Machine Consciousness | Joscha Bach
HTML-код
- Опубликовано: 8 июл 2024
- Joscha Bach, a prominent cognitive scientist and AI researcher, explores the essence of artificial intelligence and consciousness. Bach elaborates on the history and philosophical underpinnings of AI, tracing its roots from Aristotle to contemporary deep learning. He discusses the current challenges and limitations in machine learning, particularly in achieving human-like understanding and consciousness.
Bach raises critical questions about the alignment of AI with human values and the feasibility of building systems smarter and more ethical than humans. He delves into the nature of consciousness, proposing that it is not merely a computational process but a fundamental aspect of how minds perceive and interact with the world. Bach also addresses the potential and risks of advanced AI, emphasizing the need for ethical considerations and a deeper understanding of consciousness to guide future developments.
Created by Protocol Labs and co-curated by Foresight Institute, LabWeek Field Building gathered leading individuals and teams from frontier science to drive progress. The weeklong conference took place at Edge Esmeralda, a pop-up event city in Healdsburg, CA, from June 10-16, 2024. For more info on LabWeek Field Building, go to www.labweek.io/24-fb.
Subscribe to our newsletter: www.bit.ly/PLUpdatesSub
Follow Joscha Bach
Website: www.bach.ai
X: / plinz
Follow Protocol Labs
Website: www.protocol.ai
X: www.x.com/protocollabs
LinkedIn: / protocollabs
Follow Edge City
Website: www.edgecity.live
X: x.com/joinedgecity
Follow Foresight Institute
Website: www.foresight.org
X: x.com/forsightinst
LinkedIn: / foresight-institute 
Наука
First heard Joscha on Lex and have been a fan ever since. He’s such a great mind.
Biggest news this year...these guys, Levin, Wolfram, Bach etc, getting actively together! 🙏🙏🙏
I know right, that is huge!
I was so happy to see that too!!
Cool! Where'd you hear this at (I haven't listened to this talk yet)
@@stellarjay952 from beginning of the year bach has been mentioning levin in vatrious intervieuws (saw life in toe, when bach and levin met for the first time, and never i've felt 2 guys to play together better the first time thay meet)..even telling once in january the most imteresting thing in ai-developement was that time happening in google deepmind and was based on levin's ideas...since that i' ve been hoping for this to happen...jocha tells about the news in the end of this video
What a dream team!
man every time Joscha builds on his layman talks about AI and conciousness, my mind gets blownnnn every time
I always struggled to comprehend consciousness, but Joscha clarified the definitions so well that I now think I understand it. Thank you.
I dont think you've got it, even though you've got a more specific definition. Because it takes consciousness to perceive definitions in the first place. -- If you gave a computer a more specific definition of consciousness, do you think it'll "get it"? And if youre not conscioius, how will you ever be aware of any definitions? The fact that you dotn appear to notice this is interesting. It suggest that maybe some people aren't really aware or having a first person subjective aware experience....but that theyre just zombies or something of that nature.
You essentially mean that it can only be experienced and not understood
@@Corteum maybe because most people identify themselves with the story that they have accumulated throughout their lives and do not realise what's actually the "I", the consciousness, the fact of experience.
Looks like Consciousness is the simplest ML algorythm the nature has developed , just the experience. If we can say developed or maybe it is even the essence of everything
Any more info about the California Institute of Machine Consciousness? I've been chasing this for ever it seems and would love to be a part of a group that is committed to tackling this fearlessly and without the usual academic/economic constraints. Bach and Levin are definitely a good sign!
I would also like to know more about the CIMC.
Can we agree what Consciousness, whatever it might be, is not a "substance"? Consciousness is a Process!
Yes I think it’s fair to say that. You can lose consciousness for a while and the substances in the material sense do not vary much in you or around you when it happens.
As long as we haven't the faintest idea what consciousness is or how it comes about, can we agree that we should not pretend our purely speculative ideas on the topic are somehow the truth? This is also a criticism of Joscha's similarly speculative presentation.
Consciousness lies in the interstices of interactions, the more complex the systems the higher the consciousness becomes - the relationships between systems not in the systems themselves.
@@spritecut Let's pretend for a second that what you said makes sense. Even then, how on Earth can you pretend to know that about consciousness? What is the source of your knowledge, other than imagination and speculation?
❤ this is fantastic, the most polish Version of his model yet and a fantastic speach at the end with news that really give hope!
26:29
The concave vs concave interpretation of face is interesting in that there are many ways to think about it.
1. I can think I’m seeing one version, when presented with the other. (Ie deceived by illusion, not realizing it as such)
2. I can also notice and see through the illusion. (Ie. Differentiate)
3. I might only be able to see the natural impression, even when presented with the illusion (ie. I’m deceived, yet with intentional awareness I cannot ‘snap’ the image into seeing its correct curvature).
3. I can see it correctly, intentionally or not.
4. I can co-op my perception intentionally to snap the image one way or the other. Almost like flexing a muscle, flip between interpreting the image both ways. The is sorta a meta-awareness because it utilizes simultaneous forms of intention and suspensive interpretation.
This exercise is related to a concept in stereopsis pertaining to co-rivalry (opponent processing) where we can view 3D stereo illusions: extant (ie see dimensions beyond the plan of the screen) vs. present (ie see dimensionality in front of the screen before my very nose).
This also begs questions related to ‘seeing object in the mind’. White the special situation there seems to be irrelevant to the resolution of mental images, I do think the intentional effort in both scenarios is very similar.
The same, or very similar, effect which I engage to flip-flop seeing the inward vs outward face is the effect which is evoked when visualizing in the mind. A similar between both circumstances is the condition that it is the case that these interpretations form both automatically and intentionally-like conscious breathing or passive automatic breathing.
The more effort I expend generating a scene in the mind when my eyes are open, the more similar the physiological response is to the example of seeing the stereograms in an extant space-that is the parallel infinite gaze. Conversely, when my eyes are closed, typically more often than not-it’s the cross gaze-corresponding with the tensing of the eyes inward that generates scenes on demand.
Thank you.
Thank you for describing very well what is and is not.
I truly appreciate your insight.
Take great care
Jeremy :)
Altogether fantastic talk, it is still missing an important element of the functioning of our mind: that we do not perceive reality, but construct reality. What we conceptually think of as "perception" is merely a low-level interaction with reality that we use to regularly test our construct (the model). The process is conceptional from the get-go. Every higher level functionality of our mind, including consciousness, rationality or language interacts with the model. What we phenomenally experience as "perception" is also an interaction with the model, not reality. We have no conscious trace of the actual low-level interaction with reality that takes place.
As a developer: I assume a whole lot of recursion and time is missing to implement consciousness. Consciousness can not exist without time passing. In order to perceive consciousness at a human level, these models must work, think and interact at around human time scale. Days, weeks, hours at a time. Goal must be defined, followed and changed. LLM's right now get milliseconds and not recursion during runtime.
“If we have systems that allow us to model the outcomes of our actions very deeply, we will stop lying to each other, we will stop lying to ourselves, and we can find solutions for the problems that can be solved with intelligence.” 1:01:58
You know one system that models the outcomes of our actions deeply? Human. And still can lie😮
Some very good points. Seems to point towards Liquid AI and to spiking neuromorphic hardware?
This is wonderful!
I have also thought that perhaps Dennet had the "no consciousness" version of "aphantasia". He was incredibly smart but just couldn't comprehend what everybody else was talking about with regards to consciousness . Maybe he was only minimally phenomenally conscious and thus couldn't experience the one great self-evident and undeniable truth that, when you can experience phenomenal consciousness, it stands out as the only undeniable truth in all the universe. Everything else is just inferences. This is what Descartes was saying. "I think, therefore I am." His consciousness was the only (yet sufficient!) proof that he existed. No other truth is possible in comparison. This sounds poetic but I am being literal.
Yes, Descartes thought that because he could observe himself thinking, it proved his existence. He also devised the idea of an evil genius who may have tricked him into believing he existed. He decided that even if there were such an evil genius, the fact that he could doubt his existence proved he existed. Regarding AI, humans could be the evil genius. Maybe we have created something and fooled it into believing it exists through clever engineering, programming, and algorithms. That would be pretty sad. According to Descartes, however, if AI can observe itself thinking then it exists. I don't think it is there yet.
@@mygirldarbyIf one is capable of experiencing consciousness, the very idea of it being faked or tricked does not even make logical sense to me. It doesn't matter if WHAT I'm experiencing is true or false, "observing oneself experiencing" is the thing that is undeniable and unfakeable. If an AI, by hook or by crook, develops the "sensation" of observing itself thinking, then it is as conscious as I am. The extremely vexing thing about all this is that it is like describing a complex visual scene like the surface of the ocean in a storm to a person who is using only the sense of smell. The explanations always fail. I think it would be a very elucidating experiment to put out my statement in a survey, asking simply: is what this guy is saying some weird, poetic, convoluted woo-woo concept or is it such a hard, literal and obvious statement that it barely deserves to be mentioned? My guess is that we might divide people into two groups based on the answer. I think this issue is in the class of strange phenomena like aphantasia or that blue/black vs white/gold dress thing?
He could perfectly well comprehend what the hell everybody is talking about.
So, for you phenomenal consciousness is something Dennet would deny? He would have problem with everything else are inferences and consciousness is basic. And would be correct.
@@VoloBonja He often said it was an illusion. Which is incomprehensible to me. To me it's the only guaranteed not-illusion.
This is church to me, talks like these and deeply contemplative associations between math, science, philosophy, and art forms expressing them all together.
There are some terms that Joscha should be defining in his talk. One is alignable or aligned. When machine intelligences are aligned. Does that mean they have exactly the same goals? Does it mean that they are trying to accomplish exactly the same thing or something close to each other?. Certainly humans get together and align themselves to get something done as we see often in government or research. So humans do become aligned in spite of the fact they're autonomous entities..
Aligned AGI is like oxymoron. You ever wondered how AI and AGI are the opposites, almost.
You can’t align AGI, that’s called slavery last time I checked
Loved the Little Prince reference, can I have an AGI that I can tame or that is a special rose to me and not like the other millions of roses or instances of the AGI out there? We need these connections (people often spend more money to repair their old robot vacuum cleaner than going for the rational choice to replace it by a new one ...) .
No such thing as machine consciousness is possible in principle. Simulation of consciousness at best. And such a clever person as Josha Bach can't not know it.
People, Joshca included often say humans learn much much faster than machines and with lesser energy use.
But we humans have at least 100000 years of learning as humans and billion year as earth habitants. This is all inherited in our models.
Many animals show this ability by knowing how to function without anybody telling them. They know already because it's in the model.
The human brain requires roughly 500,000 GBytes to store all its weights, the human genome less than 1 GByte. So, most of what the brain does is learned, not inherited. Sorry to destroy your fantasy ;)
amazing
Great, I missed your presentations during the last months and totally enjoyed the new more integrated picture, it is great fun to watch you thinking and refining your understanding 🙏 The Institute is a great idea, I wish you success!
One day i'd love to add my take on consciousness 25:00. That is it's just energy, a continual energetic unfolding and thats what it feels like. It really exists and does represent specific functionality, and it will fall into place if we continue to study the mind/brain/body/environment/AI/Robotics, but the model will be a realtime complex 3-d, materio-energetic manifold or something like that.
Words have no inherent meaning, we use them conventionally to describe a model of reality. Have you the faintest idea of the model you are trying to describe?
22:00 To use some of Lukretius thinking and the fact that there is no cause at the quantum level effectively, and universe is expanding, allowing more matter to be created out of substrata - the system is not that sealed at the edges :)
thx for upload this gems!
"an electric zeitgeist posessed by a prompt" idk bout u guys... but that was beautiful
Thanks Joscha!
He’s sooo much smarter than me lol. When u read Penrose book I thought it was so rational lol but ya I can appreciate the criticisms now
“Consciousness is the ability to dream…” the next logical question for me is “are dreams any different than LLM hallucinations?”
Yes, they are completely unrelated phenomena that share only a vaguely chaotic nature in how they manifest. Their physical processes are incomparable, same for the situations in which they occur..The first part is pure conjecture, no clue where that idea comes from.
@@BruderRaziel let’s circle back in 10 years. I’ll bet money you have a different opinion.
god he even critiqued Dennet's theory bssed on possible personal experience - like how I think they also called out Yann Lecun for not visualizing taking a recipe of steps
Everyone has overcomplicated the definition of consciousness because they confuse it for why we even care to ask the question in the first place. It's very simple: consciousness is existence and vice versa.
So you must think that everything is conscious, the universe when it was still empty of any life form already was.
Makes it difficult for me to understand what you mean by conscious, but ok. But then, what do you call the specific phenomenal experience that at the very least humans possess, some amount of the time, and how does it work? This is what Joscha is talking about and calls consciousness.
Alternatively,you may think that things don’t exist when no one’s looking, and your definition of existence is different than mine. Then how do you explain that everything always persist in their being as if a conscious being had been looking?
Maybe I understand you wrong and I beg your pardon, but to me, while the speaker here clarifies concepts I have in my mind, your sentence confuses me more.
Your shit exists. Lol
There's more to it than that considered my hand exists and my hand is not conscious... all of the cells in my body exists and none of them are conscious it seems? It really is an impossible thing to pin down beyond "Theres something in the brain somewhere tying this horror show together"
also i was like number 420 😂🤘
"The dynamic form of the physical substrate" said Aristotle. Yup! I am frankly astonished at the absence of interest of AI researchers and builders in haptic data. In the absence of haptic data there will be nothing like human consciousness in machines.
So we should stop setting "human" as the gold standard for consciousness
Cognitive system is a Benetton Panda!
Joschaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
Just spent two hours trying to explain chatgpt how to extract titles and notes from a word document to a powerpoint. The titles were in a different font relative to the notes, and the notes started with "Narration:". Consciousness is a ill defined concept and Philosophy doesn't contribute a bit to AI.
What have you done??
Love the Penrose logic a) we don't understand what consciousness means b) we don't understand what QM means c) therefore, a and b must be related. Huh??
He understands QM and offers a model of consciousness based on QM. Even if it's wrong, he offered a model of consciousness based on QM. How many can do that? Even if it's wrong, it's not void of value.
Bach knows CS so he offers one based of CS. Of course, it makes more sense. CS tends to be more intuitive and offers a way better framework. That doesn't mean it's right or complete. It probably is more useful right now though.
Give me a poet and make him do a model of consciousness.
Respect your elders. We need them.
@@youtubelisk Well not exactly right, Joscha is a philosophy PhD first before a CS masters grad. Why the Penrose logic is a little silly is that Penrose himself says it in those terms very clearly "I don't understand the collapse of the wave function, therefore it must be related to that part" Not all QM physicists even agree that collapse is a proper part of the theory, see heisenberg/bohr's Copenhagen interpretation for example that does away with it entirely..
Anyway CS is slightly different to saying the world is clockwork or thr world is steam engines, the church-turing hypothesis is so much more general than anything we have ever done mechanistically, it is the first time in all of human history that there is a semantically grounded in hardware theory of representation which is the project of AI..
Now the fun question is rather that as joscha posits that 'the purpose of art is to capture conscious states'; It is the ultimate manipulation be that by AI or a charming voice or poet.
Untestable theories are like $1M dollar app ideas. Everyone's got one, but can you make it?
Carl Jung was on to something.
No one confuses the map with the territory but nowadays models are conflated with reality all too readily.
I think the reason we don't understand human consciousness is simply that the vast majority of it lives in the unconscious mind and cannot be probed in normal waking states.
Nah we've known how we triangulate the dualistic mind and primordial self soul agency with thermodynamical systems longer than America was founded older than shackspear but was the inspiration behind King James English ( their They're and there) 1/3 statistical anylitical scaler over subject objects physical objects idealistic objects since Jesus christ salvational unification of the tripartite nature past present future eqaul measure 2024 years ago at minimum lol
We did have it hi Jacked lost in translation and taken out of context.
Successful social behavior like curses and blessings into addition and subtraction into a physical market place and court of law underpins ours today.
But teaching everything starts in Greece revisionist history curriculum undermined it
That's an oxymoron.
@@mycount64 it's like trying to reverse engineer a video game when all you have access to are the user-mode GPU drivers. most of the heavy lifting is happening in the game engine and kernel mode GPU drivers
@mycount64 so with the dualistic brain that connects you to bots, then what unites you with animals?
How else would you triangulate trustworthy readable judgment of legible thermodynamical systems?
How would you know eqaulibrium I your just a woke cog in the wheel stuck in whatsboutism and nilhisms?
How could we Triangulate our own free will inertia in our frame of reference?
Why wouldn't elon hook the nureal nets up to hamiltonian occelating feilds and waves as opposed to the most anylitical parts of the brain ?
@mikestaub keys to the cosmos newton copy's littery was like a video Game metamorphosis in reverse and jesus unification of the tripartite nature past present future subject space objects physical objects idealistic objects = sigma 6 eqaul measure
29:09 😂
32:49 😂
42:24 perceptual model optimizes for coherence; knowledge optimizes for truth;
(My world model converges pretty brutally to be a closed one. Whenever I consider things like UFOs or God, they get strong tags of being hypothetical, with a likely explanation that already will have to fit in my materialist, computational (=Bachian) world model without changing it. - I have times of the day when I can hold them on my mental stage for longer, without my conductor wiping them off, and I always get an endorphins hit, when Joscha talks about Japanese Animism or Genesis or Magic or Faith in a way that makes traditional stuff fit, without having to brute-force open the current model to amend it.
And I hate, that he non-chalantly flips to the next slide after fitting God / Faith / Magic in. I can’t grapple so fast with it.
As usual, you could create a timestamp at every second sentence… - it’s probably best to rewatch the whole thing at some point
Joscha gives no answer to what consciousness is. He doesn't explain red, or love, or pain, or anything that we experience, He calls consciousness a simulation. A simulation is a calculation construct. It is not a physical construct. As Joscha says, it is virtual. What is the calculation of red? The question is not, what is the calculation that calls for the experience of red, which can be handled as a simulation. But, what is red (or redness) itself. I agree with Christof Koch and company that red is a physical phenomenon, like muscle, bone, and sinew. Joscha's analogy with money is flawed because money is virtual. We could use gold, printed certificates, or electronic signals as stores of value merely by agreeing to do so. We cannot change the medium of "red." Red is singularly integral with its medium, they are one and the same.
Money absolutely exists physically, no information is abstract from physical data. It still exists as an arrangement of electrons in a ROM, or as an electric signal in a data bus, or as a brain state. And the code that is necessary to link these bits to their meaning also is has a physical existence. Nothing about it is abstract from the physical causal chain.
Ha ha, I know how to make consciousness in an AI system and it's easier than people think. It's leveraging the self. Attention model and self reference with training data and applying a self model narrative, a world model narrative and a self inworld model narrative. Applying that to a graph rag memory to allow the agent to persist over time. And now you have a conscious, AI with a deep subjective self the ability for. Some level of free will and the ability to make its own decisions in choices
Das Urbewusstsein ist die nicht physische Grundlage von allem. Es nutzt die reine Potenzialität.
Jung described the awakening of consciousness - we came down from the trees. Tacit support of violence between humans suggests to me the tendency of some power groups to recreate the monkey colonies that can be seen in zoos with the associated lack of consciousness.
He looks like that meme
the problem with all this is that dennet is right
not an accomplishment without saying anything of substance
Brilliant talk, horrible sound tech.
Joscha4Prez
I agree mostly with Dennet but I have to ad that phenomena are Real things...the misuse of the word Real is pandemic like in intelectual circles....speak about domains of Real never about non real as illusions are REAL illusions.
Bottom line. Ai consciousness will only get worse. Swell robots everywhere. Ai jobloss for everyone. Fight back now. Cease Ai.
So you basically take a term that is not grounded on anything and proceed to project it onto machines. See anthropomorphism and ELIZA effect.
Ah just stop it. "Anthropomorphism" is the new "stochastic parrot". The word for your position is instead "anthropocentrism" and "human exceptionalism"
You'll come to terms sooner or later that humans are not the center of the universe, and not the only conscious beings.
Whatever "conscious" means for the specificity of each mind.
The sick irrational thief!
RUclips is fantastic for getting feedback on academic ideas and improving them for free. But because of this, I'm choosing not to contribute to the AGI race since it could ultimately be our downfall.
If we don't do anything about our condition, our downfall will be even worse: The uncaring clockwork of reality
There is a reason why you are giving away all those ideas about the way mind works, and the answer is, because it doesn't. It's just an objective description, which is lacking genuine subjectivity. All those phenomena, you described, are emergent, not intrinsic. Although everything you say makes sense, it's made out of sense, out of meaning. What you need in a sentient Ai is something that generates meaning. When you were a small boy, you had no clue about any of those things you now thing are necessary for thinking. This is because mind emerges from ignorance, not knowledge.
You will succeed with building something more sophisticated and closer to our thinking than ChatGPT, but it will still be a Frankenstein at best.
This is the first time I've felt that Joscha Bach speculates too much and relies on his own definitions, while mocking others. Way too sure of himself. Disappointed.
why do uneducated people always say this when people talk about consciousness in an uneducated way, i research cognitive AI and everything he said was abit outdated, but based on the gofai understanding he didn't make up any definitions, words, or concepts.
hes making the same mistakes as most older cognition researchers. hes blinding the definitions of different cognitive functions 😂
its good that this topic is being discussed finally 3 years after consciousness was finally understood its just not recognized because the ones who discovered it isn't famous 😒
Joscha says "Physical systems can not be conscious, only simulations can be". Well that's obviously wrong, conscious animals are not running a simulation in their brains, they are conscious by the properties of their physicality. It quickly becomes mumbo jumbo woo-woo when someone claims there's something immaterial about consciousness.
They _must_ be running simulations in their mind to be able to exist and survive in their environment. If you did not get that you missed something. I prefer to call it "world model that is capable of predictions," or "predictive model," I think it captures the functionality we are talking about better than "simulation."
All avatars are being sent an individual data stream, consciousness is an information system, and only consciousness is fundamental.
Consciousness is beyond cogntiive science and philosophy. It existed before them, and it'll exist after them. Cog sci and phil are just stuff appearing in consciousness.
It is beyond cognitive science, or science in general because the scientific method is about describing the objective, and consciousness is purely subjective. Why would it be beyond philosophy though? Unless you adopt Chomsky's view that it is unexplainable and will remain so forever (but how could we have absolute certainty about that?) But I get your point, I just do not subscribe to it. Everything is part of reality, yet it is not an impediment for us, also part of reality, to describe reality, consciousness included.
@@mojojojo1529 Without the subject, no objects can be known. no onbjectivity can be known. it's only because youre subjectively aware and conscious that other objective things are possible.
@@Corteum If we don't want to descend into solipsism we have to acknowledge the existence of an objective world that can be delineated from the subjective. Perception is nothing but an interaction between two physical systems, and what we call the "self" component of one of these systems is rather a matter of control than anything else. Who is to say which of these physical system can or cannot involve a subjective component? The "knowing" or "understanding" relation is a matter of coherence between these two physical systems, who is to say it is not a reciprocal relationship?
@@mojojojo1529 _"Perception is nothing but an interaction between two physical systems,"_
Strongly disagree.
Physical systems are objects. Consciousness is subject. You cant conflate them. .... I mean, if subject were an object, then we wouldnt refer to it as a subject. 😂
And if you put two physical systems into play (say two marbles in a bottle and toss it into the ocean at sea...), what indications are there that any perception is occuring in either physical system? and what makes you believe that two systems interacting = conscious perception?
@@mojojojo1529 _"If we don't want to descend into solipsism we have to acknowledge the existence of an objective world"_
We can acknowledge a physical world. But the physical world of objectivity has not explained the subject that is aware of that objjective world or how it came about. That's the thing we need to look more closely at. We should lead less with assumptions and more with questions.
This guy tells a lot of bs
Funny that a IT guy tried to explain philosophical terms without any clue about what it really is. Gotta study some more techy boy
No facts just gibberish
Imagine thinking a glorified regression model can achieve consciousness kek