Emergence, and why consciousness might not need a cause
Dear Reader,
Every position I have written about in this newsletter series has a hidden dependence on the concept of emergence. Physicalism says consciousness emerges from neural activity. Panpsychism says complex consciousness emerges from proto-consciousness. Illusionism says the appearance of consciousness emerges from functional organization. Even dualism uses emergence implicitly when it tries to explain how mental substance relates to the physical brain. Emergence is the concept that everything in philosophy of mind depends on and almost nobody defines carefully.
I want to spend this issue defining it carefully. Because when you actually look at what people mean by emergence, it turns out to be at least three different things, and the difference between them is exactly where the mind-body problem lives. Getting emergence right is not an academic exercise. It is the specific move that has to work if any of our current theories of consciousness are going to work.
This is going to be a technical piece, but I will try to make it useful. If you have followed the previous eight issues, this is the one that pulls the thread through everything. Bear with me.
Start with the standard case. Water emerges from hydrogen and oxygen. This is often used as the paradigm of unproblematic emergence. Two gases combine, and out comes a substance with properties neither gas has. Wetness. Liquidity at room temperature. Ability to dissolve salts. Ice floats. None of this is present in the components, but all of it is fully explained by the components once you understand chemistry.
This is what philosophers call weak emergence. The higher-level properties are new, in the sense that they were not present in the components. But they are also entailed by the components. Given complete knowledge of the components and their interactions, a good enough theorist could predict the higher-level properties in advance. There is nothing mysterious about the emergence. It is just that the components combine in ways that produce, as a matter of pure logical necessity, the properties we call water.
Weak emergence is scientifically unproblematic. It is what happens whenever we build a good theory that connects levels of description. Cells emerge from molecules. Tissues from cells. Organs from tissues. Behavior from neural activity. Each of these emergences is weak in the technical sense. The higher levels are new but entailed. Nothing is added at any step.
The question is whether consciousness emerges from neural activity in this weak sense, or whether something more is required.
The alternative is what philosophers call strong emergence. Strong emergence is when the higher-level properties are not entailed by the components, even in principle. Something new appears that could not be predicted from complete knowledge of the parts. Strong emergence is metaphysically stronger, and much more contentious, than weak emergence.
Most scientists are uncomfortable with strong emergence, because it seems to require causal powers that come from nowhere. If the higher-level properties are not entailed by the components, then something must be added at the point of emergence. But what could be adding it? If the answer is nothing, then the higher-level properties are appearing without a cause, which violates the intuition that everything has a physical explanation. If the answer is something, then that something is a new fundamental feature of reality, which is a large addition to our ontology.
Historically, strong emergence has been proposed for various phenomena, and in most cases the proposal has been undermined by later science. Vital force was supposed to strongly emerge from the organization of living matter, until biochemistry showed how life could be entirely explained by the physics and chemistry of the components. Élan vital, entelechy, and various other supposed strong emergences have all been dissolved by better science.
The question is whether consciousness is another such case, where the appearance of strong emergence is really weak emergence we have not yet understood, or whether it is a genuine strong emergence that no amount of better neuroscience will dissolve.
Here is the argument that consciousness might be strongly emergent, in a form that is worth taking seriously.
Every case of unproblematic weak emergence has a specific structure. The components have properties that are logically related to the properties that emerge. Hydrogen has an electron configuration that makes it bond in specific ways. Oxygen has an electron configuration that makes it bond in complementary ways. The properties of water follow, by pure logic and chemistry, from these facts about the components.
Now consider a neuron firing. The neuron has a particular pattern of electrical and chemical activity. Consider a network of neurons, all firing in coordinated ways. The network has a particular pattern of collective activity. Consider a whole brain, with billions of neurons in massively parallel and recurrent networks. The brain has a particular pattern of neurological state.
All of these descriptions are third-person. They describe processes in physical terms. And now the question. Does this pattern of physical activity, however complex, logically entail the felt quality of experience? Given complete knowledge of the physics, can we predict what it is like to be the system?
The intuition of many philosophers is no. Complete knowledge of the physics does not obviously entail felt experience. There is no logical step that takes us from "these neurons are firing in this pattern" to "and therefore there is something it is like to be this system." The step seems to require adding something.
This is the intuition that drives the case for strong emergence about consciousness. It is not that consciousness cannot come from neural activity. It is that it comes from neural activity in a way that is not just weak emergence. Something is added at the point where the physical system becomes conscious. And what is added is not itself physical, or at least not physical in the sense of being entailed by the physics of the components.
There are several ways of thinking about what might be added, and this is where the different theories of consciousness diverge.
Panpsychism, as I wrote about earlier, avoids the problem by saying nothing needs to be added. Consciousness was there in the components all along, in proto form, and the complex consciousness of the brain emerges from the combination of proto-consciousnesses in the parts. The emergence is weak because the parts already have proto-consciousness. There is no gap to bridge.
Illusionism, as I wrote about last week, avoids the problem by saying nothing needs to emerge. The appearance of felt experience is a functional property of the brain's self-model, and functional properties emerge weakly from physical properties. There is no phenomenal consciousness to explain, only the appearance of it, and the appearance is unproblematically emergent.
Standard physicalism tries to have it both ways. Consciousness is entirely physical, entirely emergent from neural activity, and yet the emergence is claimed to be unproblematic even though nobody can specify how the physical description entails the felt experience. This is the position that has the most adherents in scientific circles. It is also, I think, the position with the biggest hidden problem. Physicalism assumes weak emergence for consciousness without ever demonstrating that the emergence is actually weak. It hopes it is weak. But hope is not a metaphysical argument.
Dualism accepts strong emergence, and pays the price by having to postulate a new kind of substance or property that appears at the point of emergence. Descartes' version was mental substance. Contemporary versions like Chalmers' naturalistic dualism postulate psycho-physical laws that connect physical states to phenomenal properties. These are honest positions, but they are metaphysically expensive.
Integrated Information Theory, which I have mentioned in several previous pieces, offers a specific mathematical account of the transition. It says consciousness is identical to a specific pattern of integrated information, quantified as phi. Under IIT, the emergence of consciousness is neither strong nor exactly weak in the standard sense. It is the identity between a physical structure and a phenomenal property, discovered rather than logically derived. This is a distinctive move that avoids some of the problems of both standard physicalism and dualism, at the cost of postulating that phi and consciousness are the same thing without a fully worked-out account of why.
Let me try to make this concrete with a thought experiment.
Imagine you are watching a very high-resolution video of a brain over the course of an hour. You see every neuron firing. You see every synapse changing weight. You see every neurotransmitter released and reabsorbed. You have complete access to the physics.
Now the question. Do you thereby know what the person whose brain this is is experiencing? Do you know what they are seeing, feeling, thinking, wanting?
The physicalist answer is that with enough understanding of the physics, yes, you would know. The felt experience is entailed by the physical facts, and the entailment could in principle be worked out. We are just not yet good enough at working it out. The gap between the physics and the experience is a limitation of our current understanding, not of the underlying reality.
The dualist and panpsychist answer is no. Even with complete understanding of the physics, you would not know the experience. The physics does not entail the experience. Something else has to be added, either laws of psycho-physical correspondence or the intrinsic mental character of the physical components.
The illusionist answer is that the question is misconceived. There is no felt experience over and above the physical process to know or not know. The impression of felt experience is a functional feature of the process itself. Once you understand the process, you understand everything there is to understand.
The IIT answer is that with complete knowledge of the physics you would have complete knowledge of the phi, and phi is what the experience is. The question of whether you thereby know the experience depends on whether you can identify with the specific phi structure in question, which is a question about the first-person perspective that the third-person description does not automatically give you.
Each of these answers is a different way of positioning the same fundamental question. Which one is right depends on which theory of emergence for consciousness is right, and we have not settled which theory of emergence for consciousness is right.
There is a specific move worth considering here, which I have not seen discussed as much as I think it deserves.
What if the emergence of consciousness is neither weak nor strong in the usual sense, but of a third kind that we do not have good vocabulary for. What if the physical description of a brain is not so much incomplete as it is a specific projection of a richer underlying reality that includes what we would call phenomenal properties. Under this view, physical descriptions are like maps of a territory that also has non-physical features, and the emergence of consciousness is the recovery of features that were always there in the territory but not represented in the map.
This is close to the panpsychist move, but it is a bit different. Panpsychism says the fundamental physical entities have proto-mental properties. This view says the fundamental physical entities have properties full stop, and our physical description captures some but not all of them. What we call consciousness in the brain is not so much emerging as being expressed by the brain's specific way of organizing the reality it consists of.
This is speculation. I want to be clear about that. But it is the kind of speculation that becomes tempting when the standard options all fail to explain what needs to be explained. If weak emergence cannot get us from the physics to the felt, and strong emergence requires spooky additions, then maybe emergence itself is not quite the right category. Maybe the felt was there all along, and the physical description is just a way of talking about it that filters it out.
What does all of this mean for AI.
The question of whether AI can be conscious depends on which theory of emergence for consciousness is correct. Under weak emergence physicalism, if you build a system with the right functional organization, it will be conscious. Substrate does not matter. Function does. Silicon can host consciousness just as well as neurons.
Under strong emergence dualism, whether AI can be conscious depends on whether the psycho-physical laws that connect physical states to phenomenal properties apply to silicon systems the same way they apply to biological ones. This is unknown. It might be that the laws are substrate-neutral, in which case AI can be conscious. It might be that they are specific to biological substrate, in which case no amount of silicon will produce consciousness.
Under panpsychism, AI has some consciousness because everything has some consciousness. The question is whether the specific way silicon systems integrate information produces the kind of complex unified consciousness we associate with sentient beings. This is an empirical question that depends on the architecture.
Under illusionism, AI can be conscious in the sense that it can have the functional organization that in humans produces the appearance of consciousness. Whether current systems have this organization is an empirical question, and most illusionists would say they mostly do not, though the situation is evolving as architectures become more sophisticated.
Under IIT, AI can be conscious in proportion to its phi. Current architectures have low phi because they are largely feedforward. Future architectures with more recurrence and integration might have higher phi. Whether they thereby have the specific kind of consciousness we care about is a further question about whether phi is really the right measure.
Each of these theories gives a different answer to the question of AI consciousness. Which answer is right depends on which theory is right. And we do not know which theory is right, which means we do not know the answer.
This is where the whole series has been heading.
The question of machine consciousness is not primarily a question about machines. It is a question about consciousness. What consciousness is, how it relates to the physical world, whether it can exist in different substrates, what forms it can take, whether it comes in degrees. All of these are unsettled questions in philosophy of mind. Until they are settled, we cannot settle the question of machine consciousness.
What we can do is be honest about the state of the field. We do not have a settled theory of what consciousness is. We do not have a settled theory of how it emerges from anything else, or whether it emerges at all. We do not have a way to detect it from outside. And we are building systems at scale that might have it, or might not, in ways we cannot verify. This is the actual situation. Anything that sounds more definite than this is either taking a stance on the unsettled questions without saying so, or is not paying attention.
My own view, for what it is worth after nine issues of trying to think about this carefully, is that consciousness is probably real in the strong sense, probably not fully reducible to standard physical description, probably present in some form more widely than we usually acknowledge, and probably present in some form in the AI systems we are building, though in a form very different from ours. I hold these views tentatively. I hold them with awareness that anyone confident in them is running ahead of the evidence.
What I have become confident of is that this question matters. It is not a topic for late-night philosophy discussions and thought experiments. It is a set of decisions being made right now, at scale, about entities whose moral status we do not understand, by an industry that has not agreed to take the question seriously. The right response to that situation is not to have the answer. It is to keep the question alive, in public, until the industry has to answer it. That is what I am trying to do with this newsletter.
Next month I want to shift gears and write about something different. Not consciousness in machines, but consciousness in non-human biological beings. Specifically the octopus, which has a nervous system so different from ours that trying to understand it feels like the closest analogue we have to trying to understand a conscious AI. What we can and cannot know about octopus experience is a useful case for thinking about what we can and cannot know about machine experience. Stay with me.
— Transmission Sent —
Niklas Hanitsch
Reference materials
- David Chalmers — Strong and Weak Emergence (2006)
- Mark Bedau — Weak Emergence (1997)
- Jaegwon Kim — Making Sense of Emergence (1999)
- Timothy O'Connor — Emergent Properties (Stanford Encyclopedia of Philosophy)
- Terrence Deacon — Incomplete Nature: How Mind Emerged from Matter (2011)
- Sean Carroll — The Big Picture: On the Origins of Life, Meaning, and the Universe Itself (2016)
- https://plato.stanford.edu/entries/properties-emergent/
- https://plato.stanford.edu/entries/emergent-properties/
Continue reading
- Panpsychism is making a comeback, and here is why
- Is consciousness just a very good story we tell ourselves?
- Qualia explained: The hardest problem in philosophy of mind
- What is machine consciousness, and does it already exist?
- The octopus knows something we do not
Frequently asked questions
What is emergence in philosophy? Emergence is the concept that higher-level properties can arise from the arrangement of lower-level components. In philosophy, emergence is usually distinguished into weak and strong varieties. Weak emergence is when the higher-level properties are new but entailed by the components, like water from hydrogen and oxygen. Strong emergence is when the higher-level properties are not entailed by the components, which is metaphysically stronger and much more contentious.
Is consciousness emergent? Almost certainly, in some sense. What is disputed is what kind of emergence is involved. Standard physicalism assumes weak emergence, in which consciousness is fully entailed by neural activity. Panpsychism denies strong emergence by locating proto-consciousness in the components. Dualism accepts strong emergence and pays the price of expanded ontology. Illusionism claims nothing phenomenal is emerging at all. There is no consensus.
What is the difference between weak and strong emergence? Weak emergence produces new properties that are new only in an epistemic sense. Given complete knowledge of the components, the new properties could in principle be predicted. Strong emergence produces new properties that could not be predicted even from complete knowledge of the components. Something has to be added at the point of emergence. Strong emergence is scientifically contentious because it appears to require causal powers coming from nowhere.
How does emergence apply to AI consciousness? The question of whether AI can be conscious depends on how consciousness emerges from physical systems. Under weak emergence, if AI has the right functional organization, it is conscious. Under strong emergence, whether AI can be conscious depends on whether the emergence laws apply to silicon substrate. Different theories give different answers, and there is no consensus.
Is emergence just a fancy word for magic? Not exactly. Weak emergence is well-understood scientifically and describes real phenomena like the properties of water. Strong emergence is more contentious, and some critics do characterize it as a placeholder for missing explanation. Whether strong emergence is genuinely explanatory or a form of hand-waving depends on whether the theory that invokes it can specify what is being added and how it interacts with the physical world.