What is the relation between the mind (thoughts, perceptions, sensations, emotions) and the physical body? The traditional answer, assumed by all religions that believe in an afterlife, is dualism: Mind and body are two distinctly different things. Dualism’s main flaws are the lack of evidence for the existence of nonphysical souls and the problem of understanding how souls could interact with bodies.
Alternative to Dualism
The main alternative to dualism is materialism, the scientific view that everything that happens in the universe is the result of changes in matter and energy. Two versions of materialism are predominant. The mind-brain identity theory says that all mental processes are brain processes. An alternative view, inspired by the development of computers, says that mental processes should be viewed more generally as functional relations between inputs and outputs, which can be implemented in many different kinds of hardware besides brains. This view goes by various names, including functionalism, Turing-machine functionalism, computational functionalism, and substrate independence.
The main sticking point for materialism is the difficulty of accounting for conscious experience, which at first glance seems totally different from anything physical because it is qualitative, subjective, and private. I have attempted to fill this explanatory gap with the NBC theory, which proposes that conscious experiences result from four brain mechanisms: neural representation, binding, coherence, and competition. If this theory is true, then so is materialism, but what version?
Coherent Materialism
NBC might be taken to support mind-brain identity, but the rise of generative artificial intelligence (AI) since 2022 seems to provide support for functionalism, since models like ChatGPT, with their impressive capacity for language and inference, run on computers rather than brains. Considering both NBC and generative AI shows that we need a new version of materialism, which I call coherent materialism, or cohmaterialism for short. It provides a new solution to the mind-body problem that is less restrictive than the mind-brain identity theory, which says that only brains can think, but much less permissive than the functionalist claim that any powerful computer can think.
The advent of generative AI might mistakenly be taken as support for the functionalist view that mental capabilities are independent of their underlying physical substrates, resulting from software rather than hardware. On the contrary, generative AI provides excellent illustrations of how mental operations are subject to hardware constraints of time, space, and energy. The most successful AI models that mimic many kinds of human thinking do not run on just any kind of hardware but depend heavily on special chips made by Nvidia that are designed for fast and highly parallel operation. Without such hardware that contributes to the training and application of the new models, the computers would be too slow to seem intelligent.
These computers are also subject to space constraints, requiring huge buildings for data centers with thousands of computers, and to energy constraints, with massive amounts of energy needed to run them, along with large quantities of water for cooling. The models are also constrained by their history, such as the groups of people who work together to build them.
Moreover, despite their impressive abilities to solve problems and pass tests, AI models show no apparent capacity for conscious experiences such as felt perceptions, sensations, and emotions. Human experiences are heavily dependent on the kinds of bodies we have—for example, when happy emotions result from combinations of internal bodily changes such as heart rate, neurotransmitters such as dopamine, and interacting brain areas such as the amygdala and the ventromedial prefrontal cortex.
I use the term “coherent” to mean something much more precise than just “making sense” or “hanging together.” My theory of coherence analyzes it as consisting of the satisfaction of multiple constraints—for example, when a coherent decision does a better job of accomplishing relevant goals than its alternatives. An intelligent system consisting of a coupling of hardware and software must be coherent in that it accomplishes tasks such as perception, inference, and language while satisfying these constraints:
- Time: The coupled system must be fast enough to survive and accomplish its goals in its environment.
- Space: The system must be small enough to reproduce and operate in its environment, but big enough to have causal effects required for survival.
- Energy: The system must be efficient in acquiring and expending energy so that it can operate and survive in its environment.
- History: The development of the system must be available given its prior structures and processes.
Brains of humans and other animals do a marvelous job of satisfying these constraints, and generative AI is also succeeding to a large extent, without yet approaching consciousness.
Because satisfying these constraints is crucial for intelligent minds and because coherence is constraint satisfaction, this version of materialism can be called coherent materialism, or cohmaterialism for short. Its central claims are:
- Mental processes, like everything in the universe, result from the interactions of matter and energy, which are interconvertible in accord with E=mc2.
- The production of mental processes by physical processes depends on the coherent satisfaction of constraints that include time, space, energy, and history.
- Because such coherent satisfaction is hard to produce, mind is rare in the universe. Panpsychism is false.
- Different ways of accomplishing coherent satisfaction may result in different kinds of mental processes (e.g., conscious brains vs. current generative AI).
The NBC theory of consciousness does not depend on cohmaterialism but supports it as the most plausible general philosophy of mind consistent with scientific knowledge about thought and the universe.
