Thinker slot — The Penumbra, Week Four.
A sunflower head places each new floret about 137.5 degrees around from the one before. This is the golden angle, and it packs the most seeds into a disc with no gaps and no spokes. Nothing in the flower measures it. At the growing tip, a new floret forms where the hormone auxin has pooled, and as it grows it draws auxin away from its neighbours, so the next floret can only form in the widest gap left over. Repeat that local rule a few hundred times, across a wide range of growth conditions, and the spacing settles close to an irrational number.
David Deutsch has a plain description of what the flower is doing. Computation, in his account, is a physical process in which one object is made to mimic an abstract one, a number or a relation, so that its behaviour tracks the abstract thing's properties. By that description the growing tip qualifies, crudely: it tracks an angle, and nothing in it holds the angle as an idea. Whether that counts as a small amount of intelligence is not a question about sunflowers. It is a question about what you are claiming when you use the word — and that has been moving for four hundred years.
Thinker — Roger Penrose
The Dial
"Artificial intelligence" is not a thing that arrived. It is a setting on a dial, and the setting has been moved every time a machine reached it.
The phrase "artificial intelligence" was written down in 1955, in a proposal for a summer workshop held at Dartmouth College the next year. John McCarthy chose it to name a research programme, not to describe an event. The machines and the argument were already old: Pascal built a calculating machine in 1642, and in 1843 Ada Lovelace wrote that Babbage's proposed engine "has no pretensions to originate anything" — a position on this dial, stated a century before anyone could test it.
In 1950 Alan Turing declined to answer "can machines think" and replaced it with a game: could a person telling the difference, by typed conversation alone, do better than chance. That was not a discovery. It was a decision about where to put the mark — trading a question nobody could settle for one anybody could run.
In 1966 a program called ELIZA rephrased what you typed as a question, and people confided in it. In 1997 Deep Blue beat the world chess champion. In 2012 a network named AlexNet sorted photographs; in 2016 another won at Go. Each time, the thing that had been the mark of intelligence became, within a year or two, "just computation." McCarthy complained about it himself. Larry Tesler's version is the one that stuck: intelligence is whatever machines haven't done yet.
What moves is not the machine. It is what you are committing to when you use the word. Say a machine is intelligent and you have claimed something specific — that it does arithmetic unattended, or learns from examples, or holds a conversation, or understands. Those are different notches and they are not close together. Almost every argument about AI is two people reading the dial at different settings and thinking they disagree about the machine.
This is why Penrose is worth the hour. In 1989 he aimed at one notch and named it: "strong AI," John Searle's term from 1980 for the claim that running the right program simply is thinking. Penrose argued that notch cannot be reached by any program at all. Whether his argument works is a separate question, taken up in the piece above. What he did that almost nobody does is say exactly which setting he was talking about.
Ten thousand software agents produced a proof of something mathematicians had not proved, and the proof checked out. That is a real notch, newly reached, and it is not the top of the dial. It sits above "wins at Go" and below "understands," and the distance to the notch above it is not measured in months of engineering, because nobody has an instrument for it.
The honest way to use the word is with a number attached: which notch do you mean. Everything below your setting is machinery that already exists and mostly bores us. Everything above it is a promise. The dial has no top mark, which is the part that gets sold, and the same dial has been running since 1642, which is the part that doesn't.
Penrose planted his flag at the top notch and argued that no program reaches it. The argument is a single move, repeated, and it is easier to watch than to read. Follow the line below through six turns, and watch step five, because that is where the whole thing rests on a premise nobody has established.
Thinker — Roger Penrose
The Thread and the Frame
One argument, drawn as one line. Every system that can do arithmetic has a sentence it cannot prove. Step outside to see that the sentence is true, and you are standing in a larger system — which has one of its own.
Take a formal system strong enough for arithmetic: a fixed set of starting statements and rules for getting from one statement to the next. The thread is what the rules can reach — every proof the system can run, laid end to end. The frame is the system's edge.
Gödel showed in 1931 that if the system never proves both a statement and its opposite, it contains a sentence it cannot prove — one that says, in effect, "I cannot be proved here." It sits just outside the thread's reach. Not far away. Just out.
A mathematician follows the construction and sees that the sentence is true. Add it as a new starting statement and the frame redraws: the old sentence is now inside, provable, ordinary. The thread has not been replaced. It is the same thread, further along.
The larger system is a system, so Gödel's result applies to it too. A new sentence sits outside the new frame. Step out, redraw, repeat. The picture zooms out for ever and the thread never doubles back; nothing is added at any turn except more of the same thread.
The argument that this proves minds are not computers needs one more thing: that we know our own reasoning never contradicts itself. Feferman, Putnam and most logicians say we cannot know that. In 1902 Russell wrote to Frege to point out a contradiction sitting inside the system Frege had spent years building, and Frege had not seen it.
Penrose needs the stepping-out to be something no procedure could do — which requires a frame with nothing outside it. Deutsch's reply is that each step is itself a physical process running in a brain, so every frame a mind builds is open to the same construction. No one has drawn the last frame. The diagram is the argument: the question is whether it ever ends.
So the escape never terminates, and Penrose needs it to. His case does not fail because he was careless — he was the most careful person to make it, and he said out loud which setting he meant, which is more than his opponents usually manage. It fails as a proof, at one premise, and leaves the question open rather than closed.
The computational view has not won everything either. It explains what brains do. It has not explained why any of it is felt, and Penrose's argument, even where it fails, marks precisely where that burden sits. This week's news shows the layering in practice. A machine search produced a Navier–Stokes blowup, another machine checked the proof in Lean, and human mathematicians are still deciding whether the formal statement matches the problem the Clay Institute posed. Each check stands one frame outside the last.
Go back to the sunflower. Its growing tip tracks the golden angle without holding it as an idea, and the mathematician holds it as an idea. Penrose says that difference is one of kind. The computational view says it is a difference in how many frames a system can build and then inspect from outside. The evidence so far favours the second without settling it, and it suggests that what Gödel drew was never a wall around the human mind but the outline every frame has, the human one included. For the full exchange, search the Lucas–Penrose argument; Feferman's reply is the place to start.
source list:Phyllotaxis: auxin mechanism, Reinhardt et al., Nature 2003; golden angle from a local inhibition rule, Douady & Couder, PRL 1992.Deutsch on computation as mimicry The Fabric of Reality ch. 10Lucas 1961 (Philosophy); Putnam, NYT Book Review, Nov 1994; Feferman, Psyche, 1995; Turing 1950 (Mind, "The Mathematical Objection"); Russell–Frege letter, June 1902; Tegmark, Phys. Rev. E, 2000; Penrose's Nobel 2020; born Aug 8, 1931. All need primary-source checks.