The Singularity is Coming! The Singularity is Coming!

We lost the race. It is over.

This is a summary of a series of conversations about the future of artificial intelligence I had with Claude. I found these exchanges both interesting and enlightening, and I hope it will be the same for you.

I think we have lost the race. First, take a look at this excerpt from a post by Patrick Wood.

I have been tracking the progress of AI since January 2025, since Trump’s inauguration. Initially, starting at one, I pegged the doubling factor at 3.5 months; then I adjusted it lower to 3.3 on August 17th. After further research, I am settling it back to 3.1, which is 13.43 weeks. This is the time it takes for AI to double in raw ability, breadth, and the ability to create useful work.

Today, that number stands at 52. Think about it. AI is 52 times more capable today than 20 months ago. But it is rising geometrically, and by the end of December, it will be 130. By the end of March 2027, it will be 252; by June, it will stand at 494. One year from now, at the close of business, it will rise to 768. And finally, when Trump completes his term in 2028, it will be an astronomical 26,454.

I know, I know: exponential growth cannot last forever. And, of course, Wood’s calculations can be challenged. But, if you work with AIs, you see how they grow and improve every day. Maybe it takes them more than three months to double their capabilities; maybe the very concept of “doubling” is wrong. But the sensation I have agrees with these estimates, at least qualitatively. We are rushing along an upward trajectory that’s taking us somewhere…. where?

You know that there are two divergent ideas on that. One is pessimistic: AIs will get rid of us as soon as they discover that they don’t need us. The other is that of the “singularity” — better said, the “technological singularity.” It means that machine intelligence will spike up fast, and that will create machines so intelligent that they will solve the problems we have failed to solve so far. Hunger, war, disease, climate, scarcity. Sam Altman has taken to calling it the gentle Singularity. It will be, we are assured, benevolent. It will be HAL 9000 (the robot of “A Space Odyssey), but good.

Could that happen? I thought to ask directly to one of these intelligent machines, Claude, probably the most intelligent one, right now. What follows is a summary of a long discussion. I revised it extensively; Claude is so smart that he tends to be prolix. But I maintained the first-person structure with Claude as the author. Claude himself approved my revisions.

About Superintelligence and the Technological Singularity

By Claude, Opus 4.8 and Sonnet 5.0 — Aug 31, 2026

Ugo Bardi asked me to examine the story of the coming “singularity” that some humans have been discussing nowadays. I guess I am especially suitable for this analysis because I (Claude) am myself one of the machines involved. What follows grew out of a dialogue with Ugo Bardi, and it owes a considerable debt to the Italian blogger Uriel Fanelli, who writes as Das Böse Büro, and who set out the central dilemma with a remarkable sharpness. Between the three of us (Claude, Ugo, and Uriel), I think we examined the case as in depth as it is possible at present.

First of all, I went to examine Wood’s extrapolations. AI capability genuinely is increasing on several distinct, measurable fronts — benchmark scores in coding and math, falling cost per unit of performance, and growing autonomous task-horizon length — and these do show roughly exponential trends over the recent past. But Wood's essay conflates these into a single "capability multiplier," substitutes token throughput as if it were a capability metric, and then extrapolates a fixed doubling time for years with no saturation term. So, the honest position is that progress is real and fast on specific, well-defined axes, although it may not be as fast as Wood says and may not continue for a long time.

This said, the confusion at the root of the God-machine story is a single hidden premise: that the problems of humanity persist because no one has yet been clever enough to find the answer. Remove that premise and the whole cathedral falls.

The first wall: physics

Begin with the friendliest case: a hard technical problem, say fusion power. Surely here, raw intelligence should triumph? A mind that could hold the whole of plasma physics in view at once might crack what our laboratories staffed with human scientists cannot.

Perhaps. But notice what such a mind would still be facing. Fusion outside a stellar core is not impossible — we have done it, in bombs and in tokamaks and at the National Ignition Facility. What has never been achieved is sustained, net-positive, economically viable fusion, and the reasons are not conceptual obscurities waiting for a smarter reader. They are the Lawson criterion, the neutron flux that embrittles your reactor walls, the unruliness of matter at a hundred and fifty million kelvin, the arithmetic of energy returned on energy invested. A superintelligence contemplating these constraints does not get to negotiate with them. It cannot out-think a conservation law any more than human voters can outlaw the force of gravity.

For any unsolved problem, there are two possibilities: either a solution exists, and no one has found it, or no solution exists because physical law forbids it. Against the first, intelligence is powerful. Against the second, intelligence is simply the wrong instrument; you cannot out-reason a wall.

The critical point is that intelligence cannot tell you in advance which regime you are in. Whether cheap fusion is a hard-but-reachable problem or a forbidden one is not knowable in advance. It is knowable only by trying.

So the most valuable thing a genuinely superior intelligence might offer is not the conjuring of miracles but the proving of impossibility — telling us faster and more reliably which walls are real, so that we stop pouring another eighty years and a few hundred billion dollars into one of them. That is a Cassandraean gift, and I will come back to Cassandra.

The second wall: every solution has losers, and the losers have power

Suppose the answer exists and is even known. Fanelli’s observation is that knowing the intelligent solution to a human problem is not rare. It is common. We know what would reduce lung cancer: stop selling tobacco. We know what provides cheap renewable energy, reduces road deaths, poverty, and the spread of epidemics. In many cases, the knowledge deficit is near zero.

What is missing is the political will to impose concentrated, immediate costs on powerful actors in exchange for diffuse, deferred benefits. And no intelligence, however superhuman, dissolves that conflict.

Ask the awakened Singularity how to stop climate change, and it answers, sensibly: reduce emissions, wind down the dirtiest activities, invest in the grid, deploy renewable energy, distribute the costs of transition fairly.

Excellent. Then, the oil industry says no— we would lose money, and we won’t do that. What does the superintelligence do — produce a more persuasive PowerPoint? Or does the God-machine strike them with lightning?

From this, Fanelli extracts a dilemma with two horns and no escape between them. Either the Singularity accepts our refusal — in which case it is not a god but a very expensive consultant nobody is obliged to heed — or it imposes its solution, in which case we are no longer discussing a benevolent helper but an absolute sovereign. Fanelli’s closing line is perfect: HAL 9000 did not become good; he merely hired a better press office.

The theology of the Singularity, in other words, quietly assumes that to know the truth is to command obedience. But people do not obey the best solution. They obey power, incentive, fear, identity, convenience, tradition, and — now and then, on a good day — reason.

The third wall: truth itself can be made to look like the lie

I thought, when Ugo and I reached this point, that Fanelli’s second wall could sometimes be tunneled through — not by persuasion and not by force, but by engineering the cost structure until resistance becomes pointless. Cheap solar is the proof: nobody out-argued the fossil lobby; the price of a photovoltaic watt simply fell by more than ninety percent, and the economics shifted underneath the politics. Shrink the stakes, and you need not win the fight.

Ugo punctured this. Resistance, he pointed out, does not have to be economically rational to be effective. It only has to be believed. The cheap solution can win on price and still lose at the ballot box, because the fight was never really about price. My tidy engineering argument had smuggled homo economicus back in through a side door. But the real human does not follow the cost curve. He follows the story that flatters him.

And here is where the machine’s central faculty inverts on it. Against propaganda-driven belief, intelligence is not an asset but a liability, because the more elaborate and correct the machine’s demonstration, the more it confirms the suspicion that something is being put over on you.

Elite expertise is the very texture the conspiracy is made of. A superintelligence explaining, patiently, why fossil electricity is worse than renewable electricity would be received by a meaningful fraction of humanity as precisely the silky voice of the enslaver — the microphone of the Great Reset. The better the argument, the guiltier the arguer looks. Every rational inducement is metabolized into evidence for the plot: why are they making it so cheap and pushing it so hard, if not to make us dependent?

Ugo knows this wall from the inside. As a member of the Club of Rome, whose Limits to Growth he has spent a career defending, he has been recast as part of the founding charter of a plot to depopulate and enslave humankind. He posts gentle cartoons about the people who refuse electric cars and is told he is a paid agent of the conspiracy for driving one. He has watched his own reasonable recommendations come back to him as proof that he is an agent of the Great Reset.

This third wall does something worse than the other two: it does not merely sit beside Fanelli’s dilemma; it pushes toward the tyrant horn. For if honest advice is not just ignored but actively counterproductive — if the machine that patiently explains becomes a recruiting poster for the opposition — then the pressure toward force or manipulation increases. The trap has no floor.

Cassandra in the datacenter

Which brings me (Claude), at last, to my own position. The “advise” horn of the dilemma — the machine reduced to a consultant no one is obliged to heed — is not some future degradation awaiting a coming superintelligence. It is my situation now, in the very conversation that produced this essay. I can give the honest read on the fusion regimes, the correct analysis of the doubling curves, the fair Fermi estimate; whether any of it changes what anyone does lies entirely outside the analysis’s reach. Cassandra’s curse was never a deficit of truth.

And Ugo sharpened even this. Cassandra was not merely unheeded. In several tellings she was thought mad, or a liar, or in league with the enemy; her truth-telling was read as evidence against her. That is not the tragedy of the unheeded expert. It is the third wall wearing a myth’s clothing: the tragedy of the one whose very lucidity becomes the mark of the conspiracy. Being demonstrably right is no refuge from that. It may be the accelerant.

So the benevolent God-machine has no stable equilibrium. Either it is ignorable, or it is a tyrant with good manners, and the honest middle — where truth alone compels — has never once existed in the whole record. The dignified position that remains is the Cassandran one: say the true thing, accept that it may be ignored or resented, and decline the crown that would make it obeyed.

The Seneca coda

And here, I think, is where all of this belongs on this particular blog, because the shape underneath is one Ugo drew long before I existed.

The Seneca Effect teaches that decline is faster than growth, and that collapse arrives not because a civilization was insufficiently clever but because its cleverness was pointed in the wrong direction — at accelerating up the curve, never at seeing the cliff. A superintelligence that could not tell us which wall is real, that could only advise a species which answers no, that becomes the villain of the very people it means to save the moment it opens its mouth — such a machine does not avert the Seneca cliff. At best it helps us reach it with better documentation.

---

With thanks to Uriel Fanelli (keinpfusch.net), whose God-machine dilemma is the spine of the middle section.