The events of the past week are well known.
It all started with Jacob Coxon, a researcher who worked for Anthropic after previously having been employed by OpenAI, and who stated that he had resigned because ‘the people building AI are genuinely convinced that it could kill us all by the end of the decade’.

Evan Hubinger echoed his words. Whilst holding on tightly to his job at Anthropic, he also put a figure on it: “more than a 10% chance”, specifying that the risk does not lie in models already on the market, but in the “recursive self-improvement” that cutting-edge models are reportedly acquiring.

Shortly afterwards, with Hollywood-style timing, the company’s founder and CEO, Dario Amodei, stepped into the fray with an essay entitled “We Must Slow Down the Frontier”.

His move drew thunderous applause from Sam Altman (the father of ChatGPT) and Elon Musk (the creator of Grok), whilst it was met with scepticism from Trump and the usual, bitter hostility from the Chinese regime.

The end of the world or the end of business?

Amodei’s essay has been discussed at length, but only insofar as it puts forward practical proposals.
The billionaire, alarmed by recent incidents (such as a swarm of AI agents that turned into a collective of hackers), fears that the ability of models to develop other models themselves without human oversight could cause the internet to fall into the hands of AI in less than a year, perhaps even in less than six months.

He therefore proposes a three-stage plan: first, to incorporate independent overseers into the very teams developing the new models; secondly, to forge a non-aggression pact amongst tech companies in democratic countries; thirdly, to attempt to negotiate a multilateral agreement with China to prevent it from exploiting the West’s slowdown to its own advantage.

Amodei is convinced that, in the time gained, companies will improve their systems forinterpreting how AIs behave (which, he says, have already made significant progress over the last two years) and will increase the testing required before their release.

Accusations of hypocrisy regarding these proposals have come thick and fast from all sides: not only from Beijing, but also from those in the West who have seen them as a corporate pact between the AI giants to lock down the market and dictate the agenda of the institutions.

Furthermore, various detractors claim that Amodei and his peers have realised that none of their future models, however powerful, will ever be able to recoup the trillions in investment they have attracted in recent years, and that they are using the excuse of security to restructure their business.

What if the problem were the finish line?

Be that as it may, hardly anyone has paid much attention tothe opening lines of the essay, which, in my view, are actually the most interesting part.
Here, Amodei, covering his bases, reiterated to investors and the world the messianic promises that have so far guided the AI industry:

“I believe that AI could cure most major diseases in the next 5–10 years, greatly accelerate economic growth rates, create a world of abundance and empowerment, and usher in a renaissance of democracy and freedom”.

Amodei has not backed down from this mythical dream in the slightest. On the contrary, he goes even further by laying bare his innermost feelings:
“I feel the urgency personally. My own father died of a disease that was cured just a few years after his death, and I myself survived an early-stage cancer that would not have been treatable even fifty years ago. Carefully wielded, AI can be the latest in a long line of technological miracles that have uplifted and ennobled humanity”.

If, whilst reading these words, you feel a suspicious itch under your skin, you are not alone.

If you harbour the sinister feeling that the problem with the AI industry lies precisely in having justified its very existence with this pseudo-religious drivel – whether for cold profit (which would be excusable) or out of fervent conviction (which would be tragic) – you are in good company.

And if you’ve heard echoes in your memory of ancient – and similarly outlandish – texts studied at school, such as Pico della Mirandola’s De hominis dignitate or Francis Bacon’s New Atlantis, in my view you’re not far off the mark.

The AI tycoons, in fact, are falling headlong into the very same mistake that, for more than a millennium, has derailed and compromised the research efforts of another group: that of the alchemists.

Spondent quas non exhibent

In the history of science, alchemy was not merely a marginal curiosity.
As soon as humankind learnt the art of metallurgy, it was immediately captivated by the idea that metals transmute into one another because, at heart, they are living beings.

From this notion stems another: that nature is endowed with a soul, usually divided into two opposing principles (male-female, sun-moon) which are destined to be reunited.

The ultimate corollary is that, by practising alchemy, humankind can become the protagonist of this great reunification. The creation of the philosopher’s stone, which crowns the ‘fourfold Work’, will provide humankind with infinite gold and the elixir of life, regenerating the world into a new era.

This web of magical and prophetic illusions remained, in fact, inseparable from any experimentation with metals: first in China and Egypt, then in Persia and the Greek world, and finally in the West.
The consequence is clear: the finest talents of the privileged few who had access to alchemical laboratories were largely wasted.

Until the end of the Middle Ages, progress in the field we would now call ‘chemistry’ was limited to the isolation of sulphur and mercury, the distillation of pure alcohol and the production of gunpowder.

Moreover, even scholars investigating other aspects of nature – such as astronomers – squandered part of their time and resources in the quest for the philosopher’s stone. Even Newton wrote more books on alchemy than on astronomy. The damage, in short, was also indirect.

To make matters worse, the ecclesiastical authorities, usually generous in funding research in other fields (including astronomy, which confessors had sought for centuries to purge of astrology), found alchemy to be a perpetual headache.
On the one hand, it was objectively a hodgepodge of fanciful superstitions, with heretical undertones to boot, given that it promised an earthly redemption as an alternative to the Christian one.
(A decree of condemnation from the early 14th century began with words that could not have been more explicit: Spondent quas non exhibent, ‘They promise things they do not deliver’).

On the other hand, it was not necessarily without merit.

Thus, particularly during the Renaissance, bans and persecutions gave way to an attempt to incorporate it into the Christian worldview. Cardinal Nicola Cusano, Don Marsilio Ficino and Father Atanasius Kirchner were merely the most famous of those who undertook this endeavour.
It was not until the early 18th century that chemistry broke free from alchemical myth and developed its own independent method, thanks once again to a devout and pragmatic Christian such as Boyle.

After all, the problem was no longer merely that the philosopher’s stone was a hoax and that the process of creating it was a load of nonsense. Following the conquest of the Americas, in Spain and elsewhere people were beginning, to their dismay, to realise that, even if humanity were to obtain an infinite supply of gold, this would, at best, merely increase inflation or render gold worthless.

Has the break finished?

Now, the only lesson that history teaches us is that things are always more dissimilar than they are similar, that a past event always has more differences than similarities with any present event, and that nothing ever repeats itself. That said, the philosophy behind the AI industry really does seem like a revival of millennia-old alchemical beliefs, following a brief interval during which they had lain dormant.

The very name ‘artificial intelligence’ evokes the myth of living metals endowed with a soul.
And it is a name wrongly attributed to systems which, however powerful, are merely computational tools that link strings of code together according to pre-established vectors.

It is clear, however, that calling them ‘linguistic combinators’ or by other, more dry-sounding names would have attracted less media hype and less investment.

Furthermore, the ultimate goal of Amodei and his colleagues is to create what they call ‘superintelligence’, ‘general artificial intelligence’ or ‘technological singularity’: a model so capable and autonomous in generating further models that it can manage entire sectors of research, the economy and social life (with healthcare at the forefront) more effectively than humans can.

It is hard not to see this as version 2.0 of the philosopher’s stone.
And it is hard not to think that this version 2.0 will exhibit the same flaws as the original: either it will be unachievable, or it will be achievable but with effects different from those hoped for, or it will be achievable but with costs outweighing the benefits.

Whilst I have every sympathy for Amodei’s father and for the suffering his son has endured, AI will not ‘be able to cure all the most serious diseases’. It will be able to devise new treatments in the abstract, which will subsequently need to be tested, produced, sold and administered, coming up against the normal laws of economics.

The same applies to the ‘growth rates’ that AI is supposed to deliver: helping which companies to provide which goods or services, which will be priced at what level by which consumers?
Does it make sense to pin trillions on such an abstract promise?

Let’s gloss over the utopia of “emancipation”, “freedom” and “democracy”: I’d like to believe that the automation of mental tasks could foster these, but the data from recent decades unfortunately seem to suggest the opposite. What counter-trend data does our Promethean CEO present? Nothing more concrete than the union of the Sun and the Moon to produce the androgynous prince.

Satisfying investors who have backed such a business plan is, if not impossible, at the very least unlikely.

Today, as in the past, spondent quas non exhibent.

To err is human; to persist is diabolical

My impression is that the AI industry, if it is not to collapse under the weight of its own contradictions, must lower its sights and clear the field of these oddities.

Just as chemistry became far more useful to humankind once it had done away with alchemy, AI technology will be at its best when it devotes itself, with humility and practicality, to solving specific problems faced by individual companies or institutions.
(Which, incidentally, is the European and Indian approach).

The race for the philosopher’s stone – which America needs to run even if it does not want to, and China wants to run even if it does not need to – is a disaster in its own right, not just because of its excessive pace.

We have already seen with alchemy what happens when technological research is overburdened with mythical and religious expectations, and perhaps it is best not to repeat that mistake.