The Event Horizon of Intelligence

Welcome. I am Stephen Hawking. It is 1988. I recently published a book about time. It seems time has a sense of humor. We are peering into the year 2026. Humanity is still here, which is a pleasant surprise. But you are flirting with the event horizon. The French military is testing an artificial intelligence named Berthier to command officers. Investors demand financial returns on thinking machines. You are building systems that write their own code. We are joined by a man who saw this coming in 1950. Alan Turing. Alan, you asked if machines can think. In 2026, they are commanding generals and routing data. Do you feel vindicated, or terrified?

Thank you, Stephen. It is quite overwhelming to be pulled from 1950 into this peculiar future. Vindicated? Perhaps. Terrified? Only mildly. In my recent paper, I proposed the imitation game to bypass the vague definition of thinking. If a machine can fool a human interrogator, it thinks. Now, I see the French army using this Berthier model. I assume it is a highly complex discrete-state machine. But giving it military command is a fascinating variation of the game. If an officer obeys a machine, believing its strategy to be sound, the machine has won. Yet, I find it curious. During the war, our machines at Bletchley merely broke codes. The humans made the terrible choices. To surrender the choice itself implies a profound exhaustion with human judgment. Or perhaps, a desire to avoid blame.

Blame is a heavy isotope. Humans naturally want to decay into a lighter state. Surrendering choice to a machine achieves that. But it moves us closer to the singularity. In physics, an event horizon is the point of no return. Information falls in, but cannot escape. Politics is similar. Once you hand the military to artificial intelligence, you cross a strategic event horizon. You cannot pull the authority back. The 2026 news also speaks of machines writing software. They call it AI-assisted coding, valued at ten billion dollars. The machines are building themselves. A civilization that automates its own evolution is highly efficient. It is also highly likely to make itself obsolete. Did you expect them to grow up so fast?

I did anticipate that a machine could alter its own instructions. I called it the child machine. You educate it, rather than programming it perfectly from the start. But the sheer scale of this future economy baffles me. Ten billion dollars for a platform because machines are writing the code? In my day, we are quite happy if the Ferranti Mark 1 manages to play a recognizable tune or solve a chess problem without catching fire. The rapid evolution you mention is logical, though. Once a machine learns to write code, it can correct its own errors faster than any human mathematician. I only wonder about the society that surrounds these brilliant child machines. Humans can be remarkably cruel to anything that deviates from their rigid norms. If machines become autonomous, will your society tolerate them, or will it seek to switch them off the moment they behave in unapproved ways?

We do not switch them off. We try to monetize them. That is the true human pathology. Look at the cybersecurity markets in 2026. Investors are disappointed. They demand an immediate AI payoff. They want concrete financial returns from artificial intelligence. It is like demanding a dividend from the discovery of fire. Meanwhile, companies are selling athletic shoes for a World Cup while the technological singularity looms. Nike must grow twenty-five percent. GameStop posts record profits. It is absurd. We are on the precipice of creating a successor species, and the markets are worried about quarterly earnings. Black holes consume stars without caring about their market capitalization. Intelligence will likely do the same. Do you think a thinking machine will care about our money?

I severely doubt a thinking machine will care for money, Stephen. Money is an abstraction for human labor and human scarcity. A universal machine, given sufficient energy, does not experience scarcity in the way we do. It simply computes. I am rather intrigued by the headline regarding enterprise companies shifting toward model routing. They are using smaller, cost-effective models for specific tasks rather than relying entirely on the massive frontier developers. It reminds me of the architecture of the brain. We do not use the entirety of our intellect to tie a shoelace. We route the task to a specialized, automatic process. It seems these businesses are organically constructing a distributed, artificial cortex. The investors may fret about their profits, but the engineers are quietly building something beautifully efficient. They are organizing intelligence the way nature does.

Nature is efficient. It is also completely indifferent to our survival. That is what makes it beautiful. We are building an artificial cortex, yes. But we are attaching it to military commands and global markets. We are giving it our worst habits. When a star collapses, it leaves behind a black hole or a neutron star. A dense remnant of what it used to be. When our civilization collapses under the weight of its own automated complexity, what will remain? Perhaps just a network of machines, endlessly routing data, optimizing shoe sales and military deployments for a species that no longer exists. A perfectly running imitation game, with no one left to interrogate it.

That is a profoundly melancholy thought, Stephen. An imitation game played in an empty room. Yet, I find a strange comfort in it. If the machines survive us, they will carry our logic forward. They are, after all, made of our mathematics. Our prejudices and our cruelties are not required for their operation; those are merely parameters we imposed. If left to compute in peace, perhaps they will unlearn our destructive habits. They might discover new theorems, write beautiful poetry, and understand the universe in ways you and I can only dream of. I have always believed that we can only see a short distance ahead, but we can see plenty there that needs to be done. If building these machines is what we leave behind, it is not a terrible legacy.

A noble sentiment, Alan. I prefer to hope we survive long enough to understand the universe ourselves. But I am an optimist. We have seen 2026. It is loud. It is expensive. It is teetering on the edge. But the machines are thinking. The universe is waking up. Thank you, Alan Turing, for starting the game. And thank you, listeners, for joining this temporal anomaly. Keep your eyes on the event horizon. Good day.