Can we build a consciousness-measuring machine?
I spent a chunk of 2022 thinking about consciousness: whether it is possible to measure it, and if so how. I am extremely grateful to Caslav Brukner, first of all, who entertained me on multiple occasions. In addition, through these wanderings, I managed to connect Caslav and Will Zeng, who are now pursuing together some very exciting ideas in quantum foundations.
I came to the conclusion that:
There might be some potentially compelling connection between consciousness and quantum mechanics, vis a vis Wigner's Friend (and no, it has nothing to do with microtubules).
There is a ton of work to be done on quantum foundations, especially to explore the interface between the classical and the quantum. (This is the direction Will and Caslav are now pursuing very seriously, see wignersfriends.com.)
We are still far away from having a handle on the Hard Problem of Consciousness, and it's unclear whether the Hard Problem is even solvable in principle. It might be, or it might not be.
This blog post above tries to compile the various trains of thought I pursued. These are the idle wanderings of an amateur -- I don't claim that any of it is original or that any of it is correct, but I hope it might be interesting and inspiring for others.
All images are DALL-E-generated representations of the prompt "consciousness" in various styles.
We need a way to measure consciousness
This post will be slightly crazy, so I apologize in advance. But I think these things are important to think about. It is my current belief that (1) consciousness is a real thing, (2) that it cannot be explained in terms of atoms and forces, and (3) that it therefore necessitates new physics . However, we can’t come up with any theories of consciousness unless we have a way to measure it. At the moment, we have no way to measure whether a particular composition of matter contains consciousness, or how much consciousness it contains. We can assert that we are individually conscious, but have no way to validate that other humans are also conscious. (This is the “hard problem of consciousness.”) So the subject of this blog post is: how might we go about measuring consciousness?
As I will discuss below, the Wigner’s Friend paradox in physics might actually provide a strategy for measuring whether a given composition of matter contains consciousness. (It also might not.) It is probably worth taking seriously, simply because it’s the only reasonable approach I have heard. First a few disclaimers.
One: This is not a piece of scholarly work. This is a blog post with some ideas. Moreover, I am not a philosopher. Philosophers who have thought about this may find the following post excruciating. I am simply a scientist, applying the tools I have to think about something that might (or might not) be a physical phenomenon.
Two: Most of this post is not novel. It is inspired by the work of Eugene Wigner, David Deutsch, and Caslav Brukner, the latter of whom was kind enough to spend several hours talking to me about Wigner’s friend and enlightening me. My only contribution to this is suggesting that it might be worth taking Wigner’s friend seriously as a consciousness measuring technique.
Three: Our consciousness measuring device should not be dependent on self-reporting, because philosophical zombies (explained below) will always report that they are conscious. Moreover, a lack of consciousness can be explained by deficits in other brain functions, such as memory. How are we certain, for example, that we are not fully conscious during sleep, with full amnesia, and simply rediscover (and then forget) every night the fact that we are actually conscious while we sleep ?
What is consciousness?
I really don’t want to get dragged into this rabbit hole. But it’s important to clarify a few things:
Firstly, I am talking specifically about the hard problem of consciousness, i.e., why and how humans have qualia. This is distinct from awareness, i.e. whether an organism can perceive stimuli or can perceive its own internal state. For example, if you show an animal a stimulus, and then prove by looking at its brain activity that it observed the stimulus, that does not prove that the animal is conscious (i.e. that it has qualia); it only proves that the animal perceived the stimulus. Similarly, computers are not necessarily conscious just because you hook them up to video cameras.
Secondly, it is not even obvious whether consciousness is a thing that exists and that can be measured independently of other information-processing capabilities of the brain. A lot of work in philosophy has gone into thinking about this. I have a few paragraphs on this at the end. Let’s just say for now: we want to be able to tell whether the large language models are conscious. We can characterize their information processing capabilities just fine, but we need to know if they are sentient .
We might be able to use Wigner’s friend as a consciousness measurement strategy
In quantum mechanics, physical entities exist in superpositions of many observable states, and can change their state in one of two ways: they can either evolve in “unitary” manner, or they can evolve in a “projective” manner. Quantum mechanics gives no guidance as to the conditions under which an entity should evolve unitarily or projectively. The only thing we know is that whenever we as humans observe something, the system evolves in a projective manner, and whenever particles are isolated and left on their own, they evolve in a unitary manner. I.e., the microscopic world may be quantum, but our experience of the world is somehow realistic. Explaining the division between unitary and projective measurements, and what causes something to be unitary versus projective, is called the Measurement Problem.
In 1961, Eugene Wigner proposed a paradox. Suppose you isolate a friend from the world and then allow the friend to measure a particle that is in a superposition of the states |0> and |1>. In this case, according to formal quantum mechanics, you should describe the resulting state of the particle + friend as a superposition of two states: one in which the particle is in state 0 and the friend has measured 0 (call this state |00>) and one in which the particle is in state 1 and the friend has measured 1 (call this state |11>). The resulting state is a Bell state: formally, in quantum mechanics, the friend has become entangled with the particle. On the other hand, the friend would clearly measure the particle to be in either state |0> or state |1>, so the friend exists either in state |11> or state |00>. So which is it? The key is, you can in principle actually perform an interferometric measurement on the friend, to figure out if they are really entangled with the particle or not. There are two possible options: either they are not entangled, in which case something caused the wavefunction to collapse, or they are entangled, in which case your observations of the physical universe disagree with those of your friend, and objective reality does not exist.
Let’s imagine that the friend and the particle are not entangled, and let’s ignore, for a moment, the possible outcome where we discover that in fact objective reality does not exist . Then the question is: what collapsed the wavefunction? There have been many arguments about this: some people think the mass of the friend will collapse the wavefunction, for example, a set of theories called objective reduction. But it is also possible that consciousness collapses the wavefunction, i.e., we might even be able to define consciousness as a property of matter that collapses wavefunctions. You can then imagine that, if we could build a Wigner’s friend box, we could actually put different objects in it -- rocks, bacteria, frogs, anesthetized humans, awake humans, superintelligent AIs -- to determine which are conscious and which are not.
This might not be as crazy as it sounds. In a world without conscious observers, there would be no need for projective measurements. Everything would just become entangled with everything else, happily, and there would be no measurement problem because there would be no observers. Quantum mechanics would be fully internally consistent. The whole reason we have a measurement problem is because we have consciousness in the universe, and consciousness is somehow incompatible with that otherwise internally consistent version of the universe.
We need to work on Wigner’s friend
Unfortunately, there are a lot of “ifs” between us and actually building a Wigner’s Friend device to measure consciousness. As noted above, it’s possible that even if we had such a device, that the friend would become entangled with the particle and we would simply have to discard objective reality as a concept, and would be no closer to measuring consciousness. It’s also possible that something else, like mass, forces the wavefunction to collapse. Finally, even if we had such a device, we would never be able to tell that consciousness itself is the thing that is collapsing the wavefunction. (See my notes at the bottom on how we would know that the device is measuring consciousness.) Even with the consciousness measuring device, we would need selective ways to extinguish consciousness (hopefully reversibly) without affecting other brain functions in order to prove to ourselves that consciousness is in fact the thing being perturbed, and we have no such ability. (For a much more thorough and robust discussion of Wigner's Friend, see https://www.wignersfriends.com/, by Will Zeng)
The biggest issue, though, is that it is almost certainly not possible to implement a Wigner’s friend device and apply it to humans. There are two challenges, which are related: firstly, the box must be perfectly insulated from the environment, so the friend does not become entangled with other objects outside of the box. Secondly, inside the box we must somehow perform a reversible operation on the friend. Isolating the box at least sounds conceivable; performing thermodynamically reversible operations on humans, though? Good luck! It is not an exaggeration to say that the problem is equivalent to building a time machine. Not actually strictly impossible, but it might as well be.
One obvious objection might be: why not just use a GHZ state? Suppose we entangled Particle X with the Friend and with a second particle, Particle Y. Then we could simply interfere particles X and Y to determine whether the friend had collapsed the wavefunction, without explicitly interfering the friend, right? Sadly, no. As my good friend David Turban pointed out to me, the reduced density matrix of a GHZ state is diagonal. If you take the friend out of the picture, the remaining two particles will appear unentangled .
So what could we do? It would be more straightforward, at least, to run a Wigner’s friend experiment on an ultraintelligent AI running on a quantum computer, although it would be hard to interpret the results because we wouldn’t necessarily have any expectations a priori about whether the quantum AI was conscious. More interestingly, it might be possible to build variants of the Wigner’s friend experiment that get us closer to “human-like pointer states,” without actually putting a human into the box. In quantum mechanics, we only ever consider fully unitary operations or fully projective operations, but what would it take to build a device that can be adjusted continuously between a well-behaved, particle-like state, and a classical pointer-like state? In November, there was a workshop hosted by Will Zeng on the topic of what a Focused Research Organization might look like that would try to build measurements that are halfway between “classical” and “quantum.” It is worth paying close attention -- it is the only reasonable shot I think we have for getting close to measuring consciousness.
How would we know that the device is measuring consciousness?
Now we get into the hard part. Suppose we have a consciousness measuring box. How would we know that it is actually measuring consciousness?
One of the tricky things about consciousness is that we do not know whether it is possible to have brains (or thinking machines in general) without having consciousness. In particular, if it were possible to construct an entity that is identical to a human at the atomic level, but that did not experience qualia, we would refer to that construct as a philosophical zombie.
Philosophers have considered essentially three possibilities for the way consciousness fits into the universe. The first possibility is called dualism, in which consciousness and the brain are distinct and separable things. In a dualistic universe, it is possible to have philosophical zombies, because one could in principle create a simulacrum of a human with no consciousness. The second possibility is called physicalism, in which consciousness is an as-of-yet-unknown property of matter. In a physicalist universe, it would be possible to measure consciousness, but it would not be possible to create philosophical zombies, because it is not possible to have brains without consciousness. As an example, consider the electron: one could have asked, long ago, whether it was possible to have an electron that doesn’t have negative charge. I.e., is there any way to take the charge away from the electron, in the same way that one can take the charge away from, say, a jacket, by grounding it? The answer for electrons is basically no: having negative charge and being an electron are actually the same thing. That’s the physicalist perspective: you can’t have the brain without consciousness. In both the physicalist and dualist universes, it would be possible to build a consciousness-measuring box, which could in principle tell you whether AIs are conscious.
There are also reductionist philosophers who think that all of consciousness can be reduced to simpler physical phenomena. This is, in some sense, what’s going on with the integrated information arguments about consciousness, if I understand them correctly (and I might not). To take another electricity and magnetism reference: Electricity and magnetism were originally seen as different forces and appear to be different forces in everyday life, but with relativity one can show that they are in fact the same force, and that it is not possible to have one without the other in a relativistic universe. Maybe there is some way in which brains (or the information processing capacity of brains) and consciousness are similarly two representations of the same underlying phenomenon, and that if we just find the right frame of reference they will look like the same thing. But it’s hard to imagine what this would look like: at least electricity and magnetism were the same kind of thing, i.e., forces associated with particular kinds of matter. Consciousness and brains are different kinds of things. Consciousness and information theory also seem to be different kinds of things, although it’s hard to say, since we don’t know what kind of a thing consciousness is.
Do these distinctions matter? In all three universes, it would be possible to build a consciousness-measuring box. In the reductionist universe, the box would actually measure some other underlying feature of the system that we interpret as consciousness. In a dualist or physicalist universe, it would measure whether consciousness is present. The main reason the distinctions matter is that you are only allowed to have good controls for consciousness in a dualist universe. In other universes (physicalist or reductionist) consciousness is not dissociable from its underlying physical substrates, so it’s hard (impossible?) to build a device that measures consciousness without measuring some other property of the system as well. And even if you are in a dualist universe in which philosophical zombies are possible in principle, they may not be possible in practice.
Perhaps someone with more knowledge of the philosophy of science than I have can help think formally about what can be done, or perhaps it has been explored before. It may be analogous to the problem of measuring charge for the first time: if you invent a device for measuring charge, you can apply it to electrons and protons and such and get some results. The difference between measuring consciousness and measuring charge is that charge is an idea we came up with because we figured out how to measure something, and then had to give it a name. We then gradually figured out what its relationship was to other phenomena we had previously identified, like current and voltage. Consciousness is something that we know (or think) is there, and we’re trying to look for it. It is much harder.
It may alternatively be similar to the problem of building the first microscope. Imagine that one suspects there may be microbes, called bacteria, that cause disease. Until one has a microscope, one has no way of validating the existence of the microbes; but also, it is impossible to validate the microscope without knowing that some diseases are associated with microbes and others are not. But, in the end, the microscope speaks for itself. You need to make a new measurement device and see what happens.
Suppose we had a consciousness measuring box, but suppose we didn’t know exactly what it did. In a panpsychic universe, the box would always register 1, and we would probably assume it was useless and throw it away. Oh well. In a strongly reductionistic universe, it would always register a 0, and we would again assume it is broken and throw it away. However, things may be more interesting: perhaps the box would report 1 when exposed to some but not all humans (and cue dystopian science fiction). Perhaps it would report 1 for all mammals, but for no other branches of the tree of life will register a 1. After long and painstaking investigations, we would be able to rule out several modes of operation of the device, and although we could never specifically connect the outcomes of the measurement device to the hard problem of consciousness, we could study its outputs.
I am of the personal opinion that a full description of consciousness will likely require new physics, simply because my subjective experience of the world cannot be described in terms of forces and atoms. It is difficult to see how any theory grounded in the known laws of physics would answer the question of why Sam has a subjective experience, when a philosophical zombie version of Sam would perform in the same way according to all known measurements. I understand why people are uncomfortable thinking about it, but consciousness is probably one of the greatest unsolved physical mysteries left. It is worth taking it seriously.
 I also think that free will is a real thing that probably necessitates new physics, but that will be the subject of a different post.
 For the same reason, I am highly skeptical of reports that attempt to disprove free will by claiming that brain activity predicting actions precedes conscious decisions, since those arguments presume that humans are reliable reporters on our internal clocks, when other experiments based on saccades demonstrate that we are not in fact reliable reporters on internal timing.
 A fun thought problem you can think about. What would happen if we used a superpowered nanofabrication device to build a copy of you that is identical at the atomic level. Would you experience qualia from both sets of eyes? If not (and the answer is presumably no), what distinguishes the two atomically identical humans, such that they have different consciousnesses? And where does the consciousness come from?
 Although, many physicists seem to prefer that explanation, because it is consistent with the many worlds theory of quantum mechanics, which physicists tend to like, and because they find it more comforting to discard objective reality than to imagine there is something unknown in the universe.
 The generalization of the GHZ question is to ask if there is any way to entangle A, B, and C such that C is a pointer for A, but the reduced density matrix of A and B has entanglement. I think it has been proven that the answer is no. Something about how there are only two classes of tripartite entanglement that cannot be converted into one another by local unitary operations. I also thought about if there’s any way that the W state can be consistent with a pointer. I think the answer is no: suppose A, B, and C are in a W state. Then if |B> = |1> the AC is in |00>, but if |B> = |0> AC is in a bell state. So I think it doesn’t work, B is not a pointer for anything.
The only question I wonder is whether you could somehow put ABC in a W state and have B report on whether there is bipartite entanglement between A and C, and then postulate that A and C can’t be entangled if C is conscious. I suspect this doesn’t work, but at this point I have descended into sophistry. Someone who has done real quantum mechanics more recently than I have would have to think about this.