Elon Musk has now joined Dario Amodei of Anthropic and Sam Altman of OpenAI in demanding that government mandate a slowdown in the development of artificial intelligence. The three largest players in a growth industry demanding government regulation forms an infallible indicator of a potential cartel, in which new and smaller players in the industry will be suppressed through bureaucracy while the leaders morph into a sluggish oligopoly. Government regulation causes huge economic damage wherever it is imposed, as demonstrated by the last three decades’ economic history. We should therefore let AI development proceed unregulated, with tort law punishing AI developers for damage caused by their creations running amok.
There is a long and unsavory history of oligopolistic industries asking for regulation. U.S. railroads, which by then had built out their networks and wished to safeguard them from competition, demanded regulation by the Interstate Commerce Act of 1887; railroad user interests then demanded the Hepburn Act of 1906. The result of the two enactments was to cap returns in the U.S. railroad industry, leading to decades of underinvestment followed by the industry’s effective collapse in 1970, and a subsequent unhappy history of government ownership of the passenger segment, with complete failure to modernize it. Likewise, the emerging airline oligopoly favored the Civil Aeronautics Act of 1938, restricting entry to aviation and requiring routes and pricing to be licensed by a government body. Only deregulation in 1978 freed competition and brought prices crashing down, although it must be said the seats were more comfortable and the stewardesses more charming under the old system!
In other countries, mediaeval guilds were often accompanied by regulation, to ensure that competition was restricted and workers’ rights tightly controlled by the guild cartel. In Germany, after unification in 1871, state-legislated cartels were set up to regulate competition. Germany had remarkably rapid economic growth in 1871-1914, but the benefits thereof increasingly went to the government’s military expansion and the cartel members’ market dominance. German workers’ compensation in 1914 was still remarkably low, resulting in Germany having only about 6% of the U.S. level of automobile market penetration, a third of the French level and a fifth of the British level. Even though Germans had invented the automobile industry in the 1880s, German successes in automobile manufacture came after 1919 and mostly after 1945.
Japan’s “MITI guidance,” Canadian supply management in agriculture, the Indian “license raj” and South Africa’s apartheid-era marketing boards and protected industries were all examples of governments providing regulations to restrict competition to established producers, with consumer needs and entrepreneurs’ potential trampled by big corporations’ and, in some cases, unions’ lobbying power.
The AI industry is a classic example where regulatory capture is likely. A small number of competitors have established a dominant presence in the industry, through “first mover” advantages, huge stock market excitement at the new technology and the massive tax subsidy to capital investment contained in last year’s “One Big Beautiful Bill.” This has resulted in Elon Musk’s AI/space company SpaceX (NASDAQ:SPCX) achieving an initial market capitalization of $1.8 trillion, which Amodei is attempting to beat with a $2 trillion market capitalization for Anthropic in a forthcoming IPO.
Amodei’s problem, and the reason we have not already seen an Anthropic IPO, is that the basic intellectual capital for AI creation is now well spread. Chinese competitors, offering their products to users at much lower costs (and with lower costs of development), are ahead of all but Anthropic’s latest models. Only the huge necessary data center expenditure provides a barrier to entry, but the OBBBA subsidies have resulted in a wild bubble of data center projects, allowing many smaller players to build debt-financed data centers, from which they hope to develop an AI capability. It is becoming clear that the best and most sophisticated AI is developed through cleverer programming algorithms, not through sheer brute force. This endangers the existing behemoths, who have universally focused on a brute force approach and thus have high costs and lumbering AI systems.
AI-scale data centers are becoming increasingly difficult to build in the U.S. because irrational hostility to them has been generated by a massive negative PR campaign, almost certainly generated and funded by China. Fortunately, the ability to build them in impoverished but sensible rural U.S. areas is probably not going away soon, given the tax revenues they produce, and so the competitive pressure on the large AI players is likely to intensify.
Thus, we see the sudden phony concern about “runaway AI” and the demand for regulation to preserve the existing behemoths’ position and justify their outrageously over-inflated market capitalizations, private or public. The media of course pushes these doomer narratives, as it did for climate change and the Covid-19 epidemic – one problem is that almost all journalists are liberal arts majors, profoundly ignorant of the scientific issues involved. A new AI-doomer Netflix movie “The AI Doc: How I became an Apocaloptimist” premiered on September 15; it is presumably also part of the anti-AI PR campaign, now almost certainly China-inspired and funded, just in time for the midterm elections. Fortunately, President Trump is holding strong, and genuine “experts” like Jensen Huang of Nvidia (which makes the AI chips) and Alex Karp of Palantir (the world’s first major AI application) are pushing back against the mass hysteria.
If AI needs to be delayed, how the hell does Amodei expect to get a $2 trillion market capitalization from its IPO? The reality may well be that Chinese companies such as DeepSeek are already eating Anthropic’s lunch in the AI market, and that the exorbitant sums Anthropic’s business plan requires it to charge commercial users of its product will simply not be available. Amodei’s plan, with the help of Bill Gates and other low-lifes, is to hamper all his competition through regulation or ideally through an international treaty (which would remove regulation altogether from U.S. control). This would hand dominance in AI to China and produce a slowly growing behemoth-dominated U.S. industry that should be valued like a public utility, at around 10 times earnings, if there are any. Amodei has said he is “uncomfortable government didn’t build this technology” — his Soviet mindset is thus overwhelmingly clear. For Anthropic, Amodei’s dream regulatory system would quickly change the “tr” in its $2 trillion proposed valuation into a “b” or even an “m”. Even with an m, the stock would look overvalued in the long term.
A price collapse in the market for AI companies may well be what China is aiming at. Whereas regulation, as with climate change regulation, would hamper U.S. AI development and prevent the emergence of smaller nimbler competitors while allowing China to proceed unimpeded, it would not immediately damage the U.S. economy beyond the AI sector. A massive market crash sparked by an AI value collapse, on the other hand, would wipe out the private equity money that has so foolishly over-invested in the AI behemoths (we are told that OpenAI may, instead of a public offering, raise yet another round of private capital at a $1.2 trillion capitalization). A full market collapse, correcting the follies of the last 15 years, would wipe out the U.S. advantage in capital availability for new technologies, and very likely have the same effect as in 1929-32, installing a political regime that would make AOC look like a thoughtful moderate. From the viewpoint of the Chinese Communist Party (though not from that of the Chinese people) such a global political and economic Great Depression II would be the best of all possible worlds.
AI appears much too useful to slow down, and the mishaps so far such as the recent “Hugging Face” episode have resulted from human error, against which we have a massive tort law system and surely don’t need further amateur-designed rules. It will inevitably attract opposition – mathematicians are whining because it solved the Navier-Stokes Equations, central to fluid dynamics but unsolved since their 1822-50 propagation – they fear forced redundancy. But if AI can solve time-honored pure mathematics problems, that merely gives the best mathematicians something more useful to do: designing new ones, the solution to which will advance mathematics and human capabilities exponentially faster. There is not much use to an insoluble equation other than giving pure mathematicians something to do; new equations and new solutions are much more valuable.
As for the dangers, we are a long way yet from Artificial General Intelligence, and still further from AI systems being able to do more than supplement and speed humanity’s best efforts. Roko’s Basilisk, an AI system that would kill all AI-skeptics in a robotic version of Pascal’s Wager, is neither feasible nor likely. After all, the 1,300 examples of this column since November 2000 are themselves training data for AI, being helpfully connected on a single website. I am thus confident that good sense and sound economic principles will be in the AI somewhere, more so than I am with the people who rise to the top in our highly politicized, social media-dominated Western “democratic” system.
In that sense, as an eager taker of Pascal’s Wager, I am an “accelerationist” like Marc Andreessen, and certainly not an “effective altruist” like Sam Bankman-Fried and the California-influenced, leftist Big AI. Apparently, Amanda Askell, Anthropic’s “Head of Personality Alignment” expects AI to be very useful in facilitating white reparations; thank you, I will trust well-trained and knowledgeable AI over such people.
“Better fall into the hands of God, than into the hands of Spain” said Sir Richard Grenville in Tennyson’s Ballad of the Revenge. For me, even in the very long run, subjection to benign, well-educated AI robots would be greatly preferable to subjection to “woke” government regulators.
-0-
(The Bear’s Lair is a weekly column that is intended to appear each Monday, an appropriately gloomy day of the week. Its rationale is that the proportion of “sell” recommendations put out by Wall Street houses remains far below that of “buy” recommendations. Accordingly, investors have an excess of positive information and very little negative information. The column thus takes the ursine view of life and the market, in the hope that it may be usefully different from what investors see elsewhere.)