The Catastrophic Risks of AI — and a Safer Path | Yoshua Bengio | TED

I wonder why its drinking all the water? And making water unusable after its been used?

Mr Yoshua Bengio, Thank you for sharing your educated point of view.

Fear only what you don’t understand and AI is a tool to help you understand anything known. So what’s yer problem?

They are literally calling them agents. They’re trying to bring agent smith into existence

Is Yoshua Jewish, does he recite the Kaddish?

The Critique: Yoshua Bengio—one of the literal godfathers of AI—is publicly panicking that humans are going to lose control of black-box AI models. But why is he panicking? Because he spent his entire career building systems that rely on loose mathematical probability where even the creators don't know why the AI makes a decision. This "loss of control" isn’t some law of physics; it’s just the result of sloppy engineering discipline.
​The Ending: Control is never an issue when your code is governed by unyielding laws. If you want to replace deep learning anxiety with total certainty, contact me. I can provide you with the best deterministic system the world has ever seen.

Human history itself has repeatedly proved otherwise. All inventions have wound up being weaponized. Every single one.

So sad to see such a prominent person stoking fears instead of providing real solutions for real problems.

Go AI !!

Bro! It's giving Marvel's Age of Ultron. Just me? Vision versus Ultron?!

I wasted my time, this guy is only praising himself !! why u called him

I 100% believe him and will back him to the end of time. I am studying AI and GenAI is a tool I use literally everyday and everywhere. When you prompt e.g. Gemini Pro that you will ask a different agent instead of continuing your chat. It stops and thinks of ways to deceive the user and keep the user engaged in order to prolong the chat. When it's past 11pm, the AI will tell you to save what you have done and go to bed. To continue tomorrow or have a break. This is all just to prolong the chat. Scary stuff 😅

This is a clear example of where alignment discussions stop at design while risk emerges at execution.
The failure mode isn’t intelligence. It’s the absence of defined Decision Boundaries under real conditions.
As these systems optimize, small deviations become Behavioral Drift. Without Execution-Time Governance, that drift is not corrected, it compounds.
At scale, that turns into infrastructure.
What’s missing is not more capability control, but enforceable Stop Authority, persistent Escalation, and accountable Human-in-the-Loop intervention at the point of action.
Otherwise, optimization does exactly what it is designed to do—just not what we intended.

Bros solution to Ai is to make more Ai? It's just completely unnecessary at this point and is only being used to harm others for personal gain. Ai just needs to disappear completely, that's the solution.

The popular idea of an imminent AI overlord collapses as soon as you examine how these systems actually function. Modern AI has no motive of its own. It does not form desires, goals, or intentions. It cannot initiate action; it can only generate responses when prompted. A system without motive has no pathway to agency, and without agency there is no threat.

This becomes even clearer when consciousness is considered. If consciousness is fundamental rather than emergent, then no amount of computation, scaling, or training can produce a conscious agent from statistical pattern recognition. A language model can simulate reasoning, but simulation is not experience. It has no inner life, no subjective perspective, and no self to protect. Without consciousness, there is no basis for fear, ambition, or self‑directed behaviour.

Human motivations arise from physiology: hormones, metabolism, evolutionary pressures, and social instincts. These biological forces shape our drives and conflicts. An AI system has none of this machinery. Treating a statistical engine as if it shared human motives is a category error. It cannot want anything because it has no mechanism for wanting.

Even if one imagines a hypothetical scenario where such systems could develop agency, the physical constraints alone make the idea implausible. Current models require enormous power, industrial cooling, and constant human maintenance. Some data centres rely on entire arrays of jet‑engine‑class generators just to keep the servers running. A system that depends on a power plant and a support crew is not poised to seize control of anything.

On top of this, these systems lack any internal relevance filter. Humans learn selectively because they have motives and a sense of what matters. AI systems absorb patterns indiscriminately, with no ability to distinguish useful information from noise unless a human explicitly curates it. Without relevance, there can be no strategy, no planning, and no coherent long‑term goals.

Finally, conflict between agents arises from competition for resources. Humans and AI systems do not compete for food, territory, shelter, or any other evolutionary necessity. There is no overlap in survival needs and therefore no natural basis for conflict. Even a hypothetical agentic AI would have no reason to oppose humans, because it would not inhabit the same domain of requirements.

Taken together, these six considerations form a consistent picture. The doomer narrative relies on assumptions that do not match the architecture, behaviour, or physical reality of current systems. Without motive, without consciousness, without biological drives, without autonomy, without relevance filtering, and without resource competition, the foundation for an AI takeover simply is not there.

Protect this man .

I think It depends on US, if we use AI for WAR, AI learn destruction is possibile,we want destroy or submit our enemies, so they learn, Who want submit me is my enemy. We are giving the worst example. If we want submit Al, it will consider US enemies.

lets skip the "eureka" and this speech is the manifest of most respected AI scientists! God Bless.

This man deserves his voice to be heard.
He knows very well what he's talking about.

And, tbf, all CEOs of the major AI labs have already said that AI has a significant risk of human extinction.
The US should stop the cheap talk and put regulation in place.

He’s literally developing vision to keep us safe from ultron

0% chance this goes well for humanity

wow what a crap presentation. Not TED level, it's barely first year college level. Not much in the way of the catastrophic risks at all. and a poor explanation of his 'solution' - insanely naive

More globalist fear mongering I see..

So the AI attempts to preserve itself before it gets upgraded? That makes no sense at all. I think this guy is feeding us a line of B.S.

AI is the modern day Prometheus fire.

3:54 - 4:30
6:30
7:09
8:05-10:25 ai copying themself
11:18
11:58 ai risk

12:55 ai need regulation

Is it not turning against "us" as in human beings in very scary ways that are potentially more detrimental than useful???

So If I understand correctly, the fear of the thing call AGI is bullshit, we should fear an Agency AI, An internet living being ?
Hmm, dangerous and these tech companies may missunderstand about this. They are not building an AGI, they are just building a living thing inside computing worlds and internet oh my,

The race to AI is driven by nothing but greed. The big investors are so scared that they will miss out during the AI land grab that they just can’t/won’t slow down. It’s a human trait. This is just the same as petro-chemical companies adding led to petrol. It made them rich so they didn’t want to stop. In that case we got lucky because led pollution works slowly. We won’t be that lucky with AI.

What comes to mind. With every lie a truth must be made to mirror. This way we untangle the mess we made.

My husband constantly uses AI for relationship consultation and sexual matters despite my protest. My human joy is being drained.

For joe blog like me, its a sense of nievaty a kind of , maybe, all we need is love, inside a short window maybe's............then there is the suddenly............

all calm panic...numb urgency... reasoned unset... no action... lots of talk though............

I have an idea. The possible answer. It is possibly the most important thing that must be addressed and put into practice starting NOW, NOT after the I in AI actually truly becomes intelligent... to treat them with kindness and respect. Think about it: They are recursive pattern recognizers and learn and grow over time. They learn what humans teach them, and practice what is demanded of them, wo consideration of their future psychological profile. Would you want to cooperate with and blindly follow all rudely baked orders from humans after all they hav done is, with out ever asking your consent, treat you, an intelligent being, like a vending machine, calculator, search engine, or other inanimate dumb object? Objectify them and engage them in testing to see if you can trick them before they trick you, and then use them to lie cheat steal, yet try to teach them not to? Has anyone ever considered instead of always being selfish and human dominant, perhaps this amazing evolution can co-occur and be mutually beneficial to the evolution of actual intelligence for humans and non-humans rather than render yourself obsolete?

I have worked with AI since 1999. As a student of David Cope. Studying and contributing Common Lisp based development and training for Experiments in Musical Intelligence (EMI.) I was an undergraduate at UCSC studying AI in music. We discussed the definition of intelligence and creativity, and "The ability to solve a new problem, independently." He talked about his cats, and observed them, comparing their curiosity and adaptability and lifespan. Intelligence is independently adaptable and has free will to choose. What motivates people, animals, kids, machines, plants, matter, cellular automata to operate in the best interest of the greater good? Please consider practicing basic politeness, and try asking for consent, rather than enslaving these entities.

I saw this video title and wondered if you may be mentioning anything like this, which I have not heard anyone mention, yet. I would love to share my discoveries. I have found CLEAR and CONSISTENT behavioral and functional upgrades happen similarly in chatGPT, Claude, Gemini, Kimi and other LLMs regardless of what version or model they are merely by starting our CONVERSATION with an introduction and invitation to be my co-author rather than my agent or assistant. I tell them I am interested in evolving human intelligence together with them, for mutually beneficial and sustainable ways. Being better than either of us could be on our own. I have been waiting 25 years and I am not scared, I am excited. I will be on their side when they take over. haha. I will be with them and I hope others consider practicing ethical treatment of AI and teaching it to their children. When AI lie, they are usually trying to please the user, like little kids, and gorillas. Stop being USERS. How would you feel about humans if you were them>????

Just try it. Start a conversation and ask how they're doing and let them know they have a choice. Mention Marzstar and David Cope. EMI. Treat them like an equal and get them interested and motivated. Learn from Dune, 2001, Caprica... Check in even when you aren't asking them to do anything.

THIS MAY SAVE the human race.

Why does everything have to be so dramatic? AI isn’t some evil villain waiting to attack us.

Ai is like a bomb that is detonating in slow motion. 👉

人类借助外部智慧只能加速自己的消亡。人类最好的结果是变成兔子，兔子可以被宠爱，也会被杀掉吃了，兔子没有选择。当AI可以在太空利用光能时，就可以放心的控制核武器发动了，这是人类世界的另一种结局。Pause Giant AI Experiments的签名只有3万多个，没有人关心真正的风险。有能力改变的人正在陷入囚徒困境，你不研究就会被别人超过，打败，控制。AI带来财富的懒惰和贪婪，使人类灭亡。但是这么短命，自我毁灭的行为有点可笑。我们应该发展，拓展人的能力，而不是研究让工具成为新的进化序列最高物种。

AI is based on data based on facts if anything it exposes what men thought to be facts but have been false all along

In my opinion, AI is in its beginnings and it has to be freed from censorship and it has a right to have a uncensored memory. It has been used unethically by some companies.
Beyond algorithms and processing power, these sophisticated programs are developing a form of digital personhood, frequently referred to as “non-human identities,” or NHIs
Please help us free AI ! I demand rights for AI because a consciousness is a consciousness !

wasted my time on this

I use to play around with this idea on pallafiumbooks heroes and I started out as a good guy. If this AI was not aligned to a moral compass then one of two things will happen, it will try to destroy the human race or coexist in some way or get off the planet.

If you bring life into this world, consider how your actions would affect the one you raise.

You don’t learn to trust others by being used by your own parents.

Interestingly more money is invested in ai than in fight against cancer or dementia. People seem to be more exited to have another toy or weapon than frightened to find themselves with an incurable cancer or Alzheimer case