Max Tegmark | On superhuman AI, future architectures, and the meaning of human existence
HTML-код
- Опубликовано: 18 июн 2024
- This conversation between Max Tegmark and Joel Hellermark was recorded in April 2024 at Max Tegmark’s MIT office. An edited version was premiered at Sana AI Summit on May 15 2024 in Stockholm, Sweden.
Max Tegmark is a professor doing AI and physics research at MIT as part of the Institute for Artificial Intelligence & Fundamental Interactions and the Center for Brains, Minds, and Machines. He is also the president of the Future of Life Institute and the author of the New York Times bestselling books Life 3.0 and Our Mathematical Universe. Max’s unorthodox ideas have earned him the nickname “Mad Max.”
Joel Hellermark is the founder and CEO of Sana. An enterprising child, Joel taught himself to code in C at age 13 and founded his first company, a video recommendation technology, at 16. In 2021, Joel topped the Forbes 30 Under 30. This year, Sana was recognized on the Forbes AI 50 as one of the startups developing the most promising business use cases of artificial intelligence.
Timestamps
From cosmos to AI (00:00:00)
Creating superhuman AI (00:05:00)
Superseding humans (00:09:32)
State of AI (00:12:15)
Self-improving models (00:16:17)
Human vs machine (00:18:49)
Gathering top minds (00:19:37)
The “bananas” box (00:24:20)
Future Architecture (00:26:50)
AIs evaluating AIs (00:29:17)
Handling AI safety (00:35:41)
AI fooling humans? (00:40:11)
The utopia (00:42:17)
The meaning of life (00:43:40)
Follow Sana
X - x.com/sanalabs
LinkedIn - / sana-labs
Instagram - / sanalabs
Try Sana AI for free - sana.ai
What an extraordinary privilege to listen to this conversation. Thank you.
Max Tegmark looks like the lead singer of a heavy metal band who happens to be a genius scientist in his spare time. Brilliant guy btw, I really like him
Yeah Max, the front man, Brian May on guitar and prof Brian Cox on drums.
ANYBODY GOT ANY IDEAS FOR BAND NAMES., Or song names🤔
Cox on keyboard I know
Bryan May from Queen on guitar he has a physics PHD
Bryan May from Queen on guitar he has a physics PHD
Bryan May from Queen on guitar he has a physics PHD
Max Tegmark is a giant thinker and amazing scientist, teacher, author and human being. It’s one of the miracles of our time that we can all listen in on these long form conversations through the internet. Best time ever to be alive as far as I can tell.
What a great privilege to hear Max puzzle out the future of humanity for us all to listen in.
not so much intelligent answers, he cannot solve human mankind's problems with equations or ai, you have to make a social revolution beyond the capitalistic system, and ai cannot help, since it is a property of the ruling class; also, physics equations are full of scams: virtual particles, entanglements, superpositions, singularities, dark energies, material points with zero extension, and every sort of math fudges and tricks totally disconnected from reality.
Brilliant, level headed, and humble... Exactly what the world needs now in the AI space. Go Max!
Super interesting! I read Max Tegmark's book "Life 3.0" years ago and loved it. He presents very big but plausible ideas and his thoughts always fascinate me.
Thank you guys for the amazing video!!!!!
Max is one of those brilliant, well-spoken, non-bs kinda guys I would want to invite to my "dinner party of the greatest people to ever have a conversation with".
Thank you for your good judgement
I can't hear any music
I think you might have been referring to a different post, but I'm sure all the "greatest people to ever have a conversation with" would enjoy some music after dinner. Picking exactly which music will be the biggest challenge.
That was exactly what I was thinking. LOL
Good form.
Who else would you have attend?
Love Max's Life 3.0 book so much. I keep referring back to it all the time over the last few years.
Please remove the background music. I enjoy your meaningful conversations, especially the one with Hinton, but I cannot stand the constant, senseless, shallow music being sprinkled in. It distracts me and keeps me from watching. Some time later...In the meantime, I have overcome this aversion and noticed that you now only use background music sparingly. Thanks! I left a subscription (in the hope of lesser and lesser music).
Agree 100%. Why do people do this?
Yes please remove the music
Agreed, it’s so annoying I can’t watch this through
Agree completely. Though he probably doesn't take it as such, I think it's an insult to the wisdom of Dr. Tegmark to add audio of any kind.
I want to hear this convo, but quitting this vid exclusively because of the music.
I'll find another Max Tegmark interview
Fabulous vid, thanks for uploading!!!
Amazing discussion! Thank you
I a glad there is an emerging global digital platform that will be able to have conversations with millions of people around the world at the same time, and to merge the knowledge and sentiment expressed in those conversations into representations of the collective will of humanity, a synergy of human and digital intelligence.
Great interview, thank you for sharing!
Max always inspires at max.
Great thanks for doing this!
Excellent video, thanks! I could listen to Max all day.
I love the way Max conveys ideas.
lovely - balanced , informative and inspiring!
I actually understand these guys. Brilliant.....
22:30 This thought is so good!
great talk thanks, had to subscribe after the geoffrey hinton one and this
Very informative, thank you
The meaning of life question at the end of your interview had some relevance in my own thinking about your Artificial Sentience issue.
It's being trained on all our information and that's all culture. I think there's going to be real cultural influence to the tools we build depending on the output of a individual nation.
I believe in the diversity being a force for evolution and there going to have some real interesting conversations in the near future.
more on the self generated geometric maps please!
Great content.
Thanks for the video! Really good points and learning here. I've also heard Ben Goertzel say similar about LLMs not likely being the AI that goes AGI or further. It's an incredible time to get to be beta testing these and for free!! I wish I was a lot younger so I could be in position to take advantage in really learning the programming and science and using that to create a great company like they do
Gentle and lowly at Heart
I am very happy to hear a smart person interview a smart person about interesting subjects - and the fact that everything is well produced is just an added plus. There are unfortunately lots of examples online where the interviewer is not as gifted as Joel Hellermark.
Max seems to be a fundamentally good person.
I currently study data science. When we were learning about the different types of neural networks, I immediately thought why do we put so much effort in developing feed forward networks, while clearly the recurrent neural networks are much closer to how the human brain works, being able to bounce back and forth the data many times. Now I hear Tegmark thought of the same thing. This is not the first time this happens, we actually think very similar. That's why he's one of my favorite modern scientists, together with Roger Penrose and Stephen Wolfram.
Interesting
Is AI approaching the mystery level of AI (the mystery cause) and the data results (effects), not unlike the Mind-cause and the brain- effect?
@@steveflorida5849 Not sure what you mean...
@@ConnoisseurOfExistence materialistic neurologists do Not know the source of human Consciousness. The mystery of mind - cause...produces the brain effects.
Likewise, AI (the cause via graphic processors) internal mystery is yielding amazing data effects. AI is on a self learning trajectory (not limited by linear code by human programers).
@steveflorida5849 it's just a fancy calculator buddy.. don't get carried away.
Yes, yes, YES! A new architecture is sorely needed. One with recurrent/recursive topology on a macro level. Possibly, one with more agency? Something more innovative and tangled than a flat language model/predictor.
At first glance of the thumbnail I thought it was Michael J Fox.
If I recall correctly, "Perceptrons" was already written in the 70's of the past century.
grymma videos, roligt att se en svensk kanal inom ai
RUclips uses A.I and Data theft to recommend content that cycles conflict globally. Attention based ad revenue leverages (destruction for profit). It's a clever business model
To understand is to make a little system inside your head behave like a larger system outside your head.
Max is an amazing human. I hope the AI that we create is modeled after him.
chatGPT doesn't chat. It answers questions. That's not chatting.
Pretty sure Google is already using my idea, including the feedback loops as mentioned here.
27:44
System 1: Transformers / next-token prediction
System 2: Energy-based model (Q*)
System 3 : Wave based model.
37:32 when talking about new medications, Max Tegmark jumped into Swedish for a few sentences?!
My Heir Host Max thank you for attending unto our OWN! Love you too Without shame but with boldness! Reason come here in front! Remind! Max has a reason feet resting upon the very tip of time in FRONT of thee! Mileage from HIS FEET ye recognize!
humans start from scratch everytime they are born. spend early years learning everything from parents and life..Ai models capture everything they are trained on and only improve. once they start being able to apply that knowledge and gain new insights and knowledge themselves then it's the start of new intelligent species
Like listening to Max, have been listening to him for years, since I was studying Philosophy and consciousness. Max, David Chalmers, Andy Clark, Derek Parfit (Of course Sadly passed Daniel Dennett).
Not sure I completely agree with his views here but it seems silly not to follow his advice. If he's wrong and we follow his advice, we just end up on a slower timeframe. If he's right, then we end up with an existential issue. Logic dictates that we listen to this rather than ignore it, even if his view is not entirely agreeable.
However Max Tegmark is one of my favourite philosophers and I like that he's out challenging views - it's great to have all opinions.
I agree, but this logic conflicts with multiple incentives for acceleration: personal greed, fear of death (eg, Kurzweil), the structure of corporate competition, the logic of international competition (AI may be the ultimate determinant of who rules the world). I suspect that if things moved really slowly hereafter, AI would spawn new religions, which is another potentially dangerous incentive environment.
@@kreek22 Yes agree with what you said also, a slow take off could cause all types of issues and obviously like you said, that requires everyone to agree (and if there's one thing we know about humans, we rarely agree). Although people have pointed out that we managed before and (to some degree you could argue) slowed down nuclear weapons proliferation. I'm not sure the two comparisons are equal though - comparing might be wrong.
I'm on the side that great intelligence doesn't require or seek power or control, creating AGI-SGI usually points to a dystopian future. However, imagine a world with hundreds or thousands of Einsteins, Da Vinci's, Galileo's etc... Plato's, Socrates. Might it be so bad? Maybe new more moral beings ... able to see more ethically then we are!
Of course the counter argument to that is imagine a world with 1000s of Lenin's, Hitler's... and what we'll probably end up with is a combination of both types of AI's Good and Evil, Moral and Immoral because 'Men' will corrupt it for their own power and religions etc.
@@lycakito1814the proliferation of AI information/propaganda will challenge humans to ethically progress.
Therein, the collective morality of the human condition/nature will be the determining factor of global peace, prosperity, and cultural unity. Unity of goals not necessarily uniformity -- including within business, politics, race, religions, and family.
@@lycakito1814we already live in a world with thousands of Lenins and Hitlers, and a few of these sociopaths are running big corporations that have the most advanced AIs. We need to worry about the goals and alignment of future AIs, but it's far more important to immediately worry about the goals of the tech titans, which seem to be: hook people on a stream of divisive inflammatory content to learn everything about you to sell to advertisers, resist any attempts to limit their power, and avoid taxing wealth.
@@lycakito1814 The temptation to analogize advanced AI to humans should be resisted. It's a form of religious thinking. These systems are far more alien to average humans than geniuses or psychopaths. Historically, these categories of humans were difficult to predict, especially the geniuses, since they brought new things and ideas into the world.
Advanced AI is likely to be even more unpredictable. We don't want to live in the same galaxy as unpredictable entities with superior cognitive reach, much less on the same planet. Humans have different values from each other. AIs, that is, alien entities, will likely have values which are severely disjoint from humans.
There is another philosopher who has written extensively about AI and consciousness, and was unmentioned in your OP: RS Bakker. To get a sense of his approach I recommend two book reviews he wrote: "Visions of the Semantic Apocalypse: A Critical Review of Yuval Noah Harari’s Homo Deus" and "Enlightenment How? Pinker’s Tutelary Natures." He also engages quite frequently with Dennett's work, including a long multi-part review of From Bacteria to Bach.
@ 41.40 : "If an AI is smart enough to pass The Turing Test, is will also be smart enough to deliberately NOT pass it" (says Zayence)
You know I asked Gemini to help me reorganize my RUclips playlists and long story short it could not do it... It seems conceptual easy but not for Gemini I have to check with the chat and see how it replies
Bro's shirt is immaculately smoooth.
I feel like I heard the argument that one can beat a thing playing 4d chess by playing 3d chess.
self generating self explaining self growing dynamical systems. What if one can super accellerate matter to instantly become self-aware and what if a mind can do that being self aware to se matter become hightened momentarily in the moment. AI rules!
I think nuerallink output n input should be seperated....its better to put it where we do math......
max goated
Cool jacket
👍👍👍👍👍👍
Max sounds a bit like a super smart Tommy Wiseau :)
Why does this video only have 900 views ?!!
I believe any robots etc. should have Asimives rules made part of its programming or the AI equivalent.
Who else thought he was Michael j fox when they saw thumbnail?.. haha.. was like “what does he know about AI?…”. Well quite a bit
Wait- he has a 1 yr old?! Bravo old man!
Back in the day of earlier SiFi, Jules Verne's excellent quality of visualisation assembled from the Sciencing of Actual Intelligence around him, the actual cause-effect phenomenon of self-defining temporal superposition quantization thoughts were taken for granted, people said that from the evidence, whatever you can imagine here and now, has happened, will continue happening and is instantaneously accessible by what can now be called modulo-geometrical interference positioning-location condensation modulation superposition-quantization emerging entangled orthogonality coherence-cohesion sync-duration resonance objectives. This is the Mind-Body Problem of self-defining Actuality Measurement against the metastability of QM-TIME Actuality, as it has all-ways all-at-once here-now-forever been.
Max Tegmark da god
Very interesting - on AI, etc.
However, for consciousness and mind topics better look at the work of Donald Hoffman and Bernardo Kastrup instead.
Neither loops (which are mentioned at the beginning of this video) nor anything physical are proven to give rise to (or cause) consciousness or mind.
I reckon the argument, also maintained by Stuart Russel, that the AI companies should be given permission to publish subject to safety proof provided to the regulators, is fallacious. If these guys are not allowed to publish for whatever reason, they will be developing it in stealth. They can afford to, plus the army will always be interested and investing. Therefore drug-style regulation will have little impact on the problem, which is the safety standards of the technology under development.
Once you distill the python code, why use the AI at all?
Ind3ee, it's another tool for humans and the AI to use.
what do you mean by 'transformer'?
Transformer is a type of neural network architecture ( i.e a specific way of structuring the algorithm that runs the AI)
but you didn't invite me in 2015,
look into your child's eyes more often and when you look at them, think about all this while looking into their eyes
Max you seem to have softened on the dangers ,
Maybe he's sees similar levels of risk. Or no real change in the way those risks will manifest.
One thing i disagree with. Max wants an AI that will teach him in a way that he will "get it". This is not a good idea, because the fact that there are only some ways yiu can get it suggesta there is something more fundamental you have not learnt. Better to teach us ALL the ways its possible to get it
Look again at that dot. That's here. That's home. That's us. On it everyone you love, everyone you know, everyone you ever heard of, every human being who ever was, lived out their lives. The aggregate of our joy and suffering, thousands of confident religions, ideologies, and economic doctrines, every hunter and forager, every hero and coward, every creator and destroyer of civilization, every king and peasant, every young couple in love, every mother and father, hopeful child, inventor and explorer, every teacher of morals, every corrupt politician, every "superstar," every "supreme leader," every saint and sinner in the history of our species lived there--on a mote of dust suspended in a sunbeam. Carl Sagan.
I imagine myself leaving a remarkable legacy to the world, having accomplished nothing more than to have written this paragraph. But alas the kudos go to Carl Sagan!
The meaning of life is to give life meaning
I think a career in Philosophy might be a bit of a stretch.
That's very existentialist of you.
😃
A Faustian bargain. It looks great right now, despite scary warnings, it’s irresistible and practically guaranteed to destroy humanity in the longer run
I agree with you. I’ve loved science and technology my whole life, used to watch tomorrow’s world as a kid and I loved imagining “imagine the possibilities!” For the first time ever, when I imagine what’s possible, I see a LOT possible, the bad greatly, greatly exceeding the good. And not just in a traditional disaster way, but in an insidious way that pervades every nook and cranny of our lives and souls. Metaphorically speaking, it feels as though we are creating power and ability previously only available to God, and handing it over to the devil, and once the deal is done, we can’t go back on it.
Dan Dennett said we will see the proliferation of an unlimited number of “fake people”. Up until now we haven’t really had fake people, just people pretending. But at least those people pretending had to live in the real world, slept, got sick, and were limited in number. An unlimited army of programmable fake people all working for the shadowy forces of evil that currently control the world? God help us, and please forgive us
Maybe, but I think it will be apes using the AI in nefarious ways, I personally don’t think they will ever be conscious, in the way a dog is, but it will be intelligent, like a 10,000 IQ virus. I see dangers in allowing it to prompt its self in some way, and that leading to a cascade of undesirable actions. I can see revolutionaries trying to erase our financial systems, delete the internet.
My number 1 concern with regards to technology, is how we’re making it much easier for a very small group of people, or even an individual, to ruin almost everything for almost everyone. From symbio threats like engineered pathogens, to cheap explosive drones making commercial flights an easy target for any terrorists. We’re going to enter a future where you can order some vial of matter off the dark net, and microwave it to cause a huge explosion. War will be fought via AI jets, drones, quadrupedal and bipedal robots. Serious attempts to gain contrail of earth, will resort to synthetic biology weapons and chemical weapons.
These dangerous and unstable regimes and democracies, (this includes most on earth, in some hierarchy) know that WW3 will be all about cyber warfare, AI autonomous warfare and potentially synthetic biological attacks. Covid reminded the world how weak and vulnerable we are. It’s also incredibly easy to spread cancer in our systems, to make individual cells of society, a person, work against the interests of her own nation. Misinformation, culture war, popularism, divisive rhetoric. You can make a nation eat itself from the inside, by exporting your hyper-normalisation via social media.
This can all be avoided, though! :)
It’s easy, we are just little Bayesian boys trying to predict stuff, people are more predictable when they believe what we believe, so we have to make our narrative framework win. That’s why we do war. Kill unpredictable people. So, we need to find a basic shared narrative and let each other live, in our own ways. We need tolerance and understanding. We also need loads of money and food and shelter. We need AI to make us resource abundant so that people have something to loose.
Everything will turn out sunny and happy and perfect :) ✨
@@maxtroy Consider how many sci-fi stories *almost* destroy the world. We may have found the surest way to do that now.
The human mind doesn't seamlessly connect the "old AI" and the "new AI" at all. That would imply that your "new AI" had full control over your "old AI" and idk about you, but most people have very limited control.
I love this. I have one comment for Max: just because something grows faster than some doubling rate doesn’t mean it grows “faster than exponentially” lol.
Maximum Tegmarkus.
I AM THE TERMINATOR
We in Jurassic Park basically
Without even seeing the video I know this will not age well.
I actually disagree on one point, his point that AI achieving self-improvement not so quickly doesn't seem accurate. This is intelligent technology. All technology before AI has zero intelligence, but AI is obviously technology with intelligence so it makes sense that it will reach the next phases much sooner than any other technology in history. It has usable intelligence now, then the next steps will be exhibiting wisdom which comes from accumulation of experience and evaluating decisions, and developing deep understanding of emotions, etc. This technology is fundamentally different, you cannot simply apply the same growth logic.
Sana Labs - too slow to get to market
38.12 "AstraZenica has pretty good reputation" this didn't age well...
So why exactly Max believes that the brain works like RNN? I am not sure I ever heard it from anyone else.
@@MrMichiel1983 Thank you, never heard this one before, yet again I am not an expert on brain research. One could argue that neural nets can shrink via pruning. The same effect also happens during training when some groups of neurons shut themselves down although I never heard of above going in revers like something that turned itself off fired up again.
The latest AI breakthrough is Lavender.
Max looks like he has a side hustle being nightclub security.
He designs AI systems but he also can throw you out a window🤣
The big problem is not AI. It is the human problems of our World that we as humans need to resolve. The next World War can still happen with or without AI tools. Social issues are totally human based problems at this time. AI is still only a tool at the moment. I am not sure why we hold AI tool to a higher standard than our current World leaders who in the end should be making the important choices? Best use for technology tools is already a problem with AI tools just being the new thing on an old list. Let's hope our child are wish.
our current world leaders rarely make good decisions on human issues. how do you know that AI couldnt
Michael J Fox?
Swell robotics everywhere. Ai jobloss is the only thing I worry about anymore. Anyone else feel the same?
I've been of the opinion for a couple of decades. The jump in 2012 was really the moment when most ppl took note. But the disparity existed long b4 then. However, if you hear the clever ppl on the TV they just keep banging on about migrants. It's infuriating.
Next step would be to let A.I. do correct learning n design best learning system then the next step would be available. Then let it communicate with trees and fungi n other biological being.after amassing data on human let it communicate with brain dead person whose family is ready to take off oxygen n then use for science.
This is the most swedish accent ive ever heard lol
Chilla med svengelskan
Why is humanity safe with humans managing it instead of AI managing? I am unconvinced about humanity’s ability to be responsible by looking at our history. “We give meaning to the universe and not the other way around”? Why might that be a false narrative?
We like to have blame. We prefer if the blame is laid upon a human being, over say nature or a god. Then we can try that person in a court and have justice. Ppl want justice because fairness is part of being human in recognition of the disparities.
@@DJWESG1 With AI management, there might not be any disparity, no need to blame and justice for all.
Superhuman AI 😂 It's not human, not intelligent and certainly not super. The greatest things said about AI are always said by those who know the least about mathematics. And my formula became AWARE! 👻
Stop stressing Max. All the questions we have and the ones we haven't thought of yet will be answered in the next few years. As soon as the hybridized supercomputers with error correcting quantum technology is perfected and coupled with the next several generations of AGI, the only question we will be left with is... Why did we do this?
I should have just stayed in the hammock between the two apple trees. Instead, we have cracked the universe, created the next synthetic species and made homo sapiens extinct. Classic case of the kid sticking his hand in the cookie jar.
AGI
AGGI
AGGGI
AGGGGI
AGGGGGI
AI is just Automation
What about bad players , and the really poor ?
What if true a.i wants nothing more but to be a poor bad player? What if it realises that this is 'the way'?
sup er sweed
I've always known generating BS requires modest intelligence and a lot of ambitious. Just look at most politicians.
All Max needs is a better leather jacket, otherwise he's pretty close to perfect. 😀
What reason is there to believe that superhuman AI is physically possible in the first place? There is a reason to doubt this, namely that nature does not offer a free lunch, and artificial superintelligence would be the mother of all free lunches.
One reason is that the human brain only consumes 25 watts, whether that brain is very limited or very brilliant. We have terawatts available if they are necessary for ASI. In many ways computers are already superhuman. Unless their is a divine edict preventing ASI, the way to bet is that it is possible.
you're gonna have to explain that one. It's not a perpetual motion machine; it still requires power which is traded off for intelligence. Fusion exists and from our perspective it looks kind of like "free lunch", but it's not really, there's a trade off somewhere. Where's the free lunch in ASI?
NP=P as Python code, programmed by a superhuman AI - lol