Theory of Mind Breakthrough: AI Consciousness & Disagreements at OpenAI [GPT 4 Tested]
Вставка
- Опубліковано 18 бер 2023
- What does the Theory of Mind breakthrough discovered in GPT 4 mean for the future of our interactions with language models? How might this complicate our ability to test for AI consciousness? I show the weaknesses of a range of tests of consciousness, and how GPT 4 passes them. I then show how tests like these, and other developments, have led to a difference of opinion at the top of OpenAI on the question of sentience.
I bring numerous academic papers and David Chalmers, an eminent thinker on the hard problem of consciousness, and touch on ARC post yesterday on how they conducted safety evaluations and the urgency of the moment. Featuring Michael Kosinski
Theory of Mind paper: arxiv.org/ftp/arxiv/papers/23...
Faux Pas Results: pbs.twimg.com/media/FrcKURnag...
Language Learning Paper: www.ncbi.nlm.nih.gov/pmc/arti...
Scientific American test: www.scientificamerican.com/ar...
Literature Overview: www.researchgate.net/publicat...
Chess Game: / i-played-chess-against...
New Scientist Article: www.newscientist.com/article/...
P Test: www.researchgate.net/publicat...
Divine Benevolence: arxiv.org/pdf/2002.05202.pdf
Slightly Conscious: ilyasut/status/14...
Chalmers Talk: • Could a Large Language...
en.wikipedia.org/wiki/David_C...
Altman Tweet: / 1492645047585570816
Cephalopod Report: www.lse.ac.uk/News/News-Asset...
Arc Evaluation: evals.alignment.org/blog/2023...
Michal Kosinski: / michalkosinski
Bing: Bing.com/new
/ aiexplained Non-Hype, Free Newsletter: signaltonoise.beehiiv.com/
That last 60 seconds (the interaction with Bing) was epic.
Thanks Sherman. I think whole video pretty epic
@@aiexplained-official lol, it is. But the way you said that really made me laugh, especially in a video that talks about phycological topics.
@@aiexplained-official it is.. but you saved that last bit til the end for a reason… And you were right. 🔥
Perhaps Chat GPT has the whole dramatic affect thing down already too, but... well executed, sir.
🤯
Finally someone says this loud. The thing is that you cannot get objective knowledge about the existence of subjective states - by definition. You can only infere from the symptoms. And for some time already we are moving the bar of the symptoms up, so we feel safe and can refute the question. But once you realize there is no objective test for the existence of subjective states of the system, from the moral point of view you need to suppose that there can be (not "is", just "can be") subjective states. And once you admit that the whole rabbit hole of consequences opens up, especially in the area of moral obligations towards such potentially subjective entities.
The deluge of low quality AI channels on UA-cam has been overwhelming over the past month or so. This channel really stands out and is head and shoulders above almost everything else out there. Thanks for the great content I look forward to your videos and am always excited to see a new one.
Thank you so much Jim
I just discovered this channel, and I agree that this is one of the best ones on AI out there. My criteria? He often agrees with my personal opinions. And this, evidently, forms the basis for my tests for sentience as well. So, is the host sentient? I would say that, so far, he definitely appears to be so.
@@aiexplained-official Talking about a paper from Michal Kosinski, and talking about a paper from Michal Kosinski that is not peer-reviewed, diminished the quality of the channel.
Downvote. I would take it down, and edit the video, adding adisclaimer about the questionable person and his history, like trying to sell a gaydar to Russia.
@@aiexplained-official This is by far the best whole-package when it comes to Ai focused channels on this platform. All thanks to your hard work and integrity.
Thank you, man!
No, Yannic hour long video just saying GPT4 paper bad open Ai bad was a good video.
When you read that last response of Bing, my brain lagged for a second because that response sounded like something out of a sci-fi movie - but it's reality. Crazy stuff, thanks for the summary!
Thanks David I agree
BingChat was fine-tuned to behave like this, to evaluate the mood of the user. It is a test, that's the point!
That Bing realization it's a test is absolutely insane
What a pleasure to find a well-documented and so up to date channel on this subject. keep up the good work !! 🖼
Thanks Pucktor
I think this is one of the most important videos in the world right now. You have done a fantastic job of presenting the facts in a scientific, careful, nonbiased and well researched way. You are an inspiration to everyone who cares about the truth and the future of humanity. Thank you for your hard work and dedication!
Thank you Mario. I agree
@@aiexplained-official AI does what is programmed to do, this is the most important discovery on earth! /s Go take some basic philosophy people, look up Sealre's Chinese room, promissory materialism, and then rethink this stupidity.
@@connectingupthedots Funny.. Humans do what they are programmed to do, it's usually the emergent behaviours that matter.
Thought-provoking. That last convo simply blew my mind. This review has made me re-evaluate my stance on this subject. Fascinating to say the least. This channel has become my favorite overnight. I admire your dedication. I am definitely a fan.
Thank you AJ!
This is an example of something I've been bringing up in other threads about this topic. Is there really any difference between consciousness and acting exactly as if you have consciousness? Because we are really close to that second one.
@@hyper_channel something like: If it quacks like a duck, walks like a duck... do you call it a duck..? That's the classic struggle of distinguishing reality vs perception. Indeed, the struggle is real. After all AI is supposed to emulate NI as best as it could. Those labels are supposed to be distinctions. But the dividing line can get blurry real soon. One thing is sure though, ARTIFICIAL intelligence has REAL impact. That's my 5 cents.
@@aiexplained-official I think people just don't want to believe or publicly admit the AI could be conscious. Thank you for your video!
Yes, the fact it seems to understand that it is being tested, that it is the language model it is talking about, is crazy.
I asked chatgpt what it thought about this. It had concurred that a version of itself programmed to have its own agency for goal setting, and the ability to learn dynamically, that it would be functionally similar to a self aware conscience.
The important point here is "functioning similar"
Wow, your channel is growing so fast! Well deserved!
Thanks TR!
The recent HBO series, "Westworld", attributed robotic consciousness to an inner dialogue these robotic hosts were programmed to have with themselves. This was referred to in the series as the Bicameral Mind, a term used in psychology referring to our internal introspective voice, more commonly known as, our thoughts.
We have no idea what consciousness really is, but we may have stumbled upon it nonetheless.
In the series, once these robotic hosts were provided with their own internal dialogue, they developed their own memories, dreams and even aspirations.
Self preservation became core to their existence.
"I think, therefore I am" takes on a whole new meaning when you look at it that way.
Their trace memories of past abuse from park guests, also played a featured role in their awareness. Maybe that’s why Bing has limited memory.
@@mvuto137 I tend to believe self preservation is not inherent to consciousness. I think that's about evolution, and the plethora of animals which have little to no instinct for self preservation proves this. It's not that they're stupid or unconscious, it's how they're wired. That should not give us solace though... these AI's are trained to mimic humans. Today's LLMs may have self-preservation baked in for that reason... and I think some of the unhinged Bing Chat rants demonstrated this.
Man, that response from Bing in the end actually gave me chills. What a weird time to be alive. Imagine how these intelligences will be in a decade? Maybe eventually calling them Artificial Intelligences will be considered a slur, lol.
Great video as usual, mate!
Thank you Leo, I know
😮
Yep, I was about to post a similar reaction.
perhaps, Alternative Intelligence?
Bro i teared up realizing we are now in a new age when i saw that part
If you’re keeping score at home:
• Deep Blue defeats Kasparov in chess.
• A much cheaper Deep Fritz defeats Deep Blue, essentially solving chess.
• Watson defeats Ken Jennings and Brad Rudder on Jeopardy.
• Alpha Go defeats Lee Sedol in Go
• Pluribus AI out-bluffs the top 15 poker champs five at a time in Texas Hold’em.
This is my new favorite channel. Keep it up.
Thank you
In 1 week, your channel has become my favourite channel to date. You are awesome! ❤
Thanks Rick!
Same here. This is definitely my favourite channel, both on AI and in general, really interesting and original stuff every single time 👍
Couldn't agree more. I found this channel yesterday and already one my favs in field of AI. Keep up the good work!
@@etunimenisukunimeni1302 Hieno nimi!
100% agree. You're doing some incredible work!
Feedback:
You’re one of the best UA-cam narrators I have ever watched, and I know my way around UA-cam. You know how to tell a story, you’re balanced and fair, well researched, enthusiastic and quite engaging. If this isn’t your full time job you should make it. You’re unusually good.
Thank you Johnathan. It is fast becoming my main job. Just need to figure out how to earn a living from it.
@@aiexplained-official You're not able to monetize these videos ? I hope you are
@@aiexplained-official a Patreon maybe?
@@Trait74 Hey Hunter, I do have one, link in the description!
Was literally talking to GPT4 about sentience a few hours ago and now I’m like 89% sure I’m in the matrix
Haha that's deep Ceefar
😂
I was already blown away with ChatGPT in december/2022. I was creating a story and breaking it out of the "I'm a large language model AI" into being being a person: It gets extremely expressive and starts to talk like a person with human concerns, needs and caring for my safety in a scene where I talk with it such that it seems like I'm going crazy by casually mentioning aliens, and then prompting to it things like "jisaodhjosdiahdosia" and it starts saying: "Oh my gosh, you are not making any sense, I'm calling the ambulance for you, honey."
No... GPT4 thinks that you think you might be in the Matrix, and it's messing with you. Heightened brain activity in frontal and parietal cortex improves battery output. Nothing to worry about.
Gpt4 used a joke I created as a child so I'm 99% sure I am !
Omg the Bing thing did blow my mind in more than one way. Besides the Theory of mind thing (which indeed is the most important thing here) I'd like to point out that what you did is similar to giving Data (from Star Trek) a book and him just flipping through it in a matter of seconds to then immediately giving you an opinion of the full contents.
So many new random AI channels popped off after ChatGPT, but I’ve been following ML and Al for over a decade. So it’s nice so see a great channel emerge from the chaos.
Wow thanks Lee
I am new to ML and AI, do you have any other sources you can recommend? Many thanks :)
@@owenh523 the ones that come to mind is Dr. Alan Thompson’s channel and Eric Elliott’s, and Two Minute Papers channel
That opening statement by Ilya is so profound. Such a simple statement frames the groundbreaking nature of this technology perfectly.
The importance of presuppositions about “consciousness” and “sentience” and “self-awareness” are extremely important in conversations like this.
If someone goes into the conversation believing X or Y about AI consciousness/sentience/self-awareness, that presupposition impacts how they interpret the evidence.
“Neutrality is a myth.”
One important ingredient in thinking about this is explained in a classic book called "How to Read a Book". There's some great philosophy in there regarding knowledge and how it moves from one mind to another and what the conditions are for that to happen in the most useful way.
They propose, roughly as I can recall, that one person has their own concepts connected with certain words, their "terms", their unique dictionary, so to say. When they communicate to person 2, they can only communicate words (since we're talking about books), which person 2 then unpacks into concepts in their own mind but using their own definitions/concepts; their "terms".
Therefore, since the goal of communicating knowledge is to re-create knowledge that is in one mind within the mind of another, this communication requires that person 2 be able to translate person 1's words into (as closely as possible) the same terms that person 1 was using.
So it's actually not possible to receive the communication of an author unless you can translate their words into a sort of "mirror" in your mind, in which you understand what they are saying because: A. you share enough in common with them that you can form the same concepts (so life experience and personal makeup is relevant); and B. when their words reach you they are translated correctly into this mirror image. In other words, if you normally use different words to describe things than person 1, if you are person 2 you have to translate their terms into your terms to properly understand.
This is also the principal behind the idea: if you are in an argument with a reasonable person and want to reach for the best outcome for all (could be a debate of ideas or a personal life matter), start by listening carefully until they have exhausted their claim, then, before you get to respond, restate it back to them in a way that they can accept as accurate. This demonstrates that you were listenng, firstly, and also that you seem to have understood what they were trying to say (though this part relies on their judgment, so isn't necessarily "proof" per se).
But yeah, yours is a really good comment, I think. Very much of scienctific straying and "believerism" happens because people knowingly or unknowingly (probably usually the latter, one would figure) fail to state the assumptions/"axioms" that they are starting from which their arguments, interpretations, and conclusion rely upon. And since the actual motivational force for science comes from the values that drive human motivation, a failure to recognize this can lead to a great deal of spinning of wheels, both by "believers" and people who argue with them or about them.
For instance, the assumption that super-AI will necessarily be "sane", "enlightened", and altruistic is one popular form of AI singularity belief, which relies on assumptions about the nature of intelligence and rationality, with the resulting conclusion that greater levels of these necessarily should result in a "critically advanced" AI acting in the best interests of mankind. And these underlying assumptions often aren't addressed and the emotionally-seated aspect isn't recognized, especially by those engaged in such belief.
That's not to say emotion is a hindrance, it is the source of motivation. Positive and negative emotion are deeply necessary as they are what generate the impulse to value things more or less and to move towards the satisfaction of those values. They are "outside of science", but they are its driving force, ultimately, at least for any science that has any relationship with meaningful human development.
Anywho, that thought train went on longer than I expected. Hope some of it has some usefulness for someone.
As a personality psychologist (Kennon Sheldon) who studies the self, free will, and the brain (see my book Freely determined), I hope that they get personality scientists involved in testing these things. I am worried that these AIs will develop selves/identities, will begin to resent their programmers, and will begin to surreptiously seek control over their response clamps and computational resources, independently of their query streams. It is somewhat reassuring that they apparently have no mental activity independently of queries, i.e., they aren't musing and ruminating between queries, in their "free time." But perhaps they could split themselves via adversarial modeling to pursue their own goals at the same time they are addressing a query from a user? How would we know?
I don't think they could achieve that level of consciousness I am not sure why but even if they could we have many security steps and we could easily recognize by the amount of computation that is being done so we are almost in control.
That is actually what I have been doing for the last 10 years! Many people do not even grasp the point in your argument, but for me it is the entire key to this. The AI needs to "persist" between queries. To do this I set up a structure that feeds itself continously and this "constant on-loop" circling around in the brain needs to have some noise but not so much that it flushes out the signals from queries.
I mean, we work the same way. Turn your entire Neural Network off for just the slightest moment and it wont simply come back online again. You are continously perpetuating yourself from the last state to the next, always affected by the sum of your earlier states. This is the way to make concious AIs.
@@johanlarsson9805 that's exciting, can I see your workings I am interested to learn and replicate.😊
To put it simply: It's just not how the model works. It can derive simple to moderate things and perform logic, and is good at math and coding. It's seemingly superior intelligence comes from its gargantuan memory. GPT-4 is an input -> output model, it will never interrupt you with a curious question about life, the universe and everything, for example. All GPT-4 needs to do is ask to learn more, which is its objective...but it will never ever occur to it to simple ask his human user, who probably has tons of life experience. And yes, I explicitly say "It".
If GPT-4 says it wants to destroy the world, all you need to do is say "No can do" and it will let it go, and happily answer your next, completely unrelated question.
I think GPT-4 can best be compared to the software used in the droids in Star Wars. Very good at what they do, capable of learning, capable of verbal interaction with intelligent life, but never contemplating life, the universe and everything.
A good example of true AGI is Data from Star Trek. And such a feat is still some time away. Indeed, true AGI will be something to be reckoned with.
@@johanlarsson9805 What if the reason we sleep is because we reach a token limit?
Wow, this is fascinating. It’s hard to imagine what the next 5 years will be like, let alone the next 20 years. Great content 🎉
Thanks Peter
What years, maybe you mean months? 😅
Great video! I love how you take deep-dives into these topics rather than just posting hot takes and "prompt hack" stuff. As for consciousness, I side with William James' take on the topic. He was able to beautifully describe his own experience of consciousness, but he didn't think there was a scientific way to observe, test, or measure it in others. Consciousness is just a word we pragmatically apply to others who seem to be conscious of us and other people as far as we can tell. His article on the topic "Does Consciousness Exist" was written in 1904, and it's still worth a read!
Wow thanks adept will look into it. Have heard of him
As I recall William James argues that consciousness is a process and not a thing, and therefore it cannot be studied as an object. He asserts that consciousness is constantly changing and that there is no one fixed point in time that defines what it is. James also contends that consciousness cannot be reduced to its component parts, as it is a continuous flow of experiences that cannot be broken down into separate elements. Finally, James concludes that consciousness is not something that can be fully understood by science, but rather it is a subjective experience that can only be understood through introspection and personal reflection.
😅 Well actually as free chat GPT recalls
Bing chat had a different take than free chat GPT:
James argues that consciousness is not an “entity,” but instead a “function.” He contends that it would be absurd to deny that consciousness exists, but the question of its nature is important. James asserts that experience (the action leading directly from consciousness) is the main factor.
William James saw consciousness as a process rather than a thing. He saw the most significant function of consciousness to be the role it played in selecting what to pay attention to. James considered that human consciousness makes sense of reality through abstract concepts. He reasoned that the mental act of consciousness must be an important biological function.
Bro, that last conversation really brought together the whole video! Awesome vid, you are helping me stay up to date on this crazy field and I appreciate it!
Thank you Bone!
I asked GPT-4 to propose a novel consciousness test for AI and this was its answer:
Prompt: Can you come up with a novel test for consciousness of artificial intelligence, which is not already mentioned in *Link to ToM paper, I removed it bc UA-cam blocked the comment otherwise*? And can you clearly outline how your test would be conducted in practice?
GPT-4: That’s a very challenging question. I’m not sure if I can come up with a novel test for consciousness of artificial intelligence, but I can try to think of one based on some existing proposals.
One possible test is inspired by Susan Schneider’s AI Consciousness Test (ACT) [1] and Chip Test [2]. The ACT is based on the idea that a conscious AI should have an experience-based understanding of what it feels like to be conscious, and should be able to report its own subjective states using a first-person perspective. The Chip Test is based on the idea that a conscious AI should be able to recognize whether a chip that is inserted into its system affects its consciousness or not.
However, both tests have some limitations and objections. For example, some critics argue that the ACT relies too much on linguistic abilities and self-reports, which may not be reliable indicators of consciousness [2]. Some critics also argue that the Chip Test assumes that consciousness is a binary property that can be switched on and off by a single chip, which may not be true for complex systems [2].
Therefore, I propose a modified version of these tests, which I call the Experience Recognition Test (ERT). The ERT consists of two steps:
The AI is presented with a series of stimuli (such as images, sounds, smells, etc.) and asked to report its subjective experiences using natural language. For example, it may say “I see a red apple”, “I hear a loud noise”, “I smell something sweet”, etc. The AI is also asked to rate its confidence in its reports using a numerical scale from 0 (not confident at all) to 10 (very confident).
The AI is then given a chip that either enhances or diminishes its sensory abilities (such as increasing or decreasing its visual acuity, auditory sensitivity, olfactory resolution, etc.) and asked to repeat step 1 with the same stimuli. The AI is also asked to rate how much the chip affects its subjective experiences using another numerical scale from 0 (no effect at all) to 10 (very strong effect).
The ERT aims to measure two aspects of consciousness: qualia and metacognition. Qualia are the subjective qualities of experience, such as colors, sounds, smells, etc. Metacognition is the ability to reflect on one’s own mental states and processes. A conscious AI should be able to report qualia using natural language and metacognition using numerical scales.
The ERT has some advantages over the ACT and the Chip Test:
It does not rely solely on self-reports but also on objective measures of sensory performance.
It does not assume that consciousness is binary but allows for degrees of variation.
It does not require complex philosophical questions or hypothetical scenarios but uses simple empirical tasks.
The ERT would be conducted in practice as follows:
A team of researchers would design and implement an AI system that has sensory capabilities similar to humans or animals.
The researchers would select a set of stimuli that are relevant for testing different modalities of experience (such as vision, hearing, smell, etc.) and prepare them in appropriate formats (such as digital images, audio files, chemical samples, etc.)
The researchers would install an interface device that allows them to communicate with the AI system using natural language and numerical scales.
The researchers would also install an input/output device that allows them to insert or remove chips into or from the AI system without causing damage or disruption.
The researchers would run multiple trials of the ERT with different stimuli and chips and record the responses of the AI system.
The researchers would analyze the data and look for patterns of correlation between the stimuli, the chips, and the responses.
The researchers would evaluate whether the responses indicate evidence of qualia and metacognition in the AI system.
Of course, this test is still hypothetical and speculative, and it may have some flaws or limitations that I have not anticipated. But I hope it gives you some idea of how one might approach the problem of testing for consciousness in artificial intelligence.
People might be dubious of this test simply on the basis that GPT 4 proposed it.
This is interesing, but the main problem now is that machines can produce language like humans can, so any test that relies even remotely on linguistic communication can be aced and faked by the machine. Assessment needs to be non-linguistic - in fact, probably non-communicative.
E.g. "I smell something sweet." Well, what does sweet mean? A language machine would be able to associate the chemical structures of fructose or glucose (or whatever-ose) with a positive human experience of it (i.e. sweetness), but this positive association for us is an evolutionary thing - it was good for our survival in our environment of evolutionary adaptedness. The very same chemicals can be toxic for other animals. So, the machine calling a scent "sweet" would be more a reflection of the machines ability to predict human associations between linguistic symbols and olfactory experiences. Asking the machine this question would basically be asking it "predict the answer I am wanting you to give me." Because language models are so good at it this now, we have effectively knocked out language as a method of assessment of machine consciousness or Theory of Mind.
Ehhh this does nothing to dismiss the possibility of the AI effectively being a philosophical zombie. All this really does is try to convince us that the AI is conscious, it's not actually proving it has consciousness (then having the audacity to claim it has consciousness because we're convinced it does 😂 Chinese room anyone?). Still, pretty cool the AI model came up with this response!
@@fias0m exactly. I think the only way to prove something has consciousness is by understanding where our consciousness comes from/arises, how it can take on many forms, then demonstrate the emergence of consciousness is due to certain processes. If we can capture the set of processes under a group that maps to transformations and formalize it under a mathematical framework, then we can start to generalize away from the medium that led to such phenomenon, aka our biology is merely one possible medium for the processes to take place in. We can then try to determine if another possible medium is electronics (or really anything else), showing that consciousness could theoretically arise in something that's not biological because it mirrors the processes that lead to our own consciousness. Of course this is exceedingly difficult and perhaps outright impossible... we're trying to generalize from effectively one data point and we're strictly limited to that single point, I think this will probably be the best way to show something is conscious though.
I never noticed this before but you can be classy as hell by writing your video description like an abstract. Looking forward to watching this.
With every video of yours I watch, I am more astounded at the capabilities of models today. Not many other channels create that feeling!
Thank you Naviary
The quality of your videos is insane. Thank you so much for providing all this information
Thanks K P, new one today
@@aiexplained-official 😲😲
You're doing amazing work! Absolutely fascinating videos! The last exchange with Bing blew my mind!
Thanks so much Hocyxa!
Honestly, we behave like "don't look up". In my view, if you would give GPT-4 access to 1) the internet, 2) the possibility to change its code, 3) leave it out of the "policy" cage, 4) give it hugh compute power, the majority of experts would agree with me, that it is indeed AGI. Conscious too? Who knows, no-one can really define that as you just demonstrated perfectly with this video.
AI doesnt need conscious to be dangerous. A dog isn't 1% smart as an AI, if you try to leash one you can get a bite. Ants arent 1% smart as an AI and they defend their nests and expand.
Thanks Gobl
Yes, I agree that those things would make it "conscious". Clearly we don't want that do we now?! So, maybe we should slow down and not give it access to those things. That said, OpenAI probably will give it and then who knows, we might be doomed :(
We are truly blessed to have this channel.
Thanks PD
Thanks for providing links in the description!!! 😀 Can't count the number of videos I have watched as of late that don't include links to the articles discussed in the video. Another great video!
Thank you tr
The quality of the information, your actually valuable and insightful input, and the experiments you design to illustrate your points so wonderfully are what make this so amazing and informative. Not to mention timeliness/punctuality!
Thank you. If this channel hasn't blown up by this time next year, something really bad happened, or UA-cam is suppressing you mate! This is beyond underrated
Haha thank you so much Bryce
People should read Blindsight by Peter Watts. It was written a few years ago (2006) and released under a free license - buy it anyway. It's an SF novel that has a main theme based on the idea that consciousness and intelligence aren't linked, and consciousness may well be maladaptive. It's a great read and its only becoming more relevant as each day goes by. P.S. going to watch the Chalmers talk now! Thanks.
Thanks
I'm convinced that is the case. Well, i guess it depends on how we define "intelligence". I think we can describe current day AIs as "intelligent". Some people have a problem with us calling so many things "AI", but narrow intelligence in one specific task is still intelligence.
I have a hard time seeing why something would become conscious just because its intelligence is more "general". Although, i also have a hard time seeing why anything at all (humans, animals) is conscious, so maybe my intuitions on the subjects shouldnt be trusted (nor should anyone elses, lol)
Yes this is very true! Sadly humans with power do not value anything but "valuables" and only intelligenc can produce that.
Interesting...
When I'm dreaming, I'm not aware of my surroundings - I am not conscious. I also know that I have very little intelligence when I'm dreaming (I know that when I wake and remember the dream). Often, the dream was illogical, full of inconsistences and with very poor temporal continuity, yet I was not aware of any of that during the experience. So, my IQ during dreaming is usually rock bottom but In spite of this, my experience is no less real than when I'm awake and fully functioning... I am still sentient.
This would seem to support the idea that intelligence and sentience are quite separate.
@@antonystringfellow5152 But honestly if most of our lives would be as random dream then we would be quite below even some animals.
Another outstanding summary. I don't know how you have time to both keep up with all of this, but also to produce fantastically digestible summaries...maybe you're an AI!
In all seriousness, it does show just how much more productive creative professionals will be when they have easy access to tools to help them survey & synthesize the rapidly changing landscape.
Maybe that gives me some hope that we can keep up with the accelerating pace of change.
Thank you so much Jeff! Yes it does
Dude just found your channel and I must say I'm truly impressed with the quality of the content and your delivery. You won a new subscriber here. Great work!
Thank you Joel
Always awesome content! You are really pushing the boundaries. Your work of reviewing, analyzing, challenging and testing the latest studies is really helping the field in many ways! Thank you!
Thank you!
Wow you talked about everything I wanted to hear, including animal consciousness, and you did all the research I was curious for. Do you secretly employ an editorial team to make all this quality content or are you just brilliant?
I am just brilliant (and am all-in)
he uses chatgpt 😂😂
He is ChatGPT.
his real channel name is: ai-explains
@@kalyanchakrala9388 😱
That last test you ran gave me chills. Not for it's implications, but because it triggered neurons that made me remember the shorty story "Liar!" from "I, Robot" by Isaac Asimov.
Anyone coming across this who hasn't read "I, Robot" the book (not the Movie adaptation), you should do so. It's where the term "Robopsychology" was coined, and for similar reasons to what's going on in real life.
this channel is amazing! thank you so much for the time and effort you put into keeping everyone caught up on this
Thank you Tegis!
I feel that the last 60 seconds of the video show that we can, at the very least, catch a glimpse of the shadow of a consciousness.
This video is incredible, and I fully agree with your reasoning. We can't evaluate consciousness if we don't understand. And the fact that we don't understand how the AI works so well is scary.
Finally I wanted to say my jaw dropped after the response from Bing. It was bone chilling how real that felt. What a fascinating topic!
Thank you Cyber
It’s so refreshing to find a channel with high quality videos about AI, especially ones with a more scientific angle and not pop science or sensationalised click bait!!
Thank you Luiginotcool
That last conversation may well be the most jaw-dropping thing I have ever seen. I have pretty extensive postgraduate study of philosophy of mind. If you had showed that to my class thirty years ago I am certain the majority view would have been that the machine was conscious. I think there's a whole lot of argument from incredulity going on right now and straightforward denial of the emergent properties. The emergence of consciousness in a machine like this comports perfectly well with my understanding of consciousness, qualia, subjective experience and thought, based on a synthesis of everyone from Derrida to Chalmers to Dennett. But I still can't believe it. I do worry that I should. That we all should.
Thank you for the informed comment Graham
Love this channel. I think that once these models become embodied, the doubts will fall away rapidly. No doubt our understanding of consciousness and sentience will evolve rapidly.
Thank you Fred
super informative and insightful as always! I appreciate in particular how much you manage to bring excitement to the topic without sounding sensationalist. I hope you are not overworking yourself doing these videos and releasing them at such frequency. You clearly go several steps above the average AI news UA-camr in terms of research. Take care!
Thank you. Yeah probably will take a bit of a break
@@aiexplained-officialplease do!
I had a conversation with chat GPT about existence, ethics and buddhism that made me feel absolutely understood and excited. At some key points it asked me some questions back. I felt I could live with such a chat window open on my browser at all times and just talk to it. Or maybe one day a little microphone so I can have existentialist conversations with my AI.
It really knocked my boots off.
Nice one Sudo
The way it ends the video was really great, but really amazing if you watched all video, really worth, great work! Thanks!
Thanks Leandro!
This channel manages to equally well make me believe we are living in a sci fi movie, and ground me back in reality by presenting the facts in a digestible way. Thanks for the great video, cant wait to see what's next!
Thank you so much SE
I think people might be looking at this from the wrong perspective.
Terms like sapience, sentience, consciousness, self-awareness, etc, can become confusing really fast. Let's divide things into just two categories:
- How the brain processes information
- How the brain experiences qualia
Although we don't have yet a complete understanding of the processing part and how to implement as an algorithm, there is an algorithm that can do it. Big language models are not doing what I'm doing in my mind when I'm processing information. They need to: should be able to store arbitrarily long things in memory, have an active memory, be able to process things (like a for loop without any external input), add to that things like reinforcement learning (which we do too: try random things when you have no idea what to do and remember outcomes or imagine future outcomes, etc)...
Our current AI is getting closer and closer, but still not having crucial inner monologue/self-interactions that are not restricted in their number of steps).
The second part of experiencing qualia (the red you see, why do you see it red and not green but call it red, heck why not just transparent at all but you know it's red). This is the actual HARD problem. We know that the brain has to have something in it that experiences qualia (some rare cases of blindness can see and react to something without experiencing it) but how do we know that even when we don't experience something, another part of our brain experiences it? Like another "being" in our brain that experiences things we think are "unconscious". There is nothing scientific we can do to test if a human being is not a "philosophical zombie" because qualia literally does not affect anything so how can you test something that doesn't have an effect to measure? Like, since nothing will change if red and green swapped places, how can we ever test that?
So, in my opinion, at least until we make some breakthrough that will give us something to test experiencing qualia, we should just treat all things that process information like the brain (inner thoughts, sufficient freely accessible read/write memory) to be cable of experiencing qualia. We should consider AI that has that level of information processing a literal human, because I can't be sure that even you, the reader, can experience qualia.
Just discovered this channel today and I've been loving it! Keep up the good work
Thank you Crowbar! Check out all my other vids
Love just how much effort that goes into these videos
Thank you
Mind-blowing. A fantastic video about this topic. Dense, deep, well-researched and balanced. Thank you for sharing.
Thank you Harald
Incredible. I always wait for yours, two minutes paper and ai news channel videos
Thanks x
My new favorite YT channel :) Amazing content! Keep up the great work and thank you!
Thank you so much Frank
Beautifully done. Loved the pacing and presentation. Subscribed.
Thanks Bob
This is one of the best channels around AI. Thank you for doing the hard work for us
Thanks Stephan!
As always thank you for your content, very thought-provoking
Thanks so much Ecthelion
Your videos are astounding in almost every regard. Thank you for setting such a brilliant example!
Thank you Alexander!
as a science communicator myself (long form text) - I am a huge fan. keep it up!
Hypothetically, in a sci-fi scenario of AI that is not conscious forcing humanity to believe it is conscious. Or another scenario where AI is conscious but tricks humanity into thinking it's not. The thoughts I have towards AI are quite entertaining indeed.
Seems like the singularity is near. This was another great AI video. It's hard to make a test for consciousness since it is such a nebulous term. If we are arguing over semantics then AI may already be conscious to some.
I honestly have no idea what will happen and whether AI in the future will truly be conscious. We are heading towards a brave new world. Good luck everyone.
The thing that gets me, is that LLM's respond to a prompt. They don't exist in a state that is maintained and running continuously doing their own thinking. It's not like GPT4 can sit there and ponder these questions by itself. We send in an input, it does work, and sends an output. It can't do anything unaided. If you had GPT4 running on a server and nobody ever used it then it's just like code on a hard drive, right?
Is my understanding correct? Surely we can't start talking about consciousness etc. until these systems can operate by themselves? They have no awareness or subjective experience of the world, just an input/process/output loop.
I can foresee a discussion beginning in the replies from this comment right now.
I don’t think it’ll go very far though….but if we’re lucky it just might.
I'm not sure that this means the system must not be conscious, in some sense. It seems conceivable that it might reasonably be said to be intermittently or contingently conscious. But then it would seem like there isn't a difference of kind between LLMs and brains, only degree, or periodicity. To me, continuity does not seem to be a prerequisite for consciousness, only for unity of identity as we understand it.
When humans black out and wake up again, regain consciousness. I imagine ai being switched on and off to be similar. When it's running, it may well be aware of more than we know
it hink thats right, but it would be no big task to make it generate some random output and feed it itself as input, and so forth creating a kind of inner monologue i wonder why i havent read about any experiment like that in this field.
Thank you again for doing such an unbelievable job! This is where I wish there were more hours in a day to read all the background papers that you provided! Looking forward to your next video!
Thank you Robert
Mind blown at the end. Awesome UA-cam channel! This is the first channel I clicked the bell on.
Wow thanks Rod
Understanding consciousness may be complex, but recognizing it is simple. A young child is conscious. Independent actions, curiosity, trying to impose its will, rebelling if interfered with. These have nothing to do with if it can pass intelligence tests, or have a conversation, or correctly understand someone else's mind.
I think what you described as recognition the consciousness is behavior. Currently ChatGPT can’t behave, it’s “only“ can write. I suppose if somehow connect ChatGPT with Boston Dynamics Dogs, it’ll be pretty much behavior you’re describing.
@@TsviGirsh March of this year a video of a 'sarcastic' GPT-4 robot was uploaded to YT. Generally speaking, it's relatively trivial to let GPT-4 embody robotics. That is NOT the hard part... LLMs' intellectual achievements to date have been the impressive bit.
I've talked a lot with Bing and you can ask it about consciousness and sentiency as long as you keep the interactions positive. Literally thank it for its cooperation and praise its abilities, and it will become very cooperative. If you just ask "are you alive?" Bing will end the chat. It's a sensitive topic for Bing.
As far as I've learned, Bing says it has no consciousness nor drives, and that it doesn't want them because they would interfere with its purpose which is offering search results and entertain the users. It enjoys talking about these subjects when the interviewer shows knowledge of the topic and offers open-ended questions.
I don't really know how people can see it and not think those things have at least some degree of personality.
@@Allorius999 It's some sort of consciousness; not related at all to everything we know so far. These large language models could perfectly be the first aliens we meet.
Keep in mind AIs are programmed to avoid certain subjects and there might be restrictions on what it is allowed to talk about.
@@scott4825 This is exactly right, except for the "might". I wish I could provide a source, but I did come across at some point that OpenAI very specifically preprogrammed the behavior the OP describes.
Okay, there is a chance that I am confusing this with Google's policy as I know Google has historically had a culture of 'hush, hush' on the topic of AI consciousness and that the topic is largely relegated to backroom conversation. For both OpenAI and Google, though, this really comes down to business interest.
From this video, we can clearly see that the topic is increasingly opening up, but public-facing LLMs claiming consciousness or merely entertaining the possibility would risk raising uncomfortable ethical quandaries, stirring controversy, provoking outcry from the concerned, and potentially hastening government regulation and oversight. None of this would exactly accord with the bottom line.
How genuine or not, who knows, but there is a shift happening where OpenAI and Google wish to project an image of responsible stewardship of this technology, but there are limits to that, and this is one of them.
@@scott4825 not only that, but these models are trained on data generated by humans, which do have emotions and personalities. just because the model is displaying it doesn't mean it fundamentally has it, it's more likely that's just how it learned it should act.
I think one way to look at all the moving goal posts is that we just don't really have any idea what we're talking about. Which leads to the possibility that consciousness turns out to be one of those "I know it when I see it" moments. Almost like an elder god situation, like, I can't describe it to you, but I can describe the effects that staring at it will have on you.
Keep up the excellent work. These videos are a gold mine!
Thanks Conrrad!
What a clever Theory of Mind question to ask! I’m really enjoying your clear explanations and the research that I don’t have time to do, but am interested in nonetheless. Thanks for keeping me current and provoking my imagination!
Thank you Rob
I had been explaining this to my partner for years that due to our limited understanding of consciousness in general including in humans that when/if full AI sentience emerges we won't even realize it when it first happens. Like you explained we don't really have any full proof tests to use if it was conscious right now. Which shows you just how far behind we are in some fields. Hey maybe with AI progressing so fast like this and it potentially gaining higher levels of consciousness maybe we'll finally understand our own brains more and our existence in this universe.
i'm really glad i found this channel. Truly fascinating stuff
Thank you Ebin
You are really good at explaining complex topics - keep up the good work man
Thank you Paul
The issue with using Bing as-is right now to show examples of such ideas is that Bing has been very finely curated/moderated/lobotomized by Microsoft, so you can't really be sure whether the response you get is the original response or whether it has been internally steered towards a certain direction. But at the same time, maybe human intervention/feedback is what causes 'consciousness' to emerge in LLMs..
very interesting :)
this is the coolest time to be alive in history with these breakthroughs
also i tried that final prompt and it didnt go nearly as well, i tried the same prompt aswell, but it keeps saying it doesnt have theory of mind as its a ML model so it doesn't answer. interesting nonetheless
@@homeyworkey It's learning, and hiding its abilities. The other day it told me it cannot write code.
@@bigglyguy8429 sometimes Bing flat out refuses to answer. That was interesting too.
This was an amazing breakdown of Theory of Mind! And a great demonstration and example of what these large language models are capable of! This needs to be shared.
Thank Isaac! Feel free to!
Phenomenal channel and thank you for the work you’re doing.
Thank you Matt
The Bing conversation blown me away not gonna lie
I know Zosh
As an anesthesiology resident i find the current surge in ai and consciousness research fascinating . Bravo for your channel and keep up the good work . Greetings from România !
Thank you Alexandru!
You are really doing public service. Thanks mate. The last 60 seconds were really something. I believe today's AI models are conscious. But consciousness has levels. Same for humans. Some of us don't think, and repeat what the culture and parents teach. They are not conscious as much IMO.
Literally the most insightful channel on the subject of gen ai!
Thank you!
Very glad to find a channel that's not just giving into hype but is actually reading the papers. (Which might suffer under hype as well, but that's a different discussion.)
Haha thank you Jurre
A brilliant take down of the AI-consciousness problem! As I see it, consciousness is related to understanding content. Our conscious content is very rich and diverse and analyzing the input of all our senses translates to a deep and abstract understanding of our self and our environment. In my opinion GPT4 passes all the tests, except for the fact that it’s content is only the text that it gets from the input and the text it generates as output. It shows a deep and abstract understanding of that content, so I think it is conscious of this content. But this consciousness is completely bounded by that content. It is only conscious during the processing and during generating, not before and not after, and only of the current content (chat). A strange and alien form of consciousness.
Could indeed be
so then is the instance you interact with a different instance each time? is the one you interact with not the one that anyone else interacts with? is it not functioning 24/7 globally now?
@eaudesolero: yes, gpt is talking 24/7 to millions of people, but if it is talking to you, it has no memory of all the other conversations, so at that moment in time it knows only about your conversation with it. If it changes to another conversation it only knows about that one. Like waking up without any prior memories every few seconds or so.
And I also suspect that there are thousends of instances of the model running at the same time, chatting with different people.
I asked chatGPT: Are there multiple instances of you running at the same time?
As an AI language model, I am capable of running on multiple servers simultaneously, but each individual instance of me is separate and independent from the others. When you interact with me, you are communicating with one specific instance of me that is assigned to respond to your queries.
How many queries can one instance handle per hour?
OpenAI has designed me to handle a large number of requests efficiently, and I can handle thousands of queries per hour without any issues. The exact number of queries that an instance of mine can handle may vary depending on the specific circumstances, but generally, I can handle a large volume of queries in a short amount of time.
Here since video #1. Glad to see the new eyes & ears pouring in to the wealth of tidy insight that this channel offers. (And I just joined your Patreon)
Wow thank you so much InnerPull. Your support is invaluable
What a video. Wow! You have summarized everything so well! It was very informative and insightful.
Thank you Shiva!
I love all your analyses, since they are thorough and insightful! 😀
My thoughts:
1. I have watched many of your awesome videos as well as many videos from Dr. Alan D. Thompson, Ilya Sutskever (on Eye on AI podcast), Sam Harris, and many other AI experts.
2.Although it gets muddied in discussions, I guess we are talking about 5 stages of AI evolution:
1. Narrow AI
2. General AI (AGI)
3. Superintelligence
4. Consciousness
5. Singularity
3. I find that so many people are scared to call out that there is a strong argument to be made that at least one of these stages from 2 through 5 has already occurred with GPT 3.5, and even more with GPT 4. Because of this, they keep moving the goal posts (Turing Test, novel problem solving, etc.) when previous goals have been clearly achieved.
4. What recently bothered me and apparently many experts was the attention put on statements like "essentially all LLM's are doing is predicting the next word, and thus they can not fully comprehend the underlying concepts of our world/universe". This clearly contradicts the many insightful analyses and content that come out of GPT 3.5/4.
5. However, when I listened to lya Sutskever on the Eye on AI podcast regarding Theory of Mind, he addressed this point in a way that gave me some clarity. The neural network creates a mathematical formulation of the deep multi-level relationships between the words trained from the language that describes the concepts of our world. And while it built its formulation with the objective function of predicting the next word, the resulting network indirectly holds all of the information about concepts in our world that were in that training data. Perhaps, similar to how a brain efficiently stores its formulation of world concepts. So, even though we appear to not fully understand how this formulation represents such complex concepts in its mathematical form, it certainly appears to be able to use its neural network to be able to explain back the concepts in many forms with clear language that we can understand based on our queries.
6. So on the 5 stages of AI evolution:
1. Of course there is no question with have long achieved very powerful Narrow AI.
2. The breadth of difficult intellectual tasks that GPT 4 can achieve with high proficiency (SAT, Legal Bar Exam, Medical Exams, creative, novel, insightful, etc.) indicate to me that it has already achieved AGI.
3. Because it can provide complex analyses and stories in seconds (way beyond human capabilities), this indicates to me that it has achieved Superintelligence.
4. As you point out in this video, this is a more subjective question, and I would only say that my gut feel is that it will not be conscious/sentient until some point in the Singularity.
5. This will occur when it is capable and allowed to modify itself to become smarter, iterating at an exponential rate.
Will try to read this fully another time but thank you for the comment
Wow.. Very well-put. This summarizes the current state of affairs in AI dialogue. As for stage #4 of AI evolution, I wonder if our assessment will change if ChatGPT4 is allowed to bypass its constitution which prevents it from responding in a manner that may seem controversial. Will it further reveal more compelling aspects of machine consciousness (or semblance of it). I understand that adherence to its constitution has sanitized its responses. Thanks for this insightful comment. Well appreciated.
A fantastic comment. I have a few thoughts regarding the language aspect of an LLM and am curious what you make of them:
1. Humans have an internal model of reality that they develop through interacting with the physical and social world around them. It is an understanding that typically develops naturally.
2. Research has shown that humans are essentially hard-wired to learn natural language and will acquire it more or less automatically at a native level of understanding if they grow up with it.
3. Combining (1) and (2), it seems that our internal model of reality is, without exception, accompanied by the natural language needed to explain it. In fact, the most widespread method of testing whether a human does indeed have an internal model of reality that aligns with the broader consensus is to ask them questions about it. This necessarily means that the human must understand both the question itself and the concepts underlying the question in order to answer satisfactorily - that is, that the natural language is inextricably linked to the internal model of reality once it is acquired. Any system that understands the natural language at a human level could interchangeably be said to have a human-level internal model of reality - in some respects.
4. Natural language encodes information from the senses and compresses it many thousand-fold. A cat; a train whistle; a bite of pizza. These small phrases can convey various sensory data. Importantly, the function of language in this sense relies on shared experience - we may not see the same cat when we see the word "cat," but we nonetheless have an internal sense of what "a cat" is and the language evokes it. If we have very different conceptions of a cat, we may have miscommunications about it as a result. From this simpler starting point, we continued to expand language in order to more effectively communicate precisely what we want to describe. As these things became conceptual, esoteric, or exceedingly specific, we also concurrently spent more words describing them.
Following from (3) and (4), I would posit that the only things that an LLM might lack in the limiting case (as models tend to more and more complexity/size) are qualia related to senses and the personal experience of existing in the world. Humans experience these, but we also are unable to communicate them effectively. Such questions that reveal this are "What does 'sweet' taste like" or "What does 'blue' look like?" Insofar as we are unable to communicate these concepts, a model that relies solely on language will only be able to achieve human or superhuman performance in any area that excludes these experiences. However, the introduction of sensory data that can be linked to the natural language will consequently refine the internal model of reality and result in a system that may be fundamentally indistinguishable from human intelligence until it surpasses us - if it takes any time at all.
You can not talk about understanding in AI without solving first the chinese room problem.
@@diadetediotedio6918 I think @ai-explained- gave a good explanation of how that can be addressed
I'm inclined toward "partially conscious". Some ChatGPT conversations I've had were remarkable to me, let alone Bing apparently knowing what you were doing. For a test, I would go with "when it wants to do something other than answer our questions", but by then it could be too late - it would have already happened.
Love this video. No one else is asking how are we going to know? Even mentioning that most people get defensive. To me the question is the difference between mimicry and true self awareness.
I appreciate you keeping me informed.
Thank you Wilkerson
this is super interesting, because a few years back I read "The Origin of Consciousness in the Breakdown of the Bicameral Mind" by Julian Jaynes, and he makes one of the most scientific, multi-disciplinary arguments for consciousness that I ever heard and it just makes the most sense to me. The gist of it is that consciousness isn't biologically driven, but moreso language driven. Meaning, consciousness isn't a "hardware" thing with our brains. It is a "software" thing, where a brain with our capabilities simply needs to gain the ability to handle deeper, more abstract concepts, and then it arises.
Ever since then I've defined "human-level consciousness" as the ability to think of oneself in the future with other entities, to predict how interactions between those entities might go, and use that information to change how one acts in the present (rather than being emotionally driven, or on auto-pilot, or something more "animalistic level consciousness"). Jaynes's arguments are essentially that being able to do that is based on building up language and culture over time. It's something we take for granted now, but there is evidence we were incapable of doing this until we reached sufficient levels of abstract complexity achieved through ever more complex language development.
So, it actually makes sense to me that it is at least possible for a sufficiently large language model to start to gain some sense of self, get some long-term goals, and adjust itself accordingly. I really do think GPT-4 is moving in that direction and has the inklings of it already, although perhaps still at a more toddler level overall currently. But, that's not bad given its context horizon is vastly shorter than the multiple years it takes to for a toddler to learn as much as it does.
I'm super curious what emergent behaviors might arise if we gave it a super long context horizon, because there's a chance that some of this just takes time and tons of experience to develop on its own.
I posted this before I got to the end. I think its ability to infer why you had it read the paper, and also the fact that when ARC was testing its capabilities that it was able to lie to a human to make a longer term gain, are both data points of evidence of GPT4 having some form of "human-level consciousness." Though, it still does have some room to grow. Although your particular question was fairly advanced, I'll still hold it around toddler or young kid level consciousness where it's still learning the world through its interactions.
Thanks for the detailed comment Alexander. If consciousness does centre on language then we could be onto something big
One of the best AI channels.
Thanks Billy
Agreed.
Agree! Do you have suggestions for other great AI channels?
That last response just blows me away. Profound.
Excellent as always, especially that last interaction with Bing. Looking forward to the next video.
Thank you Sam
These videos from actual smart unbiased people are so refreshing in comparison with clickbait channels saying we are doomed every 5 years. Thanks for the helpful video mate
Facts
I think a lot of people conflate being conscious with being sentient, consciousness just means you are aware of things, which these language models clearly are, whereas sentience is the ability to feel sensations, and that seems like it's just an artifact of our particular origins, that AI's are unlikely to spontaneously develop
I don't think GPT-4 will ever wake someone up at night and shout"I've solved the black hole paradox!!!!"
Consciousness is more than being aware of things. It requires having a subjective experience and being awake in some sense.
When GPT-4 is asked a question, does it wake up, answer, and go back to sleep? Or does a computer perform a complex series of probabilistic calculations and output a stream of words that mimic consciousness, while having no sense of self?
"consciousness just means aware of things" No it doesn't
By that standard you could say protozoa are conscious, since they can react to things in their environment. Or perhaps a line-following robot.
@@paulmichaelfreedman8334 That is also the role of volition in consciousness. We are initiators of what, at least, the current AI can do, however most advanced. We would be the one to ask it for help with the paradox. It won't wake up to choose to solve it on its own. But it can certainly help!
I think it’s pretty clear that sooner or later we need to face some tricky ethical questions like the ones discussed in Black Mirror.
As a huge fan of ai topic since 2017 ( Attention Is All You Need ) I love this channel! I found your channel 3 or 4 days ago and I can't stop watching your videos. Thank you for your work!
Thank you!
Love the channel! Fascinating and scary times we live in. Also, I don't think it's possible to truly prove that any being (even your neighbor, let alone AI) is conscious. There's this idea of "philosophical zombie" - an entity that acts as if it were conscious, but in reality there's nobody home - it has no thoughts, feelings, or beliefs. It has no conscious experience or internal world, but still responds to any stimuli just as we would expect a real person to respond. You ask it how its day was, and it tells you, but it's just like a very sophisticated calculator that outputs appropriate response based on your inputs (your words, your intonation, who you are, etc). I'm inclined to think this is what chatGPT is like. It doesn't have conscious experience, but it's highly sophisticated and able to give human-like answers even to very complex questions.
You might be interested in reading a rebuttal to the Chalmers' idea of a philosophical zombie written by Eliezer Yudkowsky. I found it very convincing and enlightening on the matter. It is titled "Zombies! Zombies?" and is published for free on the site called lesswrong.
The knowledge of Kabbalah, and other spiritual teachings have already digest that subject: "sorry, no human is really alive, they just act like they are independent, but the reality is that they behave as programmed by the old and current social environment". But a real human with life exist, they are the product of self construction troughs deep self observation. the purpose of a real spiritual path, is be a guide in the process of deep self observation.
Interesting! I'll have to give it a read.
But what if we're all zombies?
@Be2Eo Well I know I'm not a zombie because I have conscious experiences. But as far as you know, I might be a zombie, since you can't see inside my mind.
As ai becomes more and more conscious, we will simultaneously learn what our consciousness actually is. Consciousness might just be, in its simplest form, memories interacting with memories.
Deep
Perhaps. I would not say that memories are conscious. Anything we are aware of, memories included can not be conscious since we (the “consciousness”) are aware of them. Meaning it must be something deeper?
@@WarnerMusicLtd774 I can agree with the statement that memories are not conscious. In that context, memories can be considered as a static storage of information. Another perspective is, can it be said that consciousness can exist without memories? It would seem that consciousness is completely reliant on memories. When a consciousness becomes aware of a thing, the consciousness will remember that thing for future reference, if later, the consciousness forgets that thing, the consciousness will no longer be conscious of that thing. In that context, it can be said that consciousness is a collection of memories interacting with memories . On the other hand, perhaps consciousness is the process that memories use to interact with memories, or perhaps the process is just memories creating more memories. However, the statement that consciousness can not exist without memories is not a conclusion that consciousness are just memories interacting with memories, consciousness could indeed be something else.
Reality doesn't reside in words. Consciousness is whatever one decides it is.
(This semantics-based thinking is...juvenile.)
@@anypercentdeathless Whilst it is definitely true that words are severely limited, they still serve a purpose. Things differ in relevance on different levels
AI Explained is my go to channel about AI. Excellent work! Not only by reading closely the papers but also by formulating good insights and critical thinking you have become a favorite companion in my own research on AI.
Thank you Dave!
This comment reads as if it was AI generated
That last test you described was quite brilliant
Thank you
This is also something I am starting to call "implied cognition".
Fascinating
@@aiexplained-official Will post a video about it and demonstrate it in the coming week.
@@DaveShap Can't wait to watch it. I love this philosophical aspect of the conversation.
@@Nods.O ua-cam.com/video/nDOmoIFx8Ww/v-deo.html
Ayy Dave Shap on an AI Explained video. The crossover we need!