AI is Not Alive...Yet

“We’re made of star stuff…”

Consider an antelope standing on a small rise in the middle of the savanna. Off to one side, emanating from a patch of tall grass, there is a rustle. Now, this rustle may just be the wind, or it may be the first steps of a predator about to pounce. Millions of years of evolutionary selection have predetermined what will happen next. That antelope’s pupils will dilate. Its pulse will race. Its muscles will tense. Adrenaline floods its veins. It is primed for flight.

If it turns out that the noise was only the wind, after a short refractory period the antelope’s biological response will return to normal. It will go about its business. The fact that the antelope was ready to flee at even the possibility of a predator ensures its survival in the cutthroat environment of the savanna. Nothing more.

Now consider, instead, a primitive human in the same situation. This ancient ancestor of ours will have the same inherited response. The only difference is that when it turns out to have been only the wind, they will not simply continue as before. Instead, they will ask what caused this dramatic response, and they will tell a story about an evil spirit haunting the patch of grass. Without the knowledge or experience to understand evolution, our ancestors filled in the gaps with a narrative (some might even call this a hallucination, in the large-language model sense of that term).

Many anthropologists think this is where religious belief began.

For as long as I can remember, I was the sort of kid who would ask “why” in a loop until my parents, exasperatedly, responded with “just because!” To which I would, predictably, follow up with: “why because?” Despite being raised in a rather religious family, I drifted away from religion as I got older, not for the usual reasons, but because I found it increasingly unsatisfying. At some point I realized that religion was nothing more than the ultimate exasperated parent. When I persisted with asking “why”, religion would answer with recitations of 2,000-year-old traditions as interpreted by intermediaries more interested in preserving their own power or keeping the masses placated.

This is not an essay about religion. Or, possibly, it very much is an essay about religion. Either way, my point in bringing up religion is to set the frame for how I approach science. I do not see it as standing in place of, or in opposition to, religion. I consider science a natural progression. The deepest roots of religion grow from humans’ desire to understand the world around them and their place in it. Science merely continues that quest but grounded in observable fact, rather than story and myth.

It is not surprising, given this context, that religions are often concerned with the search for profound truths. The problem is that their “truths” often involve unknowable outside forces, and they tend to set humans apart from the rest of the universe. On the other hand, science proposes that, in the search for the how and the why of the universe, we are ourselves a product of the how and the why that we seek to understand. The great Carl Sagan expressed this view rather succinctly: “We are a way for the cosmos to know itself.”

The implications of this simple statement are no less profound than if you woke up one morning to discover that your big toe had grown a face and was now staring at you asking, “Why are we here? Oh, and how come I sometimes bump into hard things?” We humans, as intelligent, sentient beings, governed by the laws of physics, struggle to investigate and understand those same laws, gather what we’ve learned, and pass along the information to future generations.

What makes this all the more remarkable and profound is the fact that, as we now understand, the laws of thermodynamics dictate that the universe, on the whole, tends toward more disorder and less structure. That we can gather, organize, and pass along what we learn about the universe, when the universe itself is seemingly working against us, is as close to a miracle as anything religion provides.

Rush Hour at Grand Central

As an undergraduate chemistry major, thermodynamics was one of my favorite subjects. Then, in graduate school, I transitioned to biochemistry and, eventually, bioinformatics. As I dove deep into the subject of genes and genetic information, the ways in which information is passed along generation to generation, I began to see parallels with thermodynamics. One day I told my first Ph.D. advisor about this line of thinking. He responded by saying, “Josh, that’s a very interesting thought. It is also a career killing thought, so I don’t want you to talk about it again.”

He was half joking as he said this, but only half. As fate would have it, I left that university and transferred back to the institution I attended as an undergraduate. When I told my second Ph.D. advisor about how I was thinking about genetics and evolution as a thermodynamic process, he handed me a book by Ilya Prigogine.

Now, Prigogine is not the most famous Nobel Laureate, to be sure, but reading about his work on dissipative structures had a profound impact on how I viewed thermodynamics, evolution, science, and the universe. Understanding his work is also key to understanding why AI is not alive, but we’ll come to that. First, we must understand a bit about equilibrium, entropy, and how they relate to dissipative structures.

We will start with Grand Central Station in New York City. If you’ve never been, you’ve likely seen it in movies or pictures. If you haven’t, picture the busiest train station you know with many entrances and exits, people coming and going and generally milling about. To understand the concept of equilibrium and the role of disorder, imagine we were to invite a thousand people into Grand Central and then locked all the doors.

At first, our group of stand-ins for thermodynamic particles might all be gathered near each other, but as they begin wandering about, eventually we will come to a situation where the average density of the people is uniform throughout. There may be small clumps of people standing closer together or further apart, some may be walking faster and some slower, but we can take the average over the entire station and use that to describe the station as a whole. The important feature of equilibrium is that, if we were to erect a barrier across the middle of the station, we would expect the average density and speed of people walking about on either side of the barrier to be the same. If we divided each half again, the same should remain true. No mater how many times, or what arrangement of divisions we create, so long as these divisions are significantly larger than the size of a single person, all our divisions should look the same. Disorder reigns. Order can not be restored without putting an effort, say in the form of someone shouting for everyone to gather to one side, into the system.

Now imagine we unlock the doors and it is the middle of rush hour. As one train full of passengers arrives, another will depart. As some people enter from 42nd Street, others exit from 45th. While there is a great commotion of people through the station, the average speed they walk is likely pretty consistent. Similarly, while some parts of the station will be slightly more crowded and others slightly less, the density of bodies in the building is likely fairly constant. We can no longer say that Grand Central is at equilibrium, because now if we erect a barrier we might expect everyone on one side of the barrier to exit the building. If we, instead, took a photo of Grand Central at rush hour, and compared it to a snapshot of our Grand Central shut-in, they’d probably look nearly identical. For this reason, we can say that Grand Central at rush hour is “near” equilibrium (and, therefore, as a neat trick chemists and physicists use, we can write nearly identical equations to describe a system near equilibrium as we do for one at equilibrium).

Dissipative structures, as described by Prigogine, arise from systems that are far from equilibrium. So, to understand them we will need to make a small modification to Grand Central Station. We will need to take that famous, picturesque, main hall of Grand Central and stretch it like taffy, until the two ends are far apart joined only by a thin corridor just wide enough for two people walking in opposite directions to pass each other.

Now, as a rush hour train approaches the station and the passengers all exit the station at one end of this corridor, many of them will need to exit from the other. No longer is the density of people uniform throughout the station. Our system is now “far” from equilibrium. As a crush of people dissipates from where the train stopped to the other end of the station, something miraculous will occur. The taffy corridor, rather than containing people walking past each other in all directions, will arrange itself into two lines of people, one moving in each direction. Furthermore, if you watch where the crowded end of the station narrows toward the corridor, you’ll see more lines of people arrange themselves, and possibly even zipper merge together.

What Prigogine was able to show with his work was that, if you take a system that is very far from equilibrium, but one in which energy is allowed to dissipate in the direction of equilibrium, it is possible for entropy (the thermodynamic measure of disorder) to decrease, locally, for as long as the energy continues to dissipate. The only difference between Prigogine’s dissipative structures and our Grand Central Taffy Station is that the narrow corridors in Prigogine’s structures are available states of a system, rather than physical walls, but the effect is the same.

Humans and Hurricanes

What does any of this have to do with AI? Nothing…yet. The question you should be asking is: “what does any of this have to do with being alive?”

One of my favorite riddles is: “What do humans and hurricanes have in common?” The answer: both are dissipative structures.

I’ve always found hurricanes to be alluringly beautiful (even despite having had to ride out eight separate major hurricanes so far), but it wasn’t until I encountered Prigogine that I began to understand why. Hurricanes, especially the powerful ones, possess exquisite order and symmetry. The source of this order is the fact that the summer atmosphere is a system far from equilibrium. By dissipating heat from the warm sea to the cold atmosphere above, and from the equator to the poles, hurricanes are able, so long as they do, to create order out of the chaotic atmosphere.

In much the same way, we humans possess quite a bit of order, what with our various organs and vessels keeping our blood flowing in the right direction. The source of our order is food. By taking the energy locked up in food molecules and dissipating it to the environment, we are able, so long as we do, to maintain that order. Take a hurricane away from the water that feeds it heat, or take a human away from the food that they eat, and neither will maintain order for very long.

In other words, at least one criteria for being “alive” is that we dissipate energy. To be sure, it is not the only criteria (you’d be hard pressed to find someone to argue that hurricanes are alive), but it is a necessary one. Ceasing to dissipate energy is, quite literally, to die. So what about AI?

The challenge with the question of whether AI is “alive” or not is that science hasn’t really satisfactorily arrived at a definition of what it even means to be alive. As with so many things in biology, the devil is in the details. Sure, humans and dogs and cats and ants and plants are alive. Once we get down to the level of bacteria and viruses, though, the lines begin to blur for most attempts at a formal definition of “life”. Does life replicate? On its own? If so then viruses don’t count, but neither do some bacteria that can only replicate inside of host organisms.

The problem with most attempts at biological definitions of life is that they tend to focus on mechanisms like cellular reproduction, copying of genetic material, and processing energy. The issue is that these definitions are borderline tautologies, as the mechanisms used to define life are those we observe in things that we already agree are alive. If we want to come up with a definition for life that is more generally applicable, we need to search for a deeper truth.

I would propose that “life” could be more generally defined as any dissipative structure that functions to organize and pass along information, in opposition to the universe’s push toward disorder, and in doing so participates in the larger project of the cosmos coming “to know itself”.

This is where you are expecting that I would say that, “of course, AI is not a dissipative structure,” but unfortunately the reality is a bit more complicated. When AI is running, there is certainly a far-from-equilibrium system in the electrical lines that carry power to a server. As AI is doing inference, or whatever other calculations it might be doing, it is dissipating this energy to the environment in the form of heat, and in doing so it does cause the bits in its volatile memory to hold ones and zeros in a specific order. When the power is cut off, and the energy stops dissipating, those ones and zeros become scrambled and chaos returns.

So, is AI alive?

No…not yet.

To understand why, we have to return to our Grand Central Station metaphor, as well as Carl Sagan’s wise observation. Where the Grand Central Station metaphor comes up short in describing a dissipative structure is where we had to bring the walls in close to enforce the order that arose from dissipating the non-equilibrium station arrangement. In much the same way, the ones and zeros of an AI running inference are determined, not spontaneously, but by the weights pre-determined during the AI’s training and stored, statically, somewhere that does not require constant dissipation of energy to maintain order. The reason this is important is because the truly spontaneous generation of order that Prigogine’s structures exhibit is linked, very directly, to Sagan’s cosmos getting to “know itself”.

The laws of physics that limit the states that a system far from equilibrium can explore as it dissipates energy, and that dictate the order that such a structure can exhibit, are the same laws that such a structure, like a living human asking deep questions about the universe, can learn about and understand. When the structure that AI exhibits is dictated by physics, itself, rather than instructions encoded by math…then, maybe, AI will be alive.