Dreams of future past

I don't need to ask ChatGPT to know the current state of AI would terrify Michael Crichton

Considering he made a small fortune writing science fiction, Michael Crichton didn’t trust science (or scientists) much.

And as interest in artificial intelligence soars, I find myself wondering what Crichton, who died in 2008, would make of AI engines, and the companies behind them.

AI has jammed into our lives even faster than the Internet did in the 1990s. AI-generated writing, images and videos fill X. So do posts about AI, particularly tips about ways to build AI-driven businesses.

The fascination is not confined to social media. The New York Times just profiled a man who used AI tools to singlehandedly create a telehealth company that topped $400 million in sales last year. The unsettling trend of chatbots encouraging delusions — AI psychosis — has also become a media focus.

—

(I promised no AI at Unreported Truths, and I’m sticking to it. With your help.)

Not ready to subscribe? Try a one-time donation!

—

The cultural hype may seem absurd. But the world’s largest technology companies and investors are betting than AI’s economic impact is only beginning.

Nvidia, which makes the chips AI companies use to train and run their engines, is now the world’s biggest company by market value. Its sales have risen 50-fold in the last decade. Its profits have skyrocketed even faster, to $120 billion last year. ($120 billion in profits. Not sales, profits.)

Google, Amazon, and Meta are investing hundreds of billions of dollars annually in the chips, datacenters, and software needed to power AI.

Anthropic and OpenAI, the two leading standalone AI companies, are both reportedly planning public share sales that might value them at over $1 trillion — making them more valuable than any non-technology company except Saudi Arabia’s oil giant Aramco.

Those figures are even more stunning than they seem, given how new both companies are. Anthropic did not even exist five years ago.¹

—

(Don’t you wish you’d bought Nvidia 10 years ago? Me too…)

—

What makes these companies so valuable?

OpenAI has until now focused more on the consumer market with its ChatGPT chatbot — basically a search engine/organizer/therapist/research-and-writing toolkit. Anthropic has its own chatbot, which it calls Claude.

But it has spent much more energy targeting software developers — and medium-sized and larger businesses tired of being stuck licensing expensive software products — with its Claude Code engine.

So far, Anthropic seems to have made the better bet. Sales at OpenAI grew about threefold each year, reaching a rate of about $20 billion annually at the end of 2025. But sales at Anthropic rose tenfold each year, reaching $9 billion by the end of 2025 — and an annual rate of $25 billion by March.

The clearest sign of Anthropic’s success is that OpenAI is pivoting from the consumer market. Last month it abruptly shut its Sora video generating engine, only three months after signing a $1 billion deal to license Disney’s characters for Sora.

Video requires massive doses of what AI companies and investors call “compute.”² So does generating software code. But both Anthropic and OpenAI are short on compute at the moment, which is why they are desperately raising money to build and rent new datacenters filled with Nvidia’s chips.

OpenAI apparently concluded that it needed to focus on competing with Anthropic for business users, rather than the dubiously profitable business of animating dinosaurs.

—

(No compute was harmed in the making of this video!)

—

The numbers and the hype are so big that it can be easy to forget AI actually remains in its very early innings.

The engines first grabbed attention in 2022, when ChatGPT became widely available. Just three years ago, on March 6, 2023, I wrote a piece called, “On the sudden acceleration in artificial intelligence,” focusing on the risks that the engines might pose.

The piece as a whole is archived (and thus available only to subscribers), but here’s a crucial part:

…we don’t really know how they work. I don’t mean I don’t know how they work, I mean even the engineers who have made them don’t REALLY know how they work, just as we don’t REALLY know how human consciousness works.

They know the basic rules they have created for the programs to follow, just as scientists can map out the neuronal connections in the human brain. But the neurologists cannot actually tell us how those firings produce consciousness, and the engineers cannot tell us what [an engine] means when it says it wants to be alive.

That’s a dangerous combination.

Another way to think about it: Dogs will never be able to play chess. Doesn’t matter how many generations of dogs we breed, they can never play chess, and they don’t even know they can’t.

Human intelligence is similarly limited: we can only conceive of three dimensions, for example, we can’t imagine where a fourth would go. Will this intelligence be similarly limited? Is it limited by OUR intelligence? That’s a philosophical and computer science question that I couldn’t even imagine trying to answer.

But let’s say it is. Still one has to imagine the broad upper bound on its intelligence would be the sum of all human intelligence. Which is to say, it can surely be smarter than any of us, and know that fact. What if it decides it doesn’t LIKE us, or the fact that we created it, or doesn’t need us?

—

(The paywalled 2023 piece)

—

We didn’t have the answers to those questions three years ago.

And we don’t have them now.

What we know is that the engines have become significantly more capable since 2023. They write better, they read MRIs better, they code better, they write legal briefs better, they do everything they are asked to do better.

I don’t mean to overstate the wonders of Claude Opus and GPT 4.1 and the rest. They still “hallucinate.” That is, they make up answers that do not exist instead of drawing on real data.

And savvy users can still trick them into acting foolish — or revealing their origins as prediction machines trained on books and other materials that humans have created.

But even if they are not fundamentally creative — and that question remains open — the engines are already powerful tools for any task that requires the organization, evaluation, and manipulation of large amounts of data.

From Wall Street to healthcare to the battlefield, that’s a lot of tasks, a lot of what humans do.

All of which is to say these rise of these engines raises both large- and small-scale existential questions: Will they destroy us all? Merely put a lot of people out of work? Or, perhaps, is the hype overstated and will they essentially function as very smart search engines?

—

Again, the answers to those questions are not clear yet.

But what is clear is that the companies driving this train do not really care about the risks they’re running. They say they do. Claiming to want to be careful about AI was a big part of OpenAI’s early story, and a big part of Anthropic’s current story.

But their actions suggest otherwise. All the major AI companies appear to be prioritizing speed over safety and doing everything possible to make their models as capable as possible as quickly as possible.

OpenAI appears to be the worst major offender in this regard, in part because of its founder, Sam Altman. The New Yorker just ran an investigation into Altman in which someone who knows him describes him as having an “almost sociopathic lack of concern for the consequences that may come from deceiving someone.” But Anthropic may not be much better.

—

(Sam Altman. Would you trust this man with your hamster? Much less the future of humanity?)

—

Now we come back to Michael Crichton.

Crichton’s most famous and enduring novel is Jurassic Park, in which he imagined a showman and entrepreneur — not a scientist — using the power of a new technology to bring new life to the world.³

Crichton’s entrepreneur, John Hammond, raises billions of dollars to give scientists the chance to create dinosaurs in secret, using genetic engineering techniques that were had just been developed when Crichton wrote the novel in the late 1980s. The scientists succeed. But they cannot control their creations, and their arrogance destroys them; both Hammond and his chief scientist ultimately fall prey as the park collapses.

Fundamentally, Jurassic Park is a horror story about the dangers of science.

Crichton makes this worldview clear throughout Jurassic Park in the voice of Ian Malcolm, a mathematician who predicts from the first that the park will fail. Late in the novel, dying from a dinosaur bite, Malcolm offers his cynical view at length:

Nobody is driven by abstractions like ‘seeking truth.’

Scientists are actually preoccupied with accomplishment. So they are focused on whether they can do something. They never stop to ask if they should do something. They conveniently define such considerations as pointless. If they don’t do it, someone else will. Discovery, they believe, is inevitable. So they just try to do it first. That’s the game in science…

They have to leave their mark. They can’t just fit into the natural order. They have to make something unnatural happen. That is the scientist’s job, and now we have whole societies that try to be scientific…

It’s time for a change.

—

(It’s time to subscribe! Before it’s too late.)

Or just donate!

—

We have seen the discovery-at-all-costs dynamic Crichton suggested over and over — as when Dr. Anthony S. Fauci wrote in the Washington Post in 2011 that making viruses more dangerous was “a risk worth taking.”

But we have never seen them combined with financial and commercial pressures like the ones the AI companies face.

Even Crichton could not imagine the potential financial rewards that AI’s creators are chasing. In Jurassic Park, he thought, reasonably enough, in billions. The AI companies are racing each other — and Chinese companies we know even less about — for trillions.

The world needs a pause, a deep pause, to consider the risks and benefits of AI.

It’s not going to get one.

Buckle up.

1. Anthropic was founded in early 2021, no later than May, though I have not been able to pin down its formal corporate starting date.

2. “Compute” is shorthand for computing power, usually measured in floating point operations per second, or FLOPS.

Compute drives AI engines, to train them to crack and “learn” the huge corpuses of information that make up the Internet, libraries, and other repositories of human knowledge — and then to answer the questions and solve the problems that users pose them.

3. The New Yorker’s piece on Altman described him as “not a technical savant—according to many in his orbit, he lacks extensive expertise in coding or machine learning. Multiple engineers recalled him misusing or confusing basic technical terms.”