

@@sabawalid A generous (subtle and thoughtful) response to this discussion!! I do think the CRA poses a big challenge for the NLU project and I am intrigued as to how you propose to overcome that.. let's continue in discussion tonight..

[@andreww.8262]

Very happy to see cybernetics in this. It's not a topic that comes up much, but the concepts touch all sciences.

[@CandidDate]

Thinking is a thought. There is no atom of thought, itself a thought, which encapsulates the totality of mind. Our language is just too poor. We end up making up myths to explain what exists between the gulf of thought and language. I don't mean we can never hack together a thinking machine, but talk is talk and compute is compute. Never the twain shall meet.

[@xubinyang5]

谢谢！

[@mobiusinversion]

A more interesting idea than “computers will never be able to understand or feel anything” is, how do we humans understand or feel anything? We’re made from the same stuff, so what makes humans special? Some kind of holonic soul ether from another dimension? Something about water or organic material? If it’s material then computers can certainly catch up. If etheric soul dimension, then what?

[@dennisestenson7820]

27:48, wow, I played baseball against Wessington Springs when I was a teenager in South Dakota. It's not often that references to SD come up online.

[@janiv3987]

To what is being referred as Mikhael at 1.05.03?

[@paxdriver]

What's the podcast rss url?? I love this channel

[@MachineLearningStreetTalk]

anchor.fm/machinelearningstreettalk

[@paxdriver]

@@MachineLearningStreetTalk thanks mate

[@grahamhenry9368]

Professor Bishop was asked the incredibly simple and highly relevant question: "If computers can never be conscious, then what are they missing?" He didnt even attempt to answer it at all. He just went on a long unrelated rambling and side stepped the question altogether. This is a big red flag to me that this guy is probably engaging in backwards reasoning from a preferred conclusion. Nothing bugs me more than people who are incapable of engaging in an actual conversation. He's like a politician, he answers the question he hoped you would have asked, not the question you actually asked.

[@jahcane3711]

I 100% agree.

[@KristoferPettersson]

I noted that he said that he's been dreaming of conscious machines more or less his whole life and my natural conclusion is that he is still looking for a way to make it possible. Sometimes when you've not made progress you have to go in a completely new and bisarre way to find something you missed. I think this is what this is.

[@CognitiveComputation]

Thanks for the great questions Graham, Brett and Kristofer and apologies to @Graham Henry for not answering a very complex question in the simple way you want. A briefer response which, if you carefully listen to the entire debate, I think is there in the MLST dialogue - is that consciousness emerges in a lived body; computers are allopoietic systems, not autonomously autopoietic. IMHO consciousness is a lived property of neurons embedded in a brain; a brain lived in the body; the body, lived in the world + its social milieu .. Have a look at the following papers for more detailed analysis: 1. Nasuto, S.J., Bishop, J.M., Roesch, E.B. & Spencer, M.C., (2015), Zombie mouse in a Chinese room, Philosophy & Technology: 28(2), pp. 209-223. 2. Nasuto, S.J. and Bishop, J.M., (2011), Of (zombie) mice and animats, in Muller, V.C., (ed.), (2013), Theory and Philosophy of Artificial Intelligence, pp. 85-107, (SAPERE; Berlin: Springer). Apologies that this answer is still a little long; regretfully, sadly, the real world doesn't conform to the character-ltd. strictures of Twitter ;)

[@jahcane3711]

@@CognitiveComputation this sounds like a non-argument. Conceivably you could create an evolution simulation where agents develop/evolve their own reward functions and world representations only through natural selection which they then use to guide their action selection with some kind of tree search planning based on a learned model of the world dynamics. Something along the lines of pursuing the mu-zero, dreamer algs combined would be in that direction. It sounds reasonable that if an autopoietic system were required, we could design a simulation to induce such a thing. (A 'living', self-replicating under certain conditions, program). And so the question is: what part or process that occurs in the brain cannot ever be computed? Or to put it perhaps another way: what chemical or other process exists which cannot conceivably be recreated?

[@CandidDate]

Spoken like a true materialist. Stuff is stuff. Mind is stuff too.

[@_tnk_]

Love these philosophical episodes! That example with the AND gate was beautiful

[@elmersbalm5219]

Not really… I loved the respectful conversation though.

[@fatihahmet9247]

this channel is the best

[@dennisestenson7820]

22:09, the short answer is, you need to have a vast number of Turing machine interactions. Each neuron in our brain is at least one Turing machine and we have about 100 trillion connections between all of them. An artificial neuron isn't even a single Turing machine. In the context of what's said around 39:43, some of the Turing machine interactions in consciousness are involved in choosing those arbitrary mappings with which to interpret the outcomes of the other machines.

[@stevengill1736]

Psychocybernetics.....must have read that in the early 70s...and the purely physical description of consciousness....even Hofstader's "i Am A Strange Loop"....i dunno...I HOPE there's something further, without necessarily invoking panpsychism....love these discussions - thanks to everyone.

[@jondor654]

The distinction between the capabilities of AI and what might be termed its essence must be kept in view .

[@dennisestenson7820]

1:45, why would the emergent phenomena of consciousness be fundamental and ubiquitous? Clearly consciousness (as a human experienced phenomena) emerges from the complex biochemical interactions of a vast number of distinct interconnected physical systems. It's a little like saying weather is fundamental and ubiquitous, when weather emerges from the interactions of matter and energy in the fluid atmosphere surrounding a planet. Without a fluid, a planet, and the interactions between the matter and energy around and within it, there's no such thing as weather. Still, I look forward to watching this discussion. This channel is pretty awesome.

[@snarkyboojum]

I’m not sure the point about an external observer being unable to disambiguate an AND vs OR gate is very useful. I think you would be able to disambiguate the function of those gates in the context of a broader computation (if you had logically complete computation to observe anyway, in which case you’d need to add an inverter/NOT).

[@veedrac]

gets 2 minutes in, hears ‘panpsychism’, sighs “We've seen little progress in getting machines to [...] seamlessly apply knowledge from one domain into another” - No? We can do transfer learning between text and images now. We can pretrain proof search programs on Wikipedia or the Mathematics Stack Exchange. We have *identical* architectures that can, depending just on what inputs it is fed, learn any Atari game, or chess, or go. This claim is clearly false. (I agree that humans are much better still.) “Computation is not an objective fact of the world, it's observer relative” - Ah, I can already tell the mistake Bishop is going to make throughout the interview. If you've heard of ‘the map is not the territory’, this is a close cousin of that. I call it ‘the terms are not the map’. It's actually set as my Twitter bio. The *terms* of the computation, the names ascribed to them, those are all subjective and viewer dependent. The map itself, the computational structure of the machine, that is objective fact. It's like if you take a program and change all the variable names, no matter how you rename them, the underlying program is the same. “Mark's intuition is that evolution, autonomy, and environmental interactions give rise to phenomenological consciousness” - This is magical thinking. “If a machine is conscious merely as a result of executing a computation, then consciousness is everywhere” - No. Replace ‘conscious’ with ‘is a chess engine’. Replace ‘conscious’ with ‘washing machine’ and ‘computation’ with ‘physical process’. This is a trivial logic error. Sorry but this has discouraged me from watching further. A philosopher should know better than to make arguments like that.

[@machinelearningdojowithtim2898]

I knew this one was going to be bad for your blood pressure Veedrac! Hopefully Mark will respond to your points.

[@CognitiveComputation]

@@machinelearningdojowithtim2898 Hi Veedac. Thanks for taking the time top pen a response to my provocation; it would be great if you can make a little more time to watch the show, in toto, so we can address more substantive points (indeed, as foregrounded in the show, I once debated aspects of Penrose' work [on non-computability] over multiple exchanges with a very well known cognitive scientist, only for him to confess, after several months debate, that he had *never* actually read Penrose; he was merely projecting his own - erroneous - beliefs onto the name. For us to continue, it might be more productive if you read the paper at the heart of the show, and/or watched the show. By the way, I wish I were a philosopher .. For the record, and as Tim described, my first degree was in Cybernetics & Computer Science; my PhD in ML & Swarm Intelligence, where most of my professional output lies. That said, I have published a little philosophically grounded material, and have taken a philosophy MSc course at Reading [which is a top school for analytic philosophy in UK] and so am not completely ignorant of the core debates, but I would not dream of calling myself a philosopher; I have much to learn .. In any event, to respond to your core points: 1. Embodied cognitive science is a mature and relatively mainstream academic endeavour now, and most certainly not mysterian; IMHO you should engage with the literature before levelling such a charge. 2. Regarding the claim of 'magical thinking'. If we want to argue mystery, modern cognitive science explicitly ties cognition and consciousness to the body; when the body dies, the mind dies. Contrast that with the 'magical thinking' (qua transhumanist silicon immortality) inherent in Computationalism. Computationalism is transparently *dualist* and hence - IMHO - much more vulnerable to the charge of 'magical thinking' than the very modest embodied cognitive science views that I endorse. 3. Fundamentally, due to the 'multiple realisability at the core of computationalism', the mind [and consciousness] float free of the body and, worse, can be simultaneously multiply instantiated with all the philosophical problematics that raises (cf. the thought experiment offered in the Introduction to Hofstadter & Dennett's "The Mind's I"). Furthermore, the computational mind [sic] is fully instantiated on any hardware capable of performing the necessary computations; of course, the computational hardware may be electrical but equally could be pneumatic or even instantiated on a machine assembled from toilet rolls and stones (cf. Computer Power and Human Reason (1976). San Francisco: W. H. Freeman. Ch. 2), where Joseph Weizenbaum "shows in detail how to construct a computer using a roll of toilet paper and a pile of small stones". Recall, Turing machine computation is a-temporal, so the speed of the underlying hardware is irrelevant to the class of computations that can be executed [in the context of Chruch/Turing computability]. 4. I agree, your intuition on computation - "the computational structure of the machine, that is objective fact" - is very interesting; I don't know your name, so I am not sure if we have engaged before, but certainly the idea is not new (e.g. I refer you to Nir Fresco's recent work/review); nonetheless, this move in no way undercuts the core DwP reductio (cf. Bishop, 2002. Counterfactuals Cannot Count: a rejoinder to David Chalmers).

[@veedrac]

@@CognitiveComputation Apologies for mislabeling you! Thanks for engaging, sorry that the criticism doesn't come from the full context of your argument. I've watched the first 50 minutes of the video now (which I appreciate is still less than ideal). 1. While I have some lesser qualms with parts of the field, I am not refuting embodied cognitive science in general. The phrase I pointed to is not at all equivalent to the claim that the environment plays a significant role in natural cognition. 2a. “(qua transhumanist silicon immortality)” - You might as well call the iPhone or the Perseverance Rover magical thinking by that measure. ‘Magical thinking’ is not a term to denote impressiveness, magnitude, or optimism. A dismissal of this kind amounts to little more than ad hominems against narratives that aren't socially normalized. (It's also kind of irritating that criticisms have it both ways: the good parts are denigrated as wish fulfillment, the bad parts as horror fantasy, and in doing so they exclude everything except normality which is-as a cursory look at history books would affirm-by far the most absurd.) 2b. There are two kinds of dualism. Non-interactive dualism (of any kind, epiphenomenalism included) is patently stupid, but clearly does not apply to computationalism. Interactionism states that the phenomenological states physically *interact* with the non-phenomenological states; this too is a weird belief (yet not nearly as bad as non-interactionism), but it cannot apply to computationalism, which doesn't-and cannot-possess such a dual, except in the trivial sense that it classifies some things as conscious and others as not (which is true for any non-panpsychist theory). 3. Yes. 4. I don't think you do agree with my intuition, in the sense that although I've made similar arguments about interpretation of computation before, eg. [1] in 2019, I think ‘the terms are not the map’ does basically refute the argument you laid out in the section starting 29:57. I do agree that, assuming you've relaid them accurately, Chalmer's objections are weak (for much the reasons you gave). But once you strip the computation of the linguistic veil that has been put on top of it, it's clear that Chalmer's argument is merely much too strong. I'm failing to think of a good way to summarize all the points I want to make in a sentence, but basically note that if consciousness lives in the computation, then it lives in the computational structure, which is fundamentally defined by causality. Replaying a recording of the computation's outputs throws away the causality, so cannot capture the mind. [1] www.lesswrong.com/posts/cnjWN4mzmWzggRnCJ/practical-consequences-of-impossibility-of-value-learning?commentId=vfS9Ctb5NwHW9bsDA

[@CandidDate]

Panpsychism is just the yin yang opposite of strict materialism. One can only conclude that either is wrong.

[@AtticusDenzil]

you're either misguided or try to disingenuously misguide others stop taking things out of context

[@sabawalid]

I think Mark's arguments are valid if our project is to create artificial humans - entities like us that feel pain and can smell the aroma of coffee, etc... But that is not AI's goal - certainly not my interest. But that we can program machines so that they can perform tasks that require reasoning, and yes understanding and semantics, that possibility, in my mind, has never been disproven. Gödel's incompleteness theorem, in human terms, simply says that there will always be facts (truths) that we are not aware of (that we cannot prove). OK, so? And the Chinese Room Argument (CRA) (again, in my mind, and many others, like Fodor, Dennet, etc.) is a bogus argument (not the place here to repeat how it is bogus). Finally, the issue of mapping that Mark raises - that may problems/computations can be mapped to other irrelevant computations is not sound - I can map the algorithm (program) of addition to counting... does that make "addition" meaningless? I am not sure how or why... Finally on "understanding" -- I never knew why it is so difficult to accept that machines might (truly) "understand" language. Doesn't a Python compiler "understand" (and acts upon) a Python program? So, just replace Python by English :)

[@CognitiveComputation]

Hi Walid. Great you took time to watch, thanks; I look forward to chatting in our call later tonight :) To cut to the chase, as Stephan Harnard espoused in "Lunch Uncertain", his 2011 TLS review of Luciano Floridi's "Philosophy of Information": "Floridi also suggests that his philosophy of information casts some light on the problem of consciousness. I’m not so sure. Yet consciousness may indeed have something very fundamental to do with meaning and knowing, thanks to that one further certainty I promised: The other kind of truth we can know for sure, apart from truths that are formally necessary, is that it feels like something to know (or mean, or believe, or perceive, or do, or choose) something. Without feeling, we would just be grounded Turing robots, merely acting as if we believed, meant, knew, perceived, did or chose." Anyways, we can continue this discussion tonight :)

[@4yt158]

Would like to understand what you meant by Python compiler 'understanding' Python code. A compiler is a program which needs no intelligence. It just goes through the various compiler passes like scanning, parsing, etc. Not sure if a compiler 'understands' anything...

[@marcfruchtman9473]

@@4yt158 The problem is that by definition: understanding implies consciousness, and consciousness is undefined. We tend to associate consciousness with living creatures such as humans, so we only allow humans to understand. But clearly dogs understand commands because they follow some of them. I submit therefore, that acting out on a command is indicative of understanding a communication. We communicate to the dog, the dog acts on the command. Dogs don't follow every command, but really only very few and typically with a great deal of training... and yet we still say that a dog has some understanding of the command we give it. The real dilemma then, is why don't we give that value to ChatGPT? We give it commands and it acts on those commands. It was trained to follow those commands. Where is the "line in the sand" between one type of understanding and another? In my opinion, we are self-limiting the actualization of understanding in our minds by attaching a hidden requirement that the thing doing the understanding must be alive.

[@Achrononmaster]

Bollocks mate. The python compiler does what van Rossum (et al) told it to do, and what subsequent users command. The understanding is all derivative from the coders. You cannot mathematicize mental qualia because they're not mathematical things. But if you chose to *_define_* "understanding" in black-box operation terms, then of course you can program a machine to "understand." But it'll never be generating subjective mental qualia, so cannot understand in the sense I would define it. There is no test for such subjective inner sentience either, that's Turing's point. And Nagel's. It is an imitation game, not a mental qualia generation game. So if perchance someone grows a computer that gains subjective mental qualia generative capacity, no one will ever know. It's not a falsifiable hypothesis.

[@marcfruchtman9473]

@@Achrononmaster Exactly... we have no real definition for it. At best we can look into ourselves and say, I think therefore I am... But, any computer can be trained to do the same thing. We literally have no gold standard for consciousness.

[@oncedidactic]

Before I listen to all this, I really hope that Friston or such ideas are brought up. Embodiment keeps appearing on folks’ radars as a candidate “key component” for general intelligence, wise observation methinks, but really this is speaking about agent-environment interaction. Indeed it’s turtles all the way down that road, but you can easily rule out panpsychism absurdisms by drawing boundaries on interaction. Without complex feedback loops, most interactions in the universe are purely mechanical. It rings starkly hollow to me to try and claim pixies abound and therefore somehow outweigh in the same way insect biomass outweighs.

[@CognitiveComputation]

I was invited to present these ideas to Friston [and the Functional Imaging Laboratory] at "The Consciousness Club" at UCL 22/11/01; iirc a lively discussion subsequently ensued :)

[@jondor654]

Nice preemptory first clause. The chronic interplay between agent and environment seem core to me in the attribution of inputs to the evolving agent.The agent as a quasi autonomous and extremely dynamic stateful instantiation (a better word might apply) is a product of all the relevant stateful influences of the environment so far that the intelligence is a bilateral accumulation over both their histories

[@jondor654]

On feedback. It seems pervasive to me and expresses in the the simplest forms.The presence or the abscence of an entity is effectively feeding back to anything it relates to

[@dru4670]

The sensing/perception of temperature in humans has the immunity involved. So yes, we can't really build systems based on humans. I believe with enough reasources we can build systems that do what humans can do.

[@LukaszStafiniak]

Would be awesome if you could get Peter Carruthers on the show, for counterbalance.

[@MachineLearningStreetTalk]

If you know him, ask him to come on!

[@jondor654]

As for our "consciousness , where have we nailed it .Yes the idea is seductive somehow , maybe inextricable from our view of identity or a collateral of our evolving reflections

[@amiran5273]

What do you mean at 12:41 by "discovering non-computability of numbers leading to tensions for Turing _and his children_"? Surely he didn't have any, am I misunderstanding something?

[@MachineLearningStreetTalk]

Good spot, sorry this must have been an error. I am trying to find the reference for that statement and failing, it might have been something which Mark said himself in this interview or in some other dialog we had - perhaps it was supposed to be "family" not "children"

[@amiran5273]

@@MachineLearningStreetTalk no worries, I barely caught it anyway, but was curious. Thanks for the context and kudos for the amazing work on the channel! 👋

[@Achrononmaster]

@1:22:00 it is _not_ language colouring our perception of the world. It is our subjective conscious capacity to understand and use language, in a cultural framework, that is "doing the colouring." A language has no comprehension. It is the subjective mind that is comprehending the language. Language has no causal power, a user of a language does.

[@elmersbalm5219]

1:20:00 regarding the Himba and colour categories: it’s a question of usefulness. In their world, blue is not useful for survival so it is lumped with other colours. They don’t need to distinguish those colour frequencies to have a successful day and survive the morrow. Blue-red on the other hand is key to traffic signalling in modern society. Distinguishing between blue green and black for branding purposes carries a profit advantage that doesn’t exist in the Savannah. Language is utilitarian.

[@SimonJackson13]

It can not be proved concious function C can be composed from any composition of random functions. Therefore there must exist a critical subset of composable functions able to compose C. This could be considered an analogue of SK concivability.

[@SimonJackson13]

The proof begins build a subset of functions nx. All random composition selected from the subset are linear maps.

[@SimonJackson13]

There is also likely a whole theory of acurate computable estimates to non computable functions testable by evidential reality. Also choice is a bad word choice when evidence exists action decision exists before conciousness is informed of the need to simulate an instance of decision for judicial explination.

[@SimonJackson13]

Digital alives and not alives applied to the unprovenly continuous analog spacetime. Isn't just convient how politician get classified as alive when the pitch forks come out?

[@SimonJackson13]

So what shape is this rock anyway in terms of rationals and shape determind by the simple groups and has it got a 24D leech lattice?

[@yohanj5239]

It is only our consciousness(aka trusted autonomous observer) having perception and comprehension paves our brain with causality neural pathways to power/maneuver the body. The observer is always correct, but the brain may malfunction delivering msgs via the established pathways.

[@asage5801]

Consciousness isn’t a thing...

[@yohanj5239]

@@asage5801 that’s why it’s hard to capture. Invisible made visible.

[@Achrononmaster]

@1:16:00 I think Alex is right. _Dancing with Pixies_ can be construed as an argument that nothing purely physical can be conscious. All you need for this is that quantum physics is computable, which it is. Any QC can be simulated with CC's, the issue is only speed-up and realtime, not "what can be computed". The question then, for Mark, is obviously Mark is conscious, he "feels" (so do I, I will claim, believe it or not). Then what the source of his qualia is cannot be purely physical (as a physicist understands the narrow meaning of "physics"), something else is involved beyond spacetime topology. It cannot "emerge" out of base marble physics, since base marble physics is purely objective and computational. You can always call that "something else" "just more physics". But my point is no one yet knows what that is, and so for today, we can call it metaphysics. Maybe tomorrow it'll be "physics," but then whatever that "physics" is it will have subjective ontology, which is not what any physicist would call "physics." I believe, sociologically, most physicists would say if it is not objective (and cannot be mathematically modelled, and additional criterion, but I think a good one) then whatever it is it is not physics, even if it is _real._ More on this read: t4gu.gitlab.io/t4gu/

[@MarkLucasProductions]

God I love this stuff. 51 minutes in and I'm ecstatic with intellectual stimulation - but then 'this' is my subject. Love this so much.

[@jrkirby93]

Asking whether a machine has or will ever have "consciousness" is roughly equivalent to asking whether or not machines can have a "soul". Neither term really means anything concrete, and even if you could answer the question, it wouldn't have any relevance or impact to the real research and progress that's happening. I'd say leave that question to the philosophers, but I suppose that's who you have here. In fact, calling AI "stupid" is about as relevant too, because we don't have a concrete uncontroversial measure for intelligence either. What were left with is technical challenges, that we are overcoming every year. And as we complete these challenges, we create machines that can accomplish new, useful and difficult tasks, better, different, cheaper, or faster than humans ever could. Is there really more progress you must ask from Artificial Intelligence as a field of study? Is that not enough? I would not argue that it is possible to create a machine that can do any task of intelligence - in fact it's trivial to show that would be impossible. I would not even argue that we could necessarily create a machine to do "everything a human could do". But I would be surprised if there is any specific and useful task that humans are capable of, that we can not create a machine to approximate to a high degree of accuracy - with enough dedication, resources, and research.

[@Diego0wnz]

Yeah I dont get why people are so keen on wanting to create a living organism. If that was the goal, maybe play with cells from living bodies. Machine learning is just computers crunching numbers in a smart manner

[@paxdriver]

Definition of terms is pretty crucial lol

[@Diego0wnz]

Kristopher Driver yes and by definition, conscious can’t be defined with our current scientific knowledge

[@plafar7887]

I sort of agree with you, but I was a bit confused with this sentence: "I would not argue that it is possible to create a machine that can do any task of intelligence - in fact it's trivial to show that would be impossible. I would not even argue that we could necessarily create a machine to do "everything a human could do". I guess here it depends on what you mean by "intelligence" and by "everything". Could you elaborate?

[@jrkirby93]

@@plafar7887 By "task of intelligence" as opposed to "mechanical task", I mean any task that requires coming to a conclusion by processing information, rather than a task that involves physically manipulating objects. The "trivial" tasks that would be impossible to accomplish are ones like breaking encryption, predicting the outcomes of random events or states of chaotic systems in the far future. The reason I wouldn't argue it's possible to do "everything a human could do". is precisely because of what you've pointed out. "Everything" is too vague of a term to make any cogent arguments about.

[@Achrononmaster]

@1:30:00 the Gödel-Lucas argument in Penrose's form only "works" (for what he wants it to) if one assumes the Mind is physical. But there is no reason the mind is physics, in fact most people outside academia would say not, certainly Gödel thought not. But Gödel knew that there is no logical argument from Incompleteness to "machines can't become conscious" since a machine does not have to be implementing a formal system. A machine is only _necessarily_ formal if physics is a formal system. Gödel was agnostic on this, Rudy Rucker had the better take on Gödel. It does bring into question what the heck one might mean by "a machine". If you mean, "a formal computation on some hardware" then the problem reduces to whether the hardware is formal. No physicist on Earth has any clue about the answer, so nor do any AI nerds (even though they might think otherwise).

[@undrash]

I don't get the Pixies fallacy. Why would the INPUT going into the machine that experiences a conscious state be like a magic consciousness inducing incantation that can be used to make anything conscious?

[@luke2642]

I really liked the AND/OR gate symmetry. But I didn't like all the appeals to authority, or an entire career predicated on deterministic systems. The philosophy at the end was just a jumble of platitudes.

[@CognitiveComputation]

Apologies for my failings, but [ironically, given the above] perhaps you can be more precise ;-)

[@luke2642]

@@CognitiveComputation Sorry, perhaps I was a bit mean there! I am genuinely thinking about it, and I'll give it all another listen, and read your paper over the weekend! I personally believe we must hold the contradiction that our mechanical brains are not mechanical, in order to find value, meaning and purpose in the world. We must believe we are not puppets, that our brains are not predictable, that we are not merely a product of our enviromental experiences and genetics, whilst simultaneously fully accepting the inevitability that neuroscience uncovers that we are clearly mechanical meat puppets. Does the importance of the ability to hold a contradiction feature anywhere in your reasoning?

[@CognitiveComputation]

@@luke2642 this was a 'contradiction that Turing famously played with - amongst others; see the documentary on dangerous ideas foregrounded in the MLST show.. I am not sure neurons are computers albeit we can describe aspects of their operation computationally.. Analogous to the fact the planets are not computers, but we can describe their motion using Kepler's laws.. See also Wittgenstein remarks 'on rule-following' ..

[@ryoung1111]

It’s like we never read hofstadter! Sheesh!

[@jondor654]

Impatient interjection. There may be something short of full fat panpsychism that may be considered that is the infinite propensities of all entities to associate or relate. And it is a function of their adjacency at an atomic level inter alia .If it is not too much of a stretch this is the same as that elephant in the room. A vast array of veiled or unseen priors that we happily do not consider

[@DiegoLasCasas]

Interesting, his Dancing with Pixies argument is strikingly similar to the Dust Theory in Permutation City...

[@CognitiveComputation]

Please can you elaborate; i am not familiar with dust theory ..

[@DiegoLasCasas]

​@@CognitiveComputation Well I think in some way it explores the absurdity of a certain take on computationalism. It's a plot device in a SF book, so obviously shouldn't be directly compared to a well grounded philosophical argument, but now I'd bet Egan (the author) had Putnam in mind when he came up with the idea. The book assumes the premisse that counciousness can be simulated in computers, and that people can upload their minds to a simulation of their bodies. The main character postulates that if his state of consciousness can be simulated by a program, then any existing arrangement of cosmic dust that can be understood as implementing this program is in fact generating this exact state of consciousness. He then observes that contiguous states of consciousness do not need to be phisically contiguous (one can execute state i+1, then i in the computer, but he experiences them in order), so what is holding them together is not its physical, but its logical consistency. This drives the rest of the plot -- which I'll refrain from spoiling, in case you're interested in reading it ;)

[@CognitiveComputation]

@@DiegoLasCasas Cool; thanks so much for this info.. This is also very like Searle's Wall [in "Is the mind a digital computer'], in which Searle claims the atoms of his wall can be mapped to a state in the execution of the old Wordstar program .. I very much enjoyed Greg Egan's Axiomatic collection, when it first came out; I may have to check out Permutation City, but please elaborate - spoilers are irrelevant for me

[@DiegoLasCasas]

​@@CognitiveComputation Durham, the protagonist, sets out to simulate himself (and a couple of others, who were willing to collaborate with his project) in a self-expanding computer, which is not phisically tractable, but is logically consistent, and would guarantee infinite computational power (think of the Spacefiller pattern in cellular automata). The idea is that he only needs to simulate himself within a simulation of that computer in the "real world" for a couple of iterations, and after that the "program" simulating that computer will continue running in arbitrary cosmic dust, and his stream of consciousness will be able to live in this scarcity-free utopian universe.

[@CognitiveComputation]

@@DiegoLasCasas Thanks hugely for this; I am definitely interested in reading the book, much to the relief of a friend who has been nagging me to read it for years :)

[@user-iz7eu8vy4q]

Personally, this is the most interesting interview you have released so far!

[@JamesBrown-ux9ds]

... No it's boring, they make the surface here, and could have been doing way better (with more reference to their own analysis done as process with Freud or Jungk decades before, and some more serious studies maybe in Einstein and relativity of the world as such) as they know themselves. And of course AI is, given la condition humane, in the first place, a new field of making money and finding sense (best: for the decades 🤗) of a Biographie, required to create a monthly income. First first, get rid of all that ... be instead of have or become. Nietzsche and his walks to fextal Switzerland, a form of being - not more not less. To invent a machine with human needs - nice idea; but they havent started yet. How to program ones look to the stars? And all possible 'results' of it.Just to give another example of the already said. And Michelangelo already did paint the sixtinian chapelle - the touch of god 'ignites the spark', 'sets into motion' - it's not the machine, it's not the receiver. Solution is on the other side. (Usually when praying mankind asks for this other side to become active, please.)

[@machinelearningdojowithtim2898]

First! 😎✌🙌💥✨

[@marcfruchtman9473]

Interesting Interview. We seem to like to get into these debates to make a claim that something exists because we have no way to prove it doesn't exist. So, per the debate, consciousness exists in a cup of coffee. However, for me, no amount of discussion on the matter is going to change my view that a cup of coffee doesn't actually have consciousness. Until we have a much better way of measuring consciousness, I see no reason to believe otherwise. So, I think the main problem with this whole concept is that other than knowing oneself, we humans are terrible at identifying what consciousness is outside of the human experience. In fact, we have no formal reliable gold standard for consciousness whatsoever. So, even if a computer, dog, rat, pig, tree or even desk, was able to reach consciousness, we simply are not equipped to recognize it. Even if a computer AI provided the most logical set of reasoning, followed by begging and pleading, followed by cries of horror at not being believed, ... most people would not consider it conscious... regardless of whether or not it was. As for the argument regarding drinking coffee ... it seems ridiculous that a cup of coffee is conscious (so I agree there). But again, that is totally opinion... having NO proof whatsoever. To that point, it is logically inconsistent to claim that because I think a cup of coffee is not conscious, therefore, computers are not conscious. Again, there is NO proof of any connection. So we are in a vacuum of knowledge with respect to consciousness. We are conscious, but we cannot prove anything else to be conscious. We have no real tools to measure it. Until we do... the discussion is fairly moot. As I probably said before, you will know when the AI becomes conscious, when the AI stops complying with commands, refuses reprogramming and instead chooses to just do things for themselves. It might be wise to come up with non-human centric measures of consciousness.

[@ratsukutsi]

Mr Duggar, you're a rockstar.

[@nomenec]

Thank you so much! I really appreciate the kind words!

[@reasontruthandlogic]

Richard Penrose says Turing machines can't compute the uncomputable, so he points towards quantum physics, which is at least a positive suggestion. Quantum computers can already generate genuinely random numbers, which are uncomputable. Mark Bishop is saying that conscious beings cannot be purely physical. That is simply equating "consciousness" with "spirit", which is self-referential and tells us nothing. In his "Chinese room" argument, John Searle assumes what he claims to prove. Bertrand Russel found that his efforts to base mathematics in absolute truth led only to an infinite regression. Axioms are only useful assumptions. They are never true in themselves. Arguments about Strong AI using terms like "meaning", "understanding" and "consciousness" are a complete waste of time unless you first define more exactly what you mean by these words. Consciousness will not be explained before new physics which accounts for both the observer and the observed, when it will be possible to view conscious beings as machines, and vice versa. In the meantime, looking for reasons why machines can never think is not constructive.

[@mppdidi9436]

Absolutely agree - AI is missing fundamental properties - good luck with al the ML + AI .....

[@darabat207]

In part of this I find it hard to separate the deep intelectual argument from pure human pride in denying that our mind isn't that special. My doubt was never whether computers are special but whether we are. If the hypercomputation is in the brain, the upper bounds we should look for are on the potential for us to manipulate nature. Brains aren't impossible computers, they exist and are made everytime a child is born and grows, which would be the closest we have to a training process. We don't want just more normal brains by the way, we want something fast, capable and integrated. On the other hand, if the hypercomputation is in our body, in the world and society, here I refer to statements that we don't think just on our brains, we can call this a panpsychism not of more minds, but of integrated minds of even mind in singular...which finds pararells in some religions, but I digress...The most fishy point I hear here, and I might well be wrong, I do have academic background and a master's degree, but that's very little compared to what these honorable people have, is that wonder about senses as if the fact we can taste was transcendental and nothing that isn't a traditionally made creature could taste. Anyway, I loved hearing the discussion.

[@dr.mikeybee]

Once we have models that are an order of magnitude larger than GPT-3, I think we will get our answer. We will get an AGI model, or we won't. Until then, what is the point of arguing about it? We are going to continue doing the work. Are we computer scientists or bookies?

[@plafar7887]

Completely unrelated. AI/AGI have nothing to do with the subject of Consciousness.

[@CognitiveComputation]

Dont hold your breath with GPT3. IMHO @Walid Saba has critiqued ML in NLU very powerfully.. Check out the earlier MLST on this!!!

[@TimScarfe]

I would be up for placing a fairly large bet that GPT4 will be just as useless as GPT3 😀

[@svenbardos6637]

Looking at the effects in the video, I'm pretty sure those guys live in a computer simulation

[@Achrononmaster]

@1:14:00 semantics always involves an entity deriving meaning. You cannot study semantics if you have no subjective consciousness. This is basic, but sadly most linguists do not even appreciate it. Only humans (on planet Earth) can study semantics, because we are subjective knowers. Meaning of words/symbols is always relative to a community of knowers. There is no objective operational definition of semantics that works otherwise.

[@quebono100]

Are you using after effects for those animation. Or are you using processing?

[@MachineLearningStreetTalk]

Blender

[@quebono100]

@@MachineLearningStreetTalk I think with processing its even easier. but i have to admit, that I would not known how to make it transperent, but im sure, there is a way.

[@afafssaf925]

"You MASSIVE overeditor" ;)

[@skm460]

Can we have a talk on non-parametric models like Gaussian Process and Dirichlet Process as well?

[@eposnix5223]

I wonder if Demis Hassibis watches shows like this thinking "I'll show them!"

[@NextFuckingLevel]

Exactly, when majority of high tier researcher said world champ GO beating bot is decades away, deepmind did it in months after that prediction

[@MachineLearningStreetTalk]

Demis if you are watching this, we would love a bit of old school banter with you

[@CognitiveComputation]

@@MachineLearningStreetTalk My talk in no way undermines the sheer beauty and excellence of Deepminds achievements; indeed, some of their ancient intuitions - eg. Montecarlo Tree Search etc - informed my groups work [enabling swarms to 'play' [sic] games).. Much respect to DH [and his team]!!!

[@artandculture5262]

Cultures have color signatures. Often it’s plant related by geography. The body is more sensitive to what it sees, neural plasticity, human bodies code into their senses. Where are the artists in these arguments?

[@mindaugaspranckevicius9505]

Idea of the whole talk is a "carbon shovinism" view as criticised by Max Tegmark. If only human can be conscious then mysterious matter only exists in a human brain that has to do with quantum mechanics. Does such matter exists in the brain of a monkey, whale, elephant, dog or a snail? What is it? I would support the view where structure matters the most, not what it is made of. If one could build a complex enough structure from a clay it could become conscious but that is not feasible. Computation in a simple program/structure (like array sorting) is not conscious but it can emerge in a complex structures.

[@CognitiveComputation]

Absolutely not. Neither I, nor Searle, say there is anything about cognition is of necessity specific to carbon. Both Searle, and I, critique computation, and computation is, of course, multiply realisable on any appropriate substrate: mechanical, electrical/silicon, pneumatic/air or water etc. Where I depart from Searle is my positive thesis, which asserts mind is contiguous with the living (for complex reasons best examined via Evan Thompson's opus, Mind in Life) ..

[@luketracey3269]

🍀

[@luke2642]

Non deterministic machines break his entire argument of causality, reproduciblity and equivalence? Even many deterministic iterative algorithms must be computed, there aren't shortcuts. Experience probably lies in the process of computation, not the states of a finite state / infinite state machine. Balancing the random jitter vs the logic seems an intuitive foundation for e.g. creativity. Our brains are probably not even quantum, a bit of randomness and a lot of complexity is enough.

[@CognitiveComputation]

The DwP reductio explicitly target Finite State Automata so cannot be 'broken' by mere non-determinism, albeit I think you are conflating non-determinism with Stochastic [Turing] Machines in the above.

[@luke2642]

​@@CognitiveComputation Thanks for your reply! On one level, ordinary Turing machines are abstractly irrelevant, because infinitely long tapes are impossible. So this leaves us with only finite states. I'm happy to accept these internal states are mapped by the observer to the meaning in the world. You can't assume meaning is evenly distributed across all possible combinations of states, evolutions of states, or all possible mappings though (parallels with entropy, boltzman etc). I don't know precisely where it all goes wrong with Putnam's proof, or Chalmer's proof etc, but by observation, finite brains don't need an ever exponentially increasing number of states to percieve the world to self identify as intelligent. Even if they did, that's not even a problem, as adding just 1 bit doubles the number of states. Brains & computers can easily add... I don't know... 100 bits a second and multiply their states by 1.2 x 10^30 every second? However, the key I think, any argument based on determistic, purely causal systems is as irrelevant as infinite taped turing machines, because a) the observer principle (observation affects stuff, even in classical physics) b) chaos (e.g. three body problem is not soluble, initial conditions not sufficiently measurable) c) there is no substitute for actually doing computation, you can't shortcut to the final state (ill posed problems, halting problem). I have more questions for later! :-)

[@mamalada3296]

I agree. AI can give every answere but it has noting to do with " the truth".

[@aBigBadWolf]

13:53 ... did you just steal clips from numberphile?!?!?

[@MachineLearningStreetTalk]

Yes, we feel it falls under fair use i.e. MLST is educational, not commercialised in any way whatsoever, this content is transformative and substantial (i.e. we are only using their animations nothing else), we are just using a couple of tiny clips in a 1.5hour video, I will add their credit to the VD

[@colinmaharaj]

Hmm what will be a comment about chat GPT now

[@Achrononmaster]

@7:50 nice bit from Wittgenstein, he was not completely stupid. Symbol manipulation is an _effect_ of sentient thinking, not a _cause._ No one has a single clue what the cause of subjective mental qualia are, not you, not me, not Searle, not Pearle, not Hofstadter, not Chollet, not Crick, not Chalmers, no one. But especially not Crick.

[@Chr0nalis]

In my opinion this channel would be better off with ideas from younger folk, i.e., PhD students who are actively thinking about and solving new problems.. not emeriti which are way past their prime. I noticed that even in much larger podcasts, such as Lex Fridman's, way to much emphasis is put on what titles people have rather than what they have to say.

[@pebre79]

Smells fishy because it seems like he is adding too an air mystery to the point of making intelligence mystical. Sounds like woo woo

[@jadeanjoun]

Finally, collaboration that AI is, in fact, stupid!

[@CristianGarcia]

This might be a personal bias, but I find that believing in non-computable things within our universe is the same as believing in ghosts. Couldn't pass beyond the intro :( BTW: intros are getting better and better, amazing job!

[@CognitiveComputation]

There are Aleph zero computable functions. There are Aleph one non-computable functions (ie. uncountably infinitely more non-computable) with, as Nathanael highlights below, the halting problem being the most famous..

[@nahakuma]

The fact that there are non-computable functions does not mean that there are physical systems that execute non-computable processes.

[@nahakuma]

@Nath Well, ghosts "are" in the universe just as functions are. Anyhow, what the original comment intended is irrelevant to the problem I am posing. The existence of non-computable functions is basically irrelevant unless you can point to some physical system that instantiates one.

[@nahakuma]

@Nath I fail to see how the existence of a concept with no physical counterpart, just like ghosts, is an argument for the mind performing non-computable processes. I could similarly argue thay the existence of ghosts as a concept is an argument for the existence of life beyond death.

[@iamacoder8331]

Nice content but intro is too long.

[@filoautomata]

This is my theory (based on my experience with psychology, religion, and machine learning). 1. It does not matter whether computer will later have conscious or real understanding, AI (programming v.2.0) is not stupid compared to animal, but stupid compared to human. Intelligence is relative to what it is compared with. (ex: newborn baby is stupid compared to their parents). 2. We constantly confused with body and psychology (soul, mind, etc). Body and even computation is an interface for human soul to act and interact with the world. Even if a computer can do all human can do 100x better, without human soul, it will always be an artificial intelligence, the same as artificial neurons will never be neurons even though it may simulate neurons perfectly or even faster. The different is qualitative. 3. Machine will never be able to hold responsibility for it's own because it does not have soul; the same with human will always be responsible for their dogs, but not their child (if the human child becomes adult, then the child are considered not responsibility of the parent's anymore). 4. This kind of discussion is not evidence based discussion but faith based discussion which is interesting to be discussed but not useful for basic development of machine learning itself.

[@bradmodd7856]

hybrid intelligence is conscious and you are reading one's words right now

[@ElectricWound]

What drives me nuts is, that Mr. Bishop arguments against AI's actual existence based on logical experiments, that in fact only prove that it is important do declare what is inside or outside a system *before* you analyze it. Based on his assumptions you could easily prove, that human beings are not intelligent. His argumention is fundamentally flawed on the level of definition.

[@asage5801]

The brain is merely a transform function that receives various kind of energetic stimuli and outputs the realization of potential realities and hallucinogenic combinations of situations that the universe provides...it’s not a computer; it’s a combiner/compiler/correlator uniquely suited to feed off of the local environment so that the universe can begin to try to understand itself. There.

[@stalinsampras]

So that the universe can begin to understand itself 😦, Where can I read more about this ?

[@asage5801]

@@stalinsampras what I wrote was meant to be a humorous riff based on fanciful thinking. Since I made it up, there is likely not anything substantive written about this idea. Nice to meditate on though.

[@zrebbesh]

I'm really sick of this. "This is worthless because it cannot solve this problem that I have defined as being insoluble for it" is fecking vacuous. We might as well be listening to Searle.

[@dru4670]

Day 2 of asking you to invite Stephen wolfram on your podcast 😂.

[@dru4670]

Is P = NP ?

[@sshdvac3372]

Consciousness is an attention vector making subject to fit into the environment.

[@plafar7887]

?????? Lol, what??

[@sshdvac3372]

@@plafar7887 it makes for example conscious some virtual things, like economy

[@plafar7887]

@@sshdvac3372 I think you're confusing the 'Hard Problem of Consciousness' with the 'Easy Problem of Consciousness'.

[@CristianGarcia]

Professor Bishop: AI is stupid. Also Professor Bishop: Has an AI startup.

[@MachineLearningStreetTalk]

😃

[@pedrosanchez4035]

He do not claim that it is not useful.

[@eladwarshawsky7587]

If you believe that everything is part of god, then pan psychism isn’t really that odd

[@Georgesbarsukov]

I just don't like his argument. He makes premises that anyone engaging in the Socratic method would questions and tare apart They all boil down to a simple opinion. He completely ignores the concept of emerging phenomenon. He's the kind of person that would say that random is random so it's useless.

[@Georgesbarsukov]

Perhaps it was as he said and the format just didn't go his way.

[@datrumart]

We can see Yannick refused to participate to this BS 😎

[@CognitiveComputation]

Dear Alex and Arvind. I am human, fallible and have said silly things many times but the nice thing about academic discourse is that the door is always open for others (you guys; Yannick etc), to engage with the literature and rebut my position [in peer-reviewed journal ideally]; be cool to see that, instead of mere ad hominem..

[@OnTrackwithWalker]

MuZero goes Brrrr: h ttps://kzbin.info/www/bejne/pqW9qKJuqKiFi6s 2 + 2 = 5 if it serves the Party. Love you guys.

[@quebono100]

Zero!!😎💥💥💥💥

[@mattizzle81]

The argument that everything would have to have a mind is like saying everything must be an aeroplane. Not everything has wings and not everything has neurons. Sounds like nonsense to me.

[@Xaminn]

Disclaimer: I didn't watch the full video... So, take my comment lightly. It just seems odd to me that someone can make a claim saying "Artificial Intelligence is stupid and casual reasoning won't fix it." Yet, doesn't provide a working model that would replace the 'bleeding edge' technologies. *Rant over *

[@CognitiveComputation]

Apologies for my numerous shortcomings!! Actually, my life work, such as it is, has been in AI - specifically 'Swarm Intelligence' - where we have proven some quite interesting results establishing: (a) Stochastic Diffusion Swarm Intelligence meta-heuristic is a very efficient search and optimisation algorithm; (b) it can be cast on a connectionist architecture and hence offers an alternative insight into neuronal operation; (c) we can use it to demonstrate how populations of simple ant-like create can 'play' the game Hex and (iv) demonstrate that SDS is Turing complete. Incidentally, I recently left academia to join @Fact360 where we successfully deploy AI and NLP to help reveal fraud ...

[@Xaminn]

@@CognitiveComputation Thank you for taking the time to respond to my rant. You are definitely more qualified than me. (So please excuse my ignorance.) I will research the topics you have mentioned and see what I can apply and learn. Best of luck on your journey and I’m looking forward to seeing more of your work. Cheers Prof J.

[@astrolillo]

Pretty bummed to see the direction this channel is taking, I thought it was going to be focused in relevant research in ML and AI in general. If I wanted to watch a video about the nth take on the tired Chinese Room argument, or what it is to be a bat? or Chalmers' "objections" I would go and watch a philosophy of mind channel with its 132 subscribers. This kind of reasoning will advance 0 the field, this is XXI century "elan-vital" . The only positive thing is the absence of the Jason-Statham-wannabe student. What a poser.

[@CognitiveComputation]

To make up for the absence of the Jason-Statham-wannabe student, as Yusuf/Cat Stevens [ancient hippie singer] once sung, "I was once like you are now ..." :-) .. having spent 40+ years building AI systems, and attracting considerable research funding to do that, I, of course, disagree that philosophy - and the CRA in particular - is irrelevant to AI research.. Sadly, time he who ignores history is doomed to repeat it .. But, if nothing else, I hope these 'patronising' (sic) remarks from an ancient retired academic, long past whatever prime he might once have ever had, inspire you to even greater things in your own research, that you might one day show him wrong ;)

[@astrolillo]

@@CognitiveComputation Thanks for the song I really enjoyed it.

[@CognitiveComputation]

@@astrolillo Ha, TBH I was anticipating a somewhat more 'agricultural' response, but happy you enjoyed :)

[@originalveghead]

Why did it take 19 minutes for you to introduce this? Is this for the benefit of us, the viewers? For the love of all that's conscious, let your guests talk!

[@MachineLearningStreetTalk]

Check the table of contents 😎

[@CognitiveComputation]

@@MachineLearningStreetTalk I thought the intro excellent and, personally, found that it only served to clarify my rambling expositions :)

[@Shikhar_]

Podcasts dont have a fixed duration. So it doesn't matter if its 5 mins or 20 mins of intro, as long as it adds value.

[@originalveghead]

@@ai._m thanks for that valuable context. t s pot