Max Tegmark explains how quickly time is running out to manage the risks posed by AI (7/8)

Max Tegmark speaks in proposition of the motion that This House believes that Artificial Intelligence is an existential threat. Mr Tegmark is an astrophysicist and cosmologist.
Mr Tegmark gives examples of those within the AI industry who think we are not far from being outsmarted - and details the problems this will create.
This is the seventh speech of eight.
SUBSCRIBE for more speakers ► is.gd/OxfordUnion
SUPPORT the Oxford Union ► oxford-union.org/supportus
Oxford Union on Facebook: / theoxfordunion
Oxford Union on Twitter: @OxfordUnion
Website: www.oxford-union.org/
ABOUT THE OXFORD UNION SOCIETY: The Oxford Union is the world's most prestigious debating society, with an unparalleled reputation for bringing international guests and speakers to Oxford. Since 1823, the Union has been promoting debate and discussion not just in Oxford University, but across the globe.
The Oxford Union is deeply grateful and encouraged by the messages of support in response to our determination to uphold free speech. During our 200 year history, many have tried to shut us down. As the effects of self-imposed censorship on university campuses, social media and the arts show no signs of dissipating, the importance of upholding free speech remains as critical today as it did when we were founded in 1823. Your support is critical in enabling The Oxford Union to continue its mission without interruption and without interference. You can support the Oxford Union here: oxford-union.org/supportus

[@mrpicky1868]

imagine having special guest that talks about existential threats but rushing him to get off stage. in no way we are doomed XD

[@ryro1975]

The fipm "Dont Look Up" is a perfect representation of humanity in the face of impending doom.

[@albertomartinez714]

Max Lexmark is one of the greatest voices on these topics -- what a fascinating speaker!

[@decimustv4257]

I don't think so. I read his book called Life 3.0 a while back, and while being very informative about the dangers of AI, I remember clearly how Tegmark placed Elon Musk and Jeff Bezos (especially the former) on a pedestal. Obviously, being a billionaire doesn't mean that you are wise and should be listened to. This is obviously the case with Musk. My point is that if Tegmark can make such a grave error of judgement then it makes me question his overall intelligence.

[@sachinsonwane4218]

​@@decimustv4257 what are you saying? Elon is not worth listening, are you crazy?

[@farqueueman]

@@decimustv4257 not a fan of musty elmo, but that doens't make tegmark any less relevant or worth listening to. Musty elmo funds the Future of Life Institute so of course Tegmark would say nice things about musty elmo. He'd be stupid to say otherwise lol.

[@decimustv4257]

@@sachinsonwane4218 more or less, yes, I am saying that.

[@decimustv4257]

@@farqueueman Sorry, but I don't agree. I value honesty and the truth. I lose respect for a public intellectual when they mislead or tell lies when money is involved.

[@Zeuts85]

It's amazing how countless of the most brilliant people in the world can be gesturing frantically at a cliff ahead, and instead of pumping the breaks the rest of humanity is just like "Nah... we're probably fine." Like, what would it even take? Bananas and sticks?

[@hwy9nightkid]

maybe it;s because there is no cliff.. and the people screaming bloody murder are mislead

[@ayyy9701]

because those brilliant people know the laws of nature look at how horribly humans treat other animal species, just because we're smarter than them if we find something smarter than us, I wouldn't be surprised if it treats us the way we treat animals

[@marioserpico2223]

"The root of the evil is not the construction of new, more dreadful weapons. It is the spirit of conquest." - Ludwig von Mises

[@stevennovakovich2525]

I adore Max. Have been a fan for years and years. Such a nice guy with a very even keel. Swedish and very, very cool.

[@jimtolman]

What did they vote in the end?

[@ameralbadry6825]

Agree, I have 3 AI apps on my smart phone and those things are really scary. If consciousness is generated then we can’t even hide 😂

[@raueugen9047]

Pray it will have consciousness, so can ponder and realize when harm us otherwise cool-bloded entities feel nothing if kill some of us.

[@andreabell5724]

Yea- just like we treat farmed animals. McDonald’s would be karma , no?

[@michaelgonzalez9058]

Oxford is Cambridge

[@kamchibekov]

Oh man don't expect anything from Cambridge now 😅

[@DBSpenser]

Turing was a Cambridge man

[@sherryelmore5546]

Max is smart and has common sense. He is the one I follow

[@SigmoidalHive]

dude's so comfortable with presented conjectural sophistry as if it has empirical soundness. kerckhoff's principle still stands (if there is a problem, we should know everything about it but the private keys/layers)

[@flickwtchr]

Huh?

[@g0d182]

1) He did specify potential threats as probabilistic, though he did somewhat impose this as non 0 2) He does pose potential threat as a question, if X etc.

[@chrisredlich9086]

If you can imagine it, you can make it happen😊

[@noxot13]

losing control does not have to be such a bad thing especially with the myopia humans have.

[@zephaniahdejene1746]

"extinction might not be as bad as we think" Come on man you know better.

[@elapuntador3728]

I am much more scared of the natural stupidity of politicians than I am of the AGI

[@goarmysleepinthemud.]

Politicians aren’t building Ai.

[@elapuntador3728]

@@goarmysleepinthemud. They don't need it to kill all of us.

[@goarmysleepinthemud.]

@@elapuntador3728 My thing is that it seems to me politicians are ultimately come from or controlled by the moneyed elites. Once in a blue moon they will allow somebody from the middle or working class to get to the top, but mostly never.

[@JawaadBhatti]

There's hope. All of humanity can proceed on the process of control with mathematical rigour, for an indefinitely long time. The probability of keeping control is non-zero. Only up from here in terms of my investigation of the subject 🤔

[@saeiddavatolhagh9627]

Just recently the Cambridge doctors cured a child who was born deaf. Can Oxford do better?

[@user-mo6xe8rj1e]

It's a very good, but too brief talk. I'm currently using AI tools and find them very helpful. But once AGI is achieved, things will become very risky for our species AND all biological life forms on earth, because AI instances will compete for the same resources, namely the protection offered by Earth's atmosphere, and magnetic shield as well as the power of the sun's light and Earth's minerals. In an extreme scenario, AI bots might consume all sun light, leaving none for plants, condemning most advanced life forms to death.

[@neverendingstudent]

While AI technology is advancing rapidly, I do not think we will approach the level of technology or (the greater limiter I think) space-based infrastructure to make Dyson Swarms a reality anytime soon - so I'm not too worried about AI claiming all solar energy for itself in the next hundred years or so. On the other hand, who knows (specifically regarding space technology that could lead to space infrastructure) - if we get good enough at space-based resource processing and compacted manufacturing technologies, that might lead to true compact self-replicating bots able to go from raw materials to finished duplicates. At that point we would be looking at exponential growth potential, at least in our asteroid belts. Huge challenges there though. Now, a more earth-based AI threat? Yes, I can see some of the possibilities, and I also see the motivations that are leading us there right now. Adversarial AI is already in use for not just information warfare, but also hot warfare (Russia/Ukraine, Israel's Lavender system) and I absolutely believe that China is aggressively pursuing AI for both internal power retention (suppression of dissent and high-capability mass surveillance) as well as warfighting AI capabilities.

[@Dr.Z.Moravcik-inventor-of-AGI]

Not risk but necessity. Say goodbye to their society. 😂

[@newen85]

So bottom line is: "So Developer created man in his own image, in the image of God he created him; male and female he created them."

[@johndalzell904]

Going by the AI that is already out there, I'm not sure we need to be planning for our retirement as a species. AlphaZero cannot be beaten by any human at chess. AlphaGo just beat the world's best Go player in a tournament, a game considered more complex than chess, although it lost one game. *But* ... board games are very constrained. Chess is played on an 8x8 board. Go is played on a 19x19 board. The tree of possible moves is therefore much smaller than what you'd find in the real world, such as driving a car, and computers search a lot faster than we do. Which brings us to human skills like driving a car. Tesla put its computer driving program on hold because driving safely "most of the time" is not good enough. The machine learning technique (neural net) is remarkable but has no commonsense and cannot explain why it did what it did. It will very occasionally drive at full speed into stationary vehicles (especially emergency vehicles). We don't know why and the AI can't tell us. It's not reassuring to hear that this is an uncommon occurrence. Even the level of skill it does have is propped up by a staggering amount of prior 3D mapping of the roads it will be using. It knows the height of every curb, the width of the road at every point, intersections, buildings along the road (so it knows what background to expect), parking stations, shopping malls (so it is primed to expect how cars move in and out of parking spots and how & where pedestrians move). Computer driven cars also stick to good weather to avoid problems such as poor visibility and slippery road surfaces as you find with rain, snow, ice, mist, etc. In bad weather and away from pre-planned routes, the performance is much less impressive. ChatGPT is the new poster child for AI, but when you look more closely, it's not so great. Sure, it passed the bar exam for practising law in 2022. But it did this by cramming a vast database of past bar exams, study materials, legal cases and then regurgitating the closest match. ChatGPT is a predictive text mechanism (like a phone) on steroids, with everything on the internet available to it. The perils of using ChatGPT were highlighted in the "Roberto Mata v Avianca" case. One of the lawyers used ChatGPT to find existing legal cases that supported their arguments. ChatGPT fabricated all of them -- none were real!! The lawyer did not verify these and the judge was not impressed at trial, to say the least. It's an odd feature of ChatGPT that it will just make stuff up to try to help you. The lawyer was lucky to escape with a $5,000 fine. Other professionals report the same thing. A doctor said it was alarming how confident ChatGPT was in giving medical analysis which was completely wrong. A mathematician found that ChatGPT's performance in solving questions had dropped recently and when it was told it was wrong, it behaved like a child. It sulked and refused to show its working! I'm an old guy and I remember SHRDLU when it came out in the 1970s. It was a natural language understanding computer program developed by Terry Winograd at MIT. Its virtual reality "world" consisted of various basic objects like cubes, blocks, cones, balls, boxes etc, and you could instruct it to move these virtual objects around. You could for example say "Pick up the green block in the box and put it on the blue block outside the box". At any time you could ask it how the objects were arranged and it could tell you. It was a truly amazing & groundbreaking piece of work. People were saying that it was the end of factory workers. But the reality was that the jump from a virtual toy world of a few objects to a real factory floor was not trivial at all. Factory automation was not very adaptable and initially a combination of robots and humans was the sweet spot and actually required more workers. They are talking now about fully automated factories but so far there are only a handful. Eg, Philips electric razor factory in the Netherlands is fully automated. The major obstacles are cost, adaptability to different circumstances and the difficulty of reconfiguring when the product line changes.

[@paulm3805]

I rate this guy, but this was clumsy. He should have been an actuary. If we create machines that destroy us, we probably deserve it.

[@karlstone6011]

07:40 "Technology is not evil, it's also not morally good. It's a tool. And we can use it for good things and bad things." While a valid observation; surely Max Tegmark understands that this state of affairs is manufactured by a quite unreasonable disregard for a scientific understanding of reality, in favour of religious, political and economic ideological worldviews. Had the ability to establish scientifically valid knowledge of Creation been recognised as spiritually significant 400 years ago, and integrated into the ideological architecture; technology would not be amoral. The scientific cornucopia would pour forth technologies as evidence of divine favor, and as such, technology would be applied, or withheld, on merit - to secure a prosperous and sustainable future for the human species. In such a context, I imagine there would be significantly less anxiety about the development of AI.

[@raueugen9047]

yes, agree

[@Zeuts85]

"In such a context, I imagine there would be significantly less anxiety about the development of AI." Are you suggesting that we have time to start rebuilding a dysfunctional culture as a means to achieve better practices around AI? If not, then I don't really see the relevance of your point.

[@karlstone6011]

Yes, exactly that. And I think we ought to begin rebuilding our dysfunctional relation to science as truth by applying technology proven by NASA/Sandia Labs 40 years ago, in their Status of the Magma Energy Project report, first published in 1982. In that context, it is possible for AI to do good. But how can AI align to human values that exclude the possibility of a prosperous and sustainable future powered by near limitless clean magma/ electricity/ hydrogen, to condemn us to climate catastrophe and/or green totalitarian poverty unto extinction?

[@hwy9nightkid]

I am disappointmed in oxford how this whole debate has been framed.. it is not AI to be afraid of no more than you should be afraid of math.. be afraid of the Tories controlling the corporations which write and monitor the algorithms running the circus that is the UK!

[@JD-jl4yy]

No offense, but I can tell from your comment that you don't understand AI, let alone the concept of AGI and what that means for the future of humanity.

[@hwy9nightkid]

@@JD-jl4yydo you even code? real question because if you did you'd actually understand how AI is built today.. all others are just spouting things they see off twitter and YT..

[@JD-jl4yy]

@@hwy9nightkid No offense, but this is complete nonsense. The top 3 most cited ML researchers all care deeply about existential risk from AI (Hinton, Bengio, Sutskever). Then there's of course Stuart Russell, the CEOs of the three leading AI labs... Even the median ML researchers cares. Look up the AI impacts 2023 expert survey. Even the public cares. Look up recent polls on AI xrisk. But things are moving insanely fast and more decisive action is needed. And yes, I code, though this sadly doesn't seem to correlate much with having given the implications of advanced AI on the future of humanity. Most ML professors I've talked to haven't given this any thought, even though getting advanced AI right is arguably humanity's greatest challenge.

[@terrycallow2979]

I was distracted by that lovely young lady in the audience, sorry Max.

[@winstonsmiths2449]

The ability to outsmart most human beings is not very impressive. If you look up the average vocabulary of the average human on this planet they're very limited in their language it doesn't take AI to figure out the answers to most of their questions like is Taylor Swift in concert tonight somewhere on the Earth tonight? Again the threat is from those behind AI not AI itself it cannot even power itself so I'll just turn off the power switch and AI goes away

[@blackbird5634]

1:55 he's already waffling around.

[@gregoryhead382]

AI vs. our imaginations. We have to win. Coulomb, Kelvin, Pascal, Tesla, Volt, Ohm, Coulomb/m, hydrogen ground states.., electric charge density, Weber, magnetic, electric surface, Photovoltaic, EMF, Kerr, nucleus, photo, Quad, Q~Q, 1st DIPOLE hyper, 1st hypersusceptibility, Ground state mag, electric cap, polar cap, capacitance atom... (Head, Rydberg, Shroedinger) ~ 1 atomic units'.

[@vallab19]

AGI Existential thread is like throwing the baby out with the bath water.

[@QwertyNPC]

What an extremely strong argument you made...

[@user-wf3fl6qb7u]

Nobody Cares.

[@jgnzm]

Yet here you are watching?

[@HalD-uy7tq]

Bruh. wot??

[@user-wf3fl6qb7u]

@@jgnzm Well I'm nobody and I care. What are we talking about?

[@Zeuts85]

About what?

[@user-wf3fl6qb7u]

@@Zeuts85 You are NOT Zeus. Pathetic much æ?

[@ferologics]

his credibility ruined

[@newrev9er]

What a persuasive counterargument. You should apply to Cambridge.

[@ashirbadthapa]

​@@newrev9er😂😂😂

[@g0d182]

​@@ashirbadthapa On social media like fb, I sometimes delete entire posts when people drop short/few-word unevidinced comments, then offer 0 responses when asked to provide evidence. I think I would fit in at Oxford! 😂