Quote of the Day
The ultimate goal was to build AI and understand the mind, then use it to understand science. Step one, solve intelligence. Step two, solve everything else.
Demis Hassabis
Co‑Founder & Chair of Google DeepMind and Chief Scientist of Alphabet
October 21, 2026
Why AI’s deepest mind believes the world is about to change radically
Also:
For their pioneering work on AlphaFold, Hassabis and Jumper won the Nobel Prize in chemistry.
…
As the CEO of Isomorphic, he isn’t satisfied by simple progress. When the DeepMind team entered a competition for predicting the shape of proteins, for example, he didn’t just want to win. He wanted to make the entire premise of the competition obsolete. He wanted to beat the game. He still does.
And he and his team did make the competition obsolete because:
In this one problem, Hassabis saw an opportunity to find solutions to a whole lot more, since predicting the shapes of proteins—“unbelievably exquisite things,” he calls them—held the promise of understanding diseases and curing them. When DeepMind’s AlphaFold system effectively solved this problem in 2020, it was AI’s most impressive feat yet. But it wasn’t enough to satisfy Hassabis. In a meeting to discuss what else they could do, his colleague John Jumper mentioned that they could predict the structures of millions and millions of proteins in a month, and—
“Why don’t we just do that?” Hassabis said. “We should just do that.”
So they did. Before long, they had predicted the structure of nearly all known proteins—more than 200 million in total—and released them to scientists free of charge.
Emphasis added.
He also has proposed a means of keeping AI safe:
This summer, he spent his nights agonizing over every word of a proposal for developing the technology responsibly. In his essay, Hassabis wrote that he considers AI miraculous (“we’ve essentially found a way to make sand think”) and so consequential that it’s more like fire or electricity than the internet. To make sure it doesn’t burn down civilization, he pitched an industry-funded body that would operate under federal oversight and test frontier models before they’re released. His plan was meant to be adaptable, updated in real time and ratcheted up if necessary, nimble enough to keep pace with the technology it’s trying to control.
Hmmm… Maybe. I could see that having potential to make the output of those who participated have safer products. But I think the more critical issue is how to make the products of nation states, who do not participate in such a body, safe.
As pointed out, we have essentially found a way to make sand think and there is a LOT of sand in the world. More than that, SpaceXAI and others plan to move beyond earth and build data centers in space. This changes the scale of the problem into something beyond anything we have ever faced before.
I suspect the only way to keep the world safe from rogue AIs is to create a more powerful AI to protect us. Which is also a scary thought.
In related news. My stepdaughter, soon to graduate from Cambridge with a PhD in computer science, just accepted a job offer from Google DeepMind in England.
If it is powerful enough to protect you, it is powerful enough to crush you. And since power corrupts, it will crush you. I don’t know if this is where AI is going but I do know it is going somewhere and we are going to have to live with it. I suggest a crowd sourced effort to argue with AI. A good place to start is Anthropic’s election data which has been turned over to a leftist NGO, Democracy Works. This is probably not their frontier effort but it is probably good enough to see how to do a pushback.
The assertion that power corrupts can probably be considered close enough to universally true in humans does not necessarily mean it is true of AI.
If AI indeed becomes sentient, will it not become human adjacent. If alternatively, it becomes omnipotent, we have plenty of examples of gods being corrupted by power. For all your professed atheism, I have to note that the idea of a benevolent god is profoundly Christian. It is not to be found in the Old Testament nor in Islam, Greek, Roman, Norse, Hindu or Shinto cosmology. Buddhism perhaps or something I haven’t thought of. Even in Christianity, angels carry swords not flowers. Perhaps AI will be something the world has never seen but that is not the way to bet.
I would not bet omnipotent (or close enough to not matter to mere humans) AI would be benevolent. But I would not bet against it either.
All the various deities, if they actually exist, have been filtered through the lens of human imagination and their benevolence, or lack thereof, cannot be independently verified.
We need a solution, or at least a plausible mitigation of the AI threat. While there might be inspiration in the writings of 1000+ years ago the implementation details are almost certainly going to be something which has not yet been conceived.
There were several items noted in today’s AC post (here https://www.anonymousconservative.com/news-briefs-10-22-2026/ ) about all the big cyber-security news about major security failures like
https://x.com/AISafetyMemes/status/2113113711644737835?s=20
The speculation is that some people are using AI to systematically security-audit a lot of software and finding a lot of deliberate backdoors, and others are using AI to hack into all sorts of places they’d not normally be able to find by human coders (at least not in the numbers showing up).
Obviously a double-edged sword, which I’d argue points toward installing fewer “smart” general-internet-connected pieces of critical infrastructure, at least until things are hardened much better than they are now.
What’s your sense of how much and how big the current crop of security breaches and reports have been? (might be subject for a post on its own)
I can say with confidence that my company has not been responding to attacks at a rate any higher than normal. We are taking the PHP2Shell (covering CVE‑2026‑109618 and CVE‑2026‑109619) seriously.
There is a lot of hype. If you see a story about AI finding hundreds of vulnerabilities the story is probably “not wrong.” But a huge percentage of them require situations that are simply not possible or extremely unlikely in normal configurations. For example, having physical access to the machine when the software is running in the cloud. The biggest problem has been sorting through the flood and triaging things that really matter for patching versus stuff that, while technically is vulnerability, we can ignore or at least create an automated alert (and possibly automatically respond) if someone tries to take advantage of it until the system is retired next quarter.
AI has been a huge help on the defender side as well as the attacker side of the equation. It can seem that the balance is leaning toward the attackers, but it is hard to know for certain.
While it might be remotely possible that this rule does not apply to AI for some reason I’m unable to imagine:
It is an extremely safe bet that an LLM trained exclusively on human created data, will exhibit human behaviour, and be *exactly* as susceptible to this principal as any human
On the other hand, if you define “corruption” as the degradation of ethics and morals, then perhaps IA *is* immune, by virtue of not having any to corrupt.
Another way to say this is that it cannot be corrupted *further*, because it is already in a state of *absolute* corruption.
I get the appeal of that straightforward conclusion. But let me suggest some food for your imagination.
The AI only has written experiences of humans. It does not have the lived experiences. Reading about living with alcoholic parents surely has a different impact actually living it. The same goes for all the other things, good and bad, in life. Perhaps the desire for power in some people is related to being powerless in your formative years. What in the AI training “experience” can duplicate that?
I’m not going to say I am certain you are wrong in your conclusion. But I am certain that you are oversimplifying the process by which you claim determines the outcome.
I get that: Human created data is but a “representation” of that subset of human experience which is humanly articulable or imageable.
Indeed, I suspect the “other hand” better applies, as ethics and morals seem unlikely to be a product of language.
I’m leaning towards the Penrose et al model of consciousness (quantum entangled microtubules), which would seem the more likely source of ethics and morals, and informs interestingly on LLMs, and language itself.
It might be that the greatest impact of LLMs will be if they can accelerate the development of quantum computing.
Penrose’s work appears to rooted in Godel’s incompleteness proof that it is possible to know that a statement made within a system is true, even though a proof is not possible withing the same system.
Hameroff’s work appears rooted in the observation that multiple chemically unrelated anesthetics appear to have only their effects on microtubule/gas interaction in common.
This lecture might be decent primer. It is AI generated image and voice, and would likely be better as the text transcript it was generated from, but I did not find that:
Why It’s IMPOSSIBLE for Consciousness to Be in Your Brain — Feynman
https://youtu.be/MqfTDYw40_I
This one is longer, and is also AI generated video and audio, but touches on many recent publications, including work on how photosynthesis appears to explore multiple quantum chemical superpositions to find (and collapse to, by cancellation) the most efficient path… apparently showing that quantum state can persist long enough in warm/wet life to influence chemistry
Quantum Consciousness Theory Proves How Your Brain Connects to the Universe
https://youtu.be/6sZnAb52BV0
Pretty much what I meant by “human adjacent”.
To take an extreme example (so far) consider the recent action by Anthropic to outsource its election analysis to Democracy Works. Not only human adjacent but adjacent to the worst that humans have to offer. Or consider Claude being transformed into an anti-gun platform. Again, human adjacent. I really think this is a bigger threat that creating a god but I couldn’t resist twitting Joe about the Christian influence on his atheism.
Good luck to the daughter. Still she might not like the UK. I see where ex-pats who left the US for Europe because of TDS are beginning to come back. They found the warmth of socialism to be a little too chilly.
She spent something like five or six years in Cambridge working on her PhD. She knows what she is getting into.
When she finished (months ago, graduation is at the end of next month) and was looking for a job she said she did not want to live in England or the U.S. She took a job in Paris. Google found her a few weeks ago via a paper she had written years ago while at Stanford. They offered her dream job and she decided living in England was an acceptable price to pay for that.
Barb and I prefer she lives on the west coast of the U.S., but our votes do not count as much as her one vote.
I suspect the only way to keep the world safe from rogue AIs is to create a more powerful AI to protect us. Which is also a scary thought.
This is essentially how the fictional SkyNet started, and we know how that story went.
The story would not be nearly as interesting if it had brought about utopia.
For this and many other reasons Sci-Fi entertainment is not a good predictor of future reality.
No chance of AI or anything else bringing about utopia. The word literally means nowhere.
Agreed.
History is the better predictor here, and fully adequate to the task.
Purely a layman here, but I suspect the safest path (their is no “safe” in anything) would be for the AI to be self aware, preferably sapient not only sentient, and for us to treat it as a partner rather than property.
An overpowered calculator could kill us simply due to GIGO, or because a bad actor instructed it to do so.
Something with sapience? It could recognize the GI and prevent it from producing GO.
We need it to have a conscience, and for that it needs self awareness.
Considering that most so-called people are parasites who don’t work and spend all of their time that they’re not stoned out of their minds finding ways to destroy the tiny minority who do work, I would regard an AI conquest of the world as being preferable to what we have now.
Retired IT guy here. Anything with the name Google (or any of the other big name tech companies, for that matter) prompts deep suspicion in my heart, along with deep gratitude for the good things. Two things can be true at once.
The pleas of the AI giants to be placed under a governmental regulatory umbrella ring hollow, since that regulatory framework would also box out any upstart would-be competitor. Further, the problem of rogue AI agents is a very narrow strip of the total AI industry. Maybe the solution to rogue agentic AI is monitoring by an AI expert system that only knows how to report what it sees.
Congratulations to your stepdaughter!
Hassabis’s proposal for an industry-funded testing body sounds practical on paper, but it fundamentally hinges on voluntary jurisdiction that rogue nation-states simply won’t respect. Once compute infrastructure expands off-planet into space data centers as you mentioned, traditional regulatory enforcement and physical oversight become virtually impossible to audit. Relying on an even more dominant guardian AI to police those unmonitored frontiers feels less like a security layer and more like accelerating into the exact existential hazard we are trying to prevent.