AGI Debate
Ғылым және технология
AI DEBATE 3 : The AGI Debate
The Pivotal Discussion in Shaping the Path of AGI's Global Discourse.
Five panels of the world's most distinguished researchers and experts on AGI :
Panel 1: Cognition and Neuroscience
Panel 2: Common Sense
Panel 3: Architecture
Panel 4: Ethics and morality
Panel 5: AI: Policy and Net Contribution
Speakers: Erik Brynjolfsson, Yejin Choi, Noam Chomsky, Jeff Clune, David Ferrucci, Artur d'Avila Garcez, Michelle Rempel Garner, Dileep George, Ben Goertzel, Sara Hooker, Anja Kaspersen, Konrad Kording, Kai-Fu Lee, Francesca Rossi, Jürgen Schmidhuber and Angela Sheffield.
Moderator and co-organizer (with Vincent Boucher) : Gary Marcus. Gary Marcus is a leading voice in artificial intelligence. Scientist, best-selling author, and entrepreneur, he was Founder and C.E.O. of Geometric Intelligence, a machine-learning company acquired by Uber in 2016. His most recent book, Rebooting AI, co-authored with Ernest Davis, is one of Forbes’s 7 Must Read Books in AI.
Vincent Boucher is President of Montreal.AI and Quebec.AI.
Website: agidebate.com/
#AGIDebate
Пікірлер: 135
Website: agidebate.com/
@dashaledinskaya
8 ай бұрын
Ш тщщ😅щ😅щт😅лш
Thank you for having Professor Chomsky speak on GPT3! What an encounter
@MontrealAI
Жыл бұрын
Your gratitude is greatly appreciated. It was indeed a historic privilege to participate in such a noteworthy encounter.
Panels 0) Intro from Gary 04:46 1) Cognition and Neuroscience 14:45 2) Common Sense 48:06 3) Architecture 01:28:07 pre 5) AI: Policy and Net Contributions (Rempel Garner) 02:18:11 4) Ethics and morality 02:23:53 5) AI: Policy and Net Contributions 03:03:08
@gmaulpker4765
Жыл бұрын
Wow someone with a brain.
This was great, very fast and content packed. I’m going to have to listen on slow speed and make a lot notes. It was hard to reflect on the amazing points being made in real-time 😊
@user-zf2ip4ck8p
Жыл бұрын
ㄢㄢㄦ
I think probably one of the most important points in this conference was the one made by Gary Marcus when referring to Terry Winograd's SHRDLU, where the AI made certain assumptions, but phrased those assumptions in a way where it explained those assumptions to the person providing the instruction. The point being, that an AI needs to be able to provide a way to address the scope of the instructions and questions it receives and be able to communicate with the person / agent that is providing the information. This is why we see a lot of failures by AI systems to properly interpret statements and questions because many of these so called "One-Shot" questions are basically flawed by this concept that we need to have "one" answer that is most correct. That's completely wrong. The answer depends on many parameters that are not actually included in the initial exchange of information. The whole point of communication is to flesh out the scope of the conversation to determine which parameters need to be addressed and which do not.
Angela Sheffield had some very polished remarks, I love the idea of a codex that highlights approaches that compliment particular use-case. ❤
What an amazing watch. Thank you all so much for having this conversation
Thanks for doing this I missed presentation but will watch it.
@MontrealAI
Жыл бұрын
Thank you greatly for the esteemed remark.
@jpber1
Жыл бұрын
Competivity is a good goal. Employees should be involved in unions, governance and social impacts. They should help those who need it most and particilarily from developping countries and neibourghoods. Aim for excellecence and quality. Be generous and fair it will be better for everybody.
Thanks for organizing this debate and to all the speakers for their opinions and presentations.
@MontrealAI
Жыл бұрын
Your appreciation for the organization of the AGI Debate and the insights provided by the speakers is greatly appreciated. Thank you for your kind words.
@l4m1s
Жыл бұрын
@@MontrealAI I do appreciate Chomsky’s intervention indeed. I am not sure about the ethicality though.
With videos this long we really need an ability to like parts of it, with the AI that cuts them into chapters you really should have the ability to like a chapter on top of the overall video 😊
@MontrealAI
Жыл бұрын
Your suggestion is excellent! We will steadily improve the experience for this video. At present, it is quite fresh and not all tools seem to be available.
It's a pretty big presumption to 1) assume that by coming together you can harness the most care and diligence 2) the utmost care and diligence is sufficient to create a technology for the betterment of mankind. I don't think there's a strong historic precedent for the first claim. I think there's a lot of theoretical reasons to be wary of the second claim. I think the only thing you can claim is that by coming together and having a candid discussion that leads to actual actionable points, you might be able to divert disaster and improve the outcome
@MontrealAI
Жыл бұрын
With all due respect, 1) and 2) are not what I said. Here’s my original statement: "By coming together as a global community and approaching the AGI debate with the utmost care and diligence, we can harness the full potential of this technology for the betterment of humanity." My sincerest apologies for not addressing the issue of mitigating malicious AGIs in my previous discourse. My desire was to allocate the maximum amount of time to our esteemed speakers. I wholeheartedly concur that the prevention of AGI-induced disaster is a vital and pressing matter that remains as yet unsolved. Vincent Boucher, President of Montreal.AI and Quebec.AI
@michalchik
Жыл бұрын
@@MontrealAI I apologize if I misrepresented what you said. I'll have to go back over the original statement and I think you're intent was good. I think Gathering like this certainly won't hurt unless it generates an unrealistic level of security. I don't have time to watch the full program right now but I will make a point of doing it within the next few days. Good luck
- EDIT - I see the KZread auto CC for this video has now been activated. Thank you 🙌🏻 Thank you recording and sharing this! Will KZread's automatic caption be activated for this (video dashboard > video's setting > English) so an .srt will be available or will we only get the one encoded into the video? Thank you again.
@MontrealAI
Жыл бұрын
We are evaluating automatic caption options to determine the most suitable solution.
@treflatface
Жыл бұрын
@@MontrealAI Grand, thank you. The quality of KZread's auto-caption has been great IMO.
This video is best experienced at 1.5x speed. Seldom do I listen to accelerated dialogue, but this one sounds completely natural when sped up and helps comprehension a lot for those of us with poor short term memory or ADHD.
Я просто уснул с включённым ютубом,Боже о чем идёт речь …
@user-zz7fl4bh1m
10 ай бұрын
я тоже, ахаххахаха
@milkkissa6620
9 ай бұрын
+, шо происходит...
There are many points here that deserve their own series of discussions with the world's leading experts, and with anyone with an interesting and important viewpoint. It is unfortunate to cut great points and topics short.
проснулся и увидел на экране это:
In a nutshell, Marcus: This is far from AGI and here are some hilarious examples, Schmidhuber: I have invented this stuff 4 decades ago, whatuallabout!? Chomsky: It does not help humanity, as I keep telling you all 🤷♂️
@ryanbigguy
Жыл бұрын
A trio of clowns.
AGI debate, 5 minutes in, technical difficulties. Seems promising.
Sadly, it feels like there isn't a proper "debate" anymore since the first one. Where are the true counter-positions? Why these short presentations instead of having two speakers argue with each other over a longer period of time?
Everyone should watch Gary Marcus vs yashua benjio debate , he's talking like he's got the whole AGI thing figured out
I think (or hope) what Yejin was trying to say with the "don't kill all trees" point was that we need to figure out a way to impart ecosystem science and ethics to a prospective AGI - the sense of a fragile whole comprised of millions of parts where every part is highly valued both for its role in the integrity of the whole as well as an evolutionary inheritance that took millions of years to create. It is a bit concerning that all the ethicists on this panel are so human-centered and the word ecosystem never came up once.
@RubanWood
8 ай бұрын
😊
Why not use GPT-4? In many cases it's getting better
@MontrealAI
11 ай бұрын
The AGI Debate was live on Dec. 23, 2022.
Comments I first thought were a negative rejection did upon reflection strike me as positive instructive precautions.
What a fantastic deus! Merci bcp pour tous vos travaux Montréal AI 👌
Personally, I'm surprised there hasn't been more talks between cognitivists with developmental biologists and materials scientists/engineers on the nature of embodied wetware systems when approaching more general/broader intelligence, especially, when discussing modular architectures. What are we afraid of, Roku's Basilisk? Perhaps I'm just pathologically daft.... Regardless, wonderful panel, Great Galaxy bless you guys for having this open to the public (much love from a person who may never aspire to being anything more than a sophisticated layperson).
@maloxi1472
Жыл бұрын
"Perhaps I'm just pathologically daft" you don't say...
@josefgross4432
Жыл бұрын
@@maloxi1472 but I did say....
Humans have the ability to compose, and teach those compositions. With composing we can bring things together in our minds, and then test them in reality. AGI - Intelligence is the ability to compose and then communicate that composition to others. Machines can compose. Static Diffusion and ChatGPT are evidence of this. But the model that they compose with is typically read only, unless in training (typically then offline). I think humans are a little dynamic during the day. But during sleep we grow, and then test our minds with dreams. The machines also do this, but during a long, drawn out, training stage. We have already achieved AGI. But we haven't embodied it. It also has no concept of its own reward. It's specific capabilities are not generally integrated (ex: ChatGPT and Static Diffusion, 2 different models, running on different machines). While machines can test, they can't determine the reward in the same dynamic way as say ChatGPT and StaticDiffusion can output different things. Could be, also, that biology also does the development of composition skills in a more time efficient way, than our machines, and better algorithms could still win over on current algorithms. But, we could always train an AI to use language as a way of describing the training/reward process, and allow it to continue the experiments, give it all the GPUs it asks for and optimize on ((GPU efficiency) + Output). Humans are great at composing. Language rules be damned. We compose words. We compose dinner. We compose the strategy to hunt for food. We compose art of course, but only because composition was so important to getting food, shelter, clothes, armaments, implements first. Our artistic composition is just a sign of the ability to do all the others well (natural selection, mating selection). English teaches are a wonderful thing, but they are basically vacuum tubes. What we are seeing is that neural nets don't use rules, yet speak clearly (if sometime getting the subject wrong). As do children. Children also draw structure, and learn to separate it from style. No other animals is as good at drawing, independently and communicately-compositionally, as humans at the age of 5 years.
Arguments based on intuition never resolve the pain, only extend it. Arguments need logic, math and a hard science. This only divides us.
Is that the real Noam Chomsky? Whoa
@MontrealAI
11 ай бұрын
Yes, we were so fortunate to have the real Noam Chomsky speaking at the AGI Debate.
я уснула за компьютером и проснулась за этим видео, где я
This “debate” seemed a little skewed in one direction. But ok, good points made.
In the beginning the awkward silence is super cringe.
bruh really Jeff Clune a mask on during a zoom chat LOL get outa here with that shhh lol yall killin me 😂 😂
lâu lâu chán xem ad chơi giải trí thật
security and regulations on a standard platform would be ideal -(unlikely) alas...
@MontrealAI
Жыл бұрын
Nevertheless, it is imperative that we continue to aspire towards this goal.
I wonder why that they make such a point of trying to trick ChatGPT with simple questions in the first hour. I've had a lot of conversation and planning with it, and clearly it does make mistakes, but it is very good at very complicated conversation and planning when the prompts are engineered well. I'd love to share some of my experiences with it. Maybe I will make a video series about it. Actually I just realized that I am on my wrong KZread Channel for this content, I will switch to my other channel Aura of Intelligence.
DiLeep George came closest re commonsense knowledge and language. Unfortunately, he is likely totally unaware of the absolute impossibility of implementing his scheme in the current computational framework that is underlying AI.
Я просто заснул под видео на ютуб, а проснулсяс этим
굿!!
Gary Marcus just doesn't realize yet that logic is a composition (one of many), it isn't a natural ability that humans have. Composition is the key skill that humans have. Even to the point, where magic is incorporated into our thinking. Young humans are in fact more easily fooled by magical thinking than our closest primates (and this was tested). Okay, so machines can do composition now. We have achieved AGI, and this debate is really over. What we need to discuss now is integration of compositional subjects, optimization (so that it model improvement can be in done realtime). Finally, there is a consolation prize in this. The way that we develop these models, incorporates a huge amount of mathematical logic. So yes, in a way, Marcus isn't wrong, logic plays a role in AGI. But it isn't only formal logic that was needed. We thought for a long time we could understand language with some finite set of rules. Turns out, language slips out from under the rules. The best decription for what natural language is, is it a description model for internal compositions that is also used to measure fealty, locality, awareness... and huge number of other (compositions), relevant to the minds of co-habitators, strangers, and enemies.
I do not really understand what can be meant by "aligning AI's ethics/values with our own". Just whose values are we talking about? Suppose you start installing AI driven technologies in slaughterhouses to make killing animals more "efficient". Will vegans be listened to when they object to such a use? Or how about the Dept. of Defense? Can they use AI to do better whatever it is that they do?
@user-nc8kj4dc2d
11 ай бұрын
😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😅😊😅😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😮😊😅
@user-nc8kj4dc2d
11 ай бұрын
😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😅😊😅😊😊😊😊😊😊😊😊😅😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😮😊😅
they need to comback and review gtp4
LLMs are one piece of a much larger research effort to build AGI. The larger community is and has been addressing all of Gary's complaints since before he thought of them. Nobody is claiming LLMs by themselves are a solution, and Gary's incessant whining about their obvious limitations is getting annoying. We know already. We've always known. Go build something that works better. That's what the rest of us are trying to do. To say LLMs have made no scientific contributions is just willful ignorance
@blackenedblue5401
Жыл бұрын
Can you give an example?
@glassrocketstair
Жыл бұрын
@@blackenedblue5401 maybe? An example of what?
@blackenedblue5401
Жыл бұрын
@@glassrocketstair ah yes, like an example of the larger community addressing one of Gary's complaints before he'd thought of it
@glassrocketstair
Жыл бұрын
@@blackenedblue5401 i guess we have to substitute "when he thought of them" with "when he presented them"... But take for instance his blog post "Noam Chomsky and GPT-3" where he says "GPT-3 is a model of how words relate to one another, not a model of how language might relate to the perceived world"... Well ya: that's exactly how the training algorithm works. It has never perceived the world, so why is he surprised? Why does he think this is an important criticism? It's like criticizing a calculator for not truly "knowing math".
@blackenedblue5401
Жыл бұрын
@@glassrocketstair @Qetion I understand your point. Marcus does strike me as hyperbolic and overly confrontational in public, the way he's always calling out Sam Altman on Twitter. Chomsky made a good point though by calling for more AI research drawing on cognitive science. I also liked the views of the guy from Stanford AI lab and of that VC investor. That last woman was the only person with nothing to say really
❤🧡❤🧡❤
How is this a debate? It's clearly no long term?
IF THEY PROGRAM AI LIKE THIS MODERATER CONTROLS HIS SOFTWARE WE ARE DOOMED
One thing one can definitely learn from this panel is that good old-fashioned AI pundits don't understand LLMs and their place in reasoning. These pundits don't begin to understand how an LLM interacts with an agent and how agents are necessary when it comes to reasoning. This is the case despite the fact that these pundits are the agents when they use an LLM. They can test the results, etc. The question here isn't if LLMs can do everything. The question is can an agent reason without a generative model. I think it's clear that we agents have internal language models. We couldn't think without them. We wouldn't come up with ideas to discard or embrace. Humans have something analogous to a critic and a policy network. We can build these and give an agent access to them. If, of course, one believes that AI is wrongheaded, it's unlikely that one bothers to know about these kinds of things.
good player
Chomsky. Genius ❤
@johnstifter
Жыл бұрын
nothing he says about AI is accurate at all
@ryanbigguy
Жыл бұрын
He's an ridiculous old man who needs to stop speaking publicly
Be careful with metacognitive architectures. This is how we'll get artificial knee-jerk reactions. It's a worthy idea, but it has dangers.
i like your game
i like it
A.i and biology need to be merged
@peterbelanger4094
Жыл бұрын
You monster. Cybernetics is abhorrent, biology and machine should NEVER mix. Transhumanism is pure evil.
@MontrealAI
Жыл бұрын
The merging of artificial intelligence and biology is a prospect that demands the utmost care and diligence and calls for careful consideration and rigorous examination.
AGI Debate
AIX runs it all now
I asked ChatGPT what it thought of this Video and this was its response. "There are some AI researchers who are unhappy about ChatGPT, which they see as a significant step towards creating artificial general intelligence (AGI). Despite this, ChatGPT is still a valuable and important achievement in the field of AI, and it has the potential to serve as a foundation for future developments in symbolic AI. Some modifications to deep learning techniques may also lead to further progress in the field." I think it was just being nice.. ChatGPT is the first AGI, I can't wait to see future iterations.
@anonymously94
Жыл бұрын
You don't know what AGI is.
@TheReferrer72
Жыл бұрын
@@anonymously94 I know ChatGPT is the second strong AGI made availble to the public, the first was Google's search Algorithm. Its fun using ChatGPT, smarter than any single human on the planet at manipulating symbols. Its breath is outstounding, depth not so but that will improve with time. Can't wait for GPT5
@dennismertens990
10 ай бұрын
@@TheReferrer72 By classifying Google's search algorithm as a form of general intelligence, it shows you have the wrong concept of "general" intelligence. By saying that ChatGPT is smarter than any single human on the planet, it shows you have a very low opinion of human intelligence.
@TheReferrer72
10 ай бұрын
@@dennismertens990 But I have a high opinion of human society as that what made these tools possible.
Self-righteous ramblings
Most of them are already talking like robots
What was this event about? You haven't seen anything more insignificant than this for a long time...
@MontrealAI
Жыл бұрын
Thank you for the gracious words. The future of artificial general intelligence holds both boundless promise and inestimable uncertainty. The potential for AGI to revolutionize our world and unlock hitherto unimagined levels of human potential is rivaled only by the necessity for scrupulous contemplation of the ethical and moral ramifications of such formidable technology. Therefore, we convened on December 23 to participate in a discourse of singular importance: The AGI Debate.
@silberlinie
Жыл бұрын
@@MontrealAI It gives the impression that no one really wanted to do this presentation. As if what had been said a thousand times had to be said again. So that one fulfils one's obligation.
@MontrealAI
Жыл бұрын
@@silberlinie Our esteemed panelists were delighted to be in attendance at this pivotal discussion, where a nuanced examination of history, best practices, and recent research was conducted in order to shape the global discourse on AGI. Some even lingered with us until the wee hours of the night, a testament to their enthusiasm!
@silberlinie
Жыл бұрын
@@MontrealAI I admire your staunch defence of the superfluous event. The music is playing elsewhere today. The old train has sailed off into the realms of the past.
they should have invited me to this, i bet nobody in here even has a cognitive humanoid ai..... gofi posers
@MontrealAI
Жыл бұрын
Your disappointment at not being invited to this event is understood.
The complaint at 2:35:00 that the effective altruism movement is gaslighting people, without any specific evidence of them making a false claim, is itself gaslighting.
LOL you cant stop A.i
Has anyone of you asked themsleves why we need AGI in the first place ... except for money obviously. This is really funny seeing hords of programmers racing towards general intelligence which - if it finally emerges - will put them out of the business :)
@daniellawson9894
Жыл бұрын
that’s the point. If there’s AGI, then the concept of business doesn’t matter. If things can be efficiently produced without human inteligence, then scarcity can be eliminated. AGI means solving many of humanities problems, and giving humans more time to solve those that it cannot.
@cynamonstudio
Жыл бұрын
@@daniellawson9894 Utopian revolution - yeah, right.
norm
government makes everything worse- always.
Who hell thought inviting Chomsky was a good fit for this.
@johnstifter
Жыл бұрын
@queerdo nothing he says about AI is accurate at all, it is embarrassing
Choam NOMsky 😋🍪🍪🍪
i like you
omg a sheeeple, sitting alone, in a mask, is so hard to take seriously
@paxdriver
Жыл бұрын
Lol depends, maybe he's locked down in a quarantine zone 😂
You are thinking of ethics from the wrong perspective, there is a method to do so easily, you are just not approaching it from the correct point of origin, however, we all know that we are not going to give our AI any ethics, it is just a complete waste of time to cover such a topic, nobody here thinks that you will give it the ability to reason on it's own, so stop it right there. We give soldiers a directive to kill and we expect them to ignore their ethics, and so we are expected to believe that you want man kind's, "Digital Child", to have thoughtful considerations; I think not!
What a rip off! This event was delayed like ten times. And in the end I could not even say a single thing! This debate was a waste of time. Nobody here do something significant for humanity. This is hubris. Must be nice to stay in an ivory tower
Im prety sure a spider or a snake understands more. Simple precise KNifes out fools ... I will show you hahaha
that one chick is kinda hot
Gary Marcus just sounds insufferable.
Gary should just stop whining and study AI before speaking.
good player
i like your game