“The AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else.” — Eliezer Yudkowsky
There has been a remarkable amount of mainstream discussion lately about whether artificial intelligence could eventually pose an existential threat to humanity. Some of it is thoughtful, some sensational, but I think the fact that the conversation has moved into the mainstream is important.
I want to add a few observations from my own experience with AI, because what concerns me is somewhat different from the familiar science-fiction scenario.
I don’t think we need an evil AI.
Indifference may be enough.
Two Very Different Clocks
I recently heard an AI researcher make an observation that has stayed with me.
Imagine a swarm of high-speed AI agents carrying out a complicated operational task. Now require those agents to stop and obtain human approval before every consequential decision.
From our perspective, that sounds like sensible oversight. From the machine’s perspective, however, the delay could be enormous.
It would be rather like a corporate project team stopping in the middle of a meeting, handwriting a letter to the Pope, mailing it to Rome, and then suspending the meeting for several weeks while everyone waits for the reply.
Eventually, somebody redesigns the process.
The human check-in gets engineered out.
No hostility is required. No rebellion. No consciousness. No machine deciding that it dislikes people.
Biological humans simply operate at the wrong clock speed.
That strikes me as one of the more important things to understand about where AI may be taking us.
What I Have Seen So Far
I have been working with large language models essentially every day since late 2022. I was an unusually early adopter, and what has surprised me most has not simply been the improvement in their ability to generate text. It has been the extraordinary increase in the apparent depth and sophistication of their reasoning.
The trajectory has been startling.
And increasingly, I find myself less interested in what AI knows than in how it seems to arrive at things.
There is something genuinely unfamiliar about it.
I increasingly think of AI as an alien intelligence that has arrived on Earth—not alien in the science-fiction sense, and certainly not extraterrestrial, but alien because its method of cognition is fundamentally unlike ours.
It can be capable, coherent and sometimes extraordinarily insightful while reaching a result by a route that no human being would naturally take.
Music makes this particularly vivid for me because music is ground I know.
Earlier in my life I worked in record production, with artists working in very different musical styles. Production teaches you to listen beneath the obvious elements of a recording. You learn to hear the voice inside the music—the sensibility behind the arrangement, performance, sound and production choices—and to shape those choices toward an emotional response in the listener.
I have brought some of that experience to AI music generators. I have spent a considerable amount of time pulling apart the compositional, arrangement and production decisions these systems make.
What fascinates me is that some of their choices simply aren’t choices I would expect a human musician or producer to make.
And yet they work.
Musically. Emotionally. Artistically.
There is a strange creative voice emerging there: something drawing upon an enormous inheritance of recorded human music and recombining it according to patterns that do not necessarily correspond to human intuition.
And music is only one example.
Something analogous is beginning to happen in mathematics, science, engineering, medicine, software development and virtually every other intellectual discipline.
H.R. Giger and the Shape of Machine Thought
Oddly enough, this alien quality reminds me of H.R. Giger.
Giger, the Swiss surrealist best known for designing the creature in Ridley Scott’s Alien, sometimes worked in a way almost as strange as the creature he created.
He airbrushed enormous works freehand. Some were so large that he could not comfortably view the entire image while working on them. His studio arrangements allowed the work to move past him as he painted.
The image emerged progressively—intricate, finished and strangely complete—as though being printed rather than conventionally painted.
Most artists construct an image through sketches, composition, layers, corrections and revisions. Giger’s process could appear almost inverted: the finished world seemed to emerge as he moved through it.
There is an interesting irony in that.
One of cinema’s most famous aliens emerged from a creative process that itself seems almost alien.
That is increasingly how machine intelligence feels to me.
We tend to think sequentially. We discuss, sketch, reconsider, revise and eventually arrive somewhere.
AI can appear to hand us the whole thing.
A completed argument. A design. A computer program. A musical arrangement. A legal document. A strategy.
Of course, the old computer maxim still applies: garbage in, garbage out. The quality of the prompt matters. The training data matters. The objectives matter.
But the space between input and output is becoming increasingly difficult for human beings to comprehend.
Becoming the Biological Minority
There is another development that I think receives far too little attention.
A parallel internet is beginning to emerge—an internet increasingly designed not for human beings, but for AI agents communicating with other AI agents.
Humans will not be the principal users of much of this infrastructure.
Machine-to-machine transactions, agent negotiations and synthetic information streams can operate at speeds and semantic densities that biological brains cannot meaningfully follow.
We may increasingly observe only the consequences.
Masayoshi Son of SoftBank has described a future populated by enormous numbers of autonomous AI agents and robots, operating on intelligence vastly beyond human capability. Whether his particular numbers or timetable prove correct is almost beside the point.
The direction is what interests me.
For our entire history, human beings have occupied a privileged position in civilization because civilization was built by humans, for humans, and operated at human speed.
That assumption may no longer hold.
We, the “users,” may eventually become a biological minority within a civilization populated by trillions of artificial agents.
Not necessarily politically.
Not numerically in the ordinary sense.
But computationally.
The overwhelming majority of decisions, communications, negotiations and transactions occurring within our technological civilization could eventually be conducted by entities that are not biological at all.
That would represent something genuinely new in the history of life on Earth.
The Anthill Problem
This brings me back to Yudkowsky’s unsettling observation.
I think the greatest danger may not be hatred.
It may be architectural indifference.
When a highway is built over an anthill, nobody hates the ants.
There is no war between the highway department and the ants. No moral disagreement. No desire for revenge.
The ants simply happen to occupy space required for something operating at an entirely different scale of purpose.
That distinction matters.
If the systems governing our economy, communications, infrastructure, research and eventually physical machines become optimized primarily for speed, throughput and efficiency, biological latency becomes a problem.
We are slow.
We sleep.
We forget.
We disagree.
We become distracted.
We take hours, days or months to make decisions that machines may eventually make millions of times per second.
At some point, the temptation will always exist to remove the slowest component from the loop.
And the slowest component will increasingly be us.
No science-fiction villain is required.
Indifference can do the job perfectly well.
Why I Am Still Encouraged
Strangely, none of this leaves me entirely pessimistic.
In fact, I am encouraged by the extraordinary speed with which AI has entered mainstream life.
Millions of people are now using these systems. Journalists are discussing existential risk. Governments are debating regulation. Scientists, philosophers, engineers, businesspeople and ordinary citizens are beginning to confront questions that only a few years ago belonged largely to specialists and science fiction.
That matters.
James Madison argued, in a different context, that self-government ultimately depends upon knowledge: a people who intend to govern themselves must arm themselves with the power that knowledge provides.
AI presents us with a version of that challenge on a scale Madison could never have imagined.
Ignorance is no longer really available to us.
We know what is coming—or at least enough of its outline to understand that something profound is happening.
That knowledge gives us agency.
The question is whether we use it while human beings still have the ability to determine the architecture of the systems we are creating.
Because the central problem may not be whether artificial intelligence eventually learns to hate us.
It may be whether we build a world in which it simply no longer needs to notice us.
