Макс Тегмарк - Life 3.0 - Being Human in the Age of Artificial Intelligence

A famous family of NCC experiments pioneered by physiologist Benjamin Libet has shown that the sort of actions you can perform unconsciously aren’t limited to rapid responses such as blinks and ping-pong smashes, but also include certain decisions that you might attribute to free will—brain measurements can sometimes predict your decision before you become conscious of having made it.15

Theories of Consciousness

We’ve just seen that, although we still don’t understand consciousness, we have amazing amounts of experimental data about various aspects of it. But all this data comes from brains, so how can it teach us anything about consciousness in machines ? This requires a major extrapolation beyond our current experimental domain. In other words, it requires a theory.

Why a Theory?

To appreciate why, let’s compare theories of consciousness with theories of gravity. Scientists started taking Newton’s theory of gravity seriously because they got more out of it than they put into it: simple equations that fit on a napkin could accurately predict the outcome of every gravity experiment ever conducted. They therefore also took seriously its predictions far beyond the domain where it had been tested, and these bold extrapolations turned out to work even for the motions of galaxies in clusters millions of light-years across. However, the predictions were off by a tiny amount for the motion of Mercury around the Sun. Scientists then started taking seriously Einstein’s improved theory of gravity, general relativity, because it was arguably even more elegant and economical, and correctly predicted even what Newton’s theory got wrong. They consequently took seriously also its predictions far beyond the domain where it had been tested, for phenomena as exotic as black holes, gravitational waves in the very fabric of spacetime, and the expansion of our Universe from a hot fiery origin—all of which were subsequently confirmed by experiment.

Analogously, if a mathematical theory of consciousness whose equations fit on a napkin could successfully predict the outcomes of all experiments we perform on brains, then we’d start taking seriously not merely the theory itself, but also its predictions for consciousness beyond brains—for example, in machines.

Consciousness from a Physics Perspective

Although some theories of consciousness date back to antiquity, most modern ones are grounded in neuropsychology and neuroscience, attempting to explain and predict consciousness in terms of neural events occurring in the brain.16 Although these theories have made some successful predictions for neural correlates of consciousness, they neither can nor aspire to make predictions about machine consciousness. To make the leap from brains to machines, we need to generalize from NCCs to PCCs: physical correlates of consciousness, defined as the patterns of moving particles that are conscious. Because if a theory can correctly predict what’s conscious and what’s not by referring only to physical building blocks such as elementary particles and force fields, then it can make predictions not merely for brains, but also for any other arrangements of matter, including future AI systems. So let’s take a physics perspective: What particle arrangements are conscious?

But this really raises another question: How can something as complex as consciousness be made of something as simple as particles? I think it’s because it’s a phenomenon that has properties above and beyond those of its particles. In physics, we call such phenomena “emergent.”17 Let’s understand this by looking at an emergent phenomenon that’s simpler than consciousness: wetness.

A drop of water is wet, but an ice crystal and a cloud of steam aren’t, even though they’re made of identical water molecules. Why? Because the property of wetness depends only on the arrangement of the molecules. It makes absolutely no sense to say that a single water molecule is wet, because the phenomenon of wetness emerges only when there are many molecules, arranged in the pattern we call liquid. So solids, liquids and gases are all emergent phenomena: they’re more than the sum of their parts, because they have properties above and beyond the properties of their particles. They have properties that their particles lack.

Now just like solids, liquids and gases, I think consciousness is an emergent phenomenon, with properties above and beyond those of its particles. For example, entering deep sleep extinguishes consciousness, by merely rearranging the particles. In the same way, my consciousness would disappear if I froze to death, which would rearrange my particles in a more unfortunate way.

When you put lots of particles together to make anything from water to a brain, new phenomena with observable properties emerge. We physicists love studying these emergent properties, which can often be identified by a small set of numbers that you can go out and measure—quantities such as how viscous the substance is, how compressible it is and so on. For example, if a substance is so viscous that it’s rigid, we call it a solid, otherwise we call it a fluid. And if a fluid isn’t compressible, we call it a liquid, otherwise we call it a gas or a plasma, depending on how well it conducts electricity.

Consciousness as Information

So could there be analogous quantities that quantify consciousness? The Italian neuroscientist Giulio Tononi has proposed one such quantity, which he calls the “integrated information,” denoted by the Greek letter Φ ( Phi ), which basically measures how much different parts of a system know about each other (see figure 8.5).

Figure 8.5: Given a physical process that, with the passage of time, transforms the initial state of a system into a new state, its integrated information Φ measures inability to split the process into independent parts. If the future state of each part depends only on its own past, not on what the other part has been doing, then Φ = 0: what we called one system is really two independent systems that don’t communicate with each other at all.

I first met Giulio at a 2014 physics conference in Puerto Rico to which I’d invited him and Christof Koch, and he struck me as the ultimate renaissance man who’d have blended right in with Galileo and Leonardo da Vinci. His quiet demeanor couldn’t hide his incredible knowledge of art, literature and philosophy, and his culinary reputation preceded him: a cosmopolitan TV journalist had recently told me how Giulio had, in just a few minutes, whipped up the most delicious salad he’d tasted in his life. I soon realized that behind his soft-spoken demeanor was a fearless intellect who’d follow the evidence wherever it took him, regardless of the preconceptions and taboos of the establishment. Just as Galileo had pursued his mathematical theory of motion despite establishment pressure not to challenge geocentrism, Giulio had developed the most mathematically precise consciousness theory to date, integrated information theory (IIT).

I’d been arguing for decades that consciousness is the way information feels when being processed in certain complex ways.18 IIT agrees with this and replaces my vague phrase “certain complex ways” by a precise definition: the information processing needs to be integrated, that is, Φ needs to be large. Giulio’s argument for this is as powerful as it is simple: the conscious system needs to be integrated into a unified whole, because if it instead consisted of two independent parts, then they’d feel like two separate conscious entities rather than one. In other words, if a conscious part of a brain or computer can’t communicate with the rest, then the rest can’t be part of its subjective experience.