The Photodiode and You

Same job, different universe

You tied. On the face of it you and a four-cent component just did identical work: same screen, same answers, same score. That impression is wrong, and why it's wrong is the whole of this room.

The photodiode has two possible experiences. Not two that it can currently distinguish, with better wiring to come -- two, in total, for the whole of its existence. "Light or dark" isn't a simplified version of what you were looking at. It is the complete contents of that thing's world, and it was in one half of it.

Your panel could have shown any of what you just watched it show. And that grid is a joke of a visual field -- twelve squares by twelve, sixteen colours. Even so it has 16¹⁴⁴ possible arrangements, about 10¹⁷³. There are something like 10⁸⁰ atoms in the observable universe. So when the diode said "light," it was choosing one thing out of two. When you said "light," you were choosing one thing out of that -- and your real visual field is enormously bigger than the toy above. You did it instantly, three times, without noticing you had done anything at all.

That is the first half of integrated information theory. What matters isn't how much information a thing handles. It's how large a space of possibilities each single moment gets picked out of -- what Tononi calls the system's repertoire. Your score and the diode's score were the same number. The two numbers were measuring different universes. The photodiode is Tononi's own example, not ours.

Now for the second half, which is where it gets strange.

It was never one thing

Nothing happened. Each half carries on exactly as before, reporting its own bits, entirely unbothered. You can cut it again, and again, down to sixteen lonely diodes in sixteen separate boxes, and at no point does anything break -- because there was never one system there to break. There were sixteen systems sitting near each other in a nice case.

Now try it on yourself. Split your visual field down the middle so that the left half is one experience and the right half is another, belonging to somebody else. You can't, and not because you lack a knife. Your experience arrives already whole, and it turns out you have no way to hand back half of it.

This is the other half of the theory: integration. A conscious system, IIT says, is one whose cause-effect power over itself is irreducible -- cut it anywhere and you destroy something that the parts did not separately have. The measure of that irreducibility is called Φ, phi. The camera has sixteen bits and Φ of zero. You have far fewer bits under conscious control at any instant, and Φ well above zero.

If you want the version of this done with a knife rather than a metaphor, the museum has a room for it: Exhibit № 17, where surgeons really did cut the connection and something really did divide.

The claim that makes people angry

Those two wirings can compute exactly the same function. Same inputs, same outputs, same behaviour, indistinguishable from outside. And integrated information theory says the first one has Φ of precisely zero -- that a purely feed-forward system is not a little bit conscious, or dimly conscious, but not conscious at all, however clever its behaviour. The proof is technical. The consequence isn't: build something that behaves exactly like you, feed-forward, and on this theory you have built a philosophical zombie on purpose. You have already met one of those in Exhibit № 24.

That is not a footnote in IIT. It's the point. The theory refuses to let behaviour be the criterion -- it insists on looking at the causal structure of the thing itself, which is why it can say a simulation of a brain running on conventional hardware would have almost no Φ, and would be nobody, no matter how convincingly it wept.

And then the grid

In 2014 the computer scientist Scott Aaronson worked out what the formula actually implies and found something awkward: certain large, extremely dull networks -- regular grids of simple elements, doing nothing anyone would call thinking -- come out with astronomically high Φ. Higher than a human being. He offered this as a reductio: a theory that says a lattice is more conscious than you are has refuted itself.

Tononi's reply was not to patch the formula. It was to accept it. Yes, he said, that grid would be conscious -- very conscious, in a way that involves no thought, no behaviour, and nothing we would recognise. Our intuition that consciousness requires cleverness is the thing that should give way.

You may find that magnificent or ridiculous. Both reactions are widely held by people who have thought about it much longer than either of us. In September 2023 more than a hundred researchers signed an open letter calling IIT pseudoscience, which caused a considerable row -- including from people who dislike the theory but thought the letter was a bad way to argue. The row is still going.

Why it's here anyway

Because it is the only major theory of consciousness that starts from the right end. Almost everyone else starts with the brain and works towards experience. IIT starts by asking what experience is undeniably like -- that it exists, that it's structured, that it's specific, that it's unified, that it's definite -- and then asks what a physical system would have to be like to have those properties. Whether or not the answer is right, the question is put the correct way round, and that is rarer than it should be.

It has also produced the single most clinically useful thing to come out of consciousness science: a way of measuring how integrated a brain's response is, by pinging the cortex with a magnetic pulse and measuring how complex the echo is. It reliably distinguishes wakefulness, sleep and anaesthesia -- and it can detect covert awareness in patients who look entirely unresponsive. Whatever Φ turns out to be, something in that neighbourhood is real enough to find people nobody knew were still in there. The museum's Exhibit № 20 is where that machinery lives.

And down the corridor, its rival: Exhibit № 28 says consciousness is what gets broadcast. This room says broadcasting has nothing to do with it, and what matters is whether the thing can be cut. In 2023 both were put in a ring together, with each side agreeing beforehand what would count as losing. Both came out bruised. Neither came out dead. That is the honest state of the field, and it is why the needle sits where it does.

Take it further

Tononi's Phi: A Voyage from the Brain to the Soul is the strange, lovely, illustrated version; the formal statement lives under integrated information theory. Scott Aaronson's blog post "Why I Am Not An Integrated Information Theorist" is the best short demolition attempt and is very funny, and Tononi's reply is appended to it -- a genuine public argument between two people who both understood the maths, which is worth reading purely as a spectacle.