Philipp Haueis and David Colaço make an excellent case for their main conclusion that metabolic considerations are highly relevant to evaluating and generating models of cognition. However, they have understated the relevance of their analysis for the central philosophical question of the relation between mind and matter. I will describe how metabolic reasoning about energy consumption should be combined with other constraints of speed, space, and history. The result provides an alternative to multiple realizability and medium independence: coherent materialism.
The key question concerns the relation between mental processes and physical processes. The mind–brain identity theory says that all mental processes, such as problem solving, are brain processes, but multiple realizability claims that the mind could be realized by a wide range of physical processes, such as digital computers and space aliens. Medium independence narrows the range by allowing that some physical media may be relevant to restrictions on realization but is vague about what these restrictions might be. Haueis and Colaço astutely notice that metabolism is highly relevant to restricting the range of possible realizations, because mental systems operating in the real world must acquire and use energy economically and efficiently to enable them to operate in their environments.
Metabolism, however, is not the only constraint on cognitive effectiveness in real-world situations. One advantage of neural networks is their massive parallelism, which allows them to perform many computations simultaneously. The brains of humans and other organisms use this parallelism to work rapidly in the world in pursuits where speed matters, in finding food, avoiding predators, and acquiring mates. Speed is a property of physical systems, as we see in the contrast between relatively slow neuronal firing and incredibly fast (trillions of times per second) operation of silicon chips.
The constraint of speed interacts with the metabolic constraint of energy consumption. If greater speed comes at the cost of energy expenditure, then it may not improve an entity’s ability to survive and reproduce in a relatively non-threatening world.
A third restriction on the physical realization of mental processes is space, because thinking must be an appropriate size to operate in their environments. Human brains could not be much bigger because they initially grow in and are birthed from mothers. Realizability and independence are therefore limited by the size of the thinking substrate. The data centers that run current artificial intelligence models are enormous, covering multiple square miles, but they cannot be unlimited because of the availability of land. The constraints of metabolism and space interact because large systems require more energy to enable them to operate.
The fourth constraint on physical realization is history, which operates because both biological evolution and technical design cannot pull new arrangements out of the air but must work with prior physical properties. Biology never managed to produce brains capable of extrasensory perception, and computer design also has to build on top of currently available technologies. Metabolism interacts with history because any new entity, evolved or designed, is limited by the energy capabilities of its predecessors.
These four constraints show the implausibility of multiple realizability and make even medium independence suspect. But we should not swing back to mind–brain identity that assumes zero independence, not because of fanciful thought experiments, but because generative AI has brought computers capable of many intelligent operations, including sophisticated language use, problem solving, learning, and many kinds of reasoning. The rise of powerful AI beginning with ChatGPT in 2022 shows that minds are realizable in special computers as well as brains.
Consideration of these four interacting constraints might be described as scant realizability or minimal independence, but I call the resulting view of mental-physical arrangements coherent materialism (Thagard, Reference Thagard2025, Reference Thagardin press). Here coherence is not just the vague idea of fitting together, but rather the precise idea, computable by neural networks and other algorithms, that coherence is satisfaction of multiple conflicting constraints (Thagard, Reference Thagard2000). According to coherent materialism, only rare physical processes will sufficiently satisfy constraints of energy, time, space, and history in their environments to achieve mental accomplishments such as intelligent problem solving. At the moment, we have only two realizations of such minds that we know about, using biological brains and engineered computer chips using GPU and similar processors. Others may turn up through technological innovation or biological evolution on other planets, but constraints that include metabolism show that we should not expect many ways of physical production of minds.
The new AI systems illustrate the interactions of the four constraints of energy, time, space, and history, sharply in contrast with how humans satisfy these constraints. Compared to humans, AI data centers are massive energy hogs, requiring hundreds of megawatts to operate compared to the 20 watts of human brains. On the other hand, they are capable of enormous speed, because brains are limited by the relative slowness of neural firing and neurotransmitter propagation. AI data centers are becoming ridiculously large, but they can be built that way without having to worry about being born or carried around. AI models are free of the historical constraints of having to evolve slowly through genetic inheritance but are becoming capable of major leaps in part because they can help to design themselves, as happens when programming models made by OpenAI and Anthropic contribute to their next generation.
The new AI and considerations of the metabolism of brains together point the way to a new solution to the mind–body problem, coherent materialism. The energy constraint of metabolism should be viewed as interacting with the additional constraints of time, space, and history to greatly restrict realizability and independence. Hence what Haueis and Colaço call the “metabolic turn” in philosophy is even more consequential than they surmise.
You have
Access
Philipp Haueis and David Colaço make an excellent case for their main conclusion that metabolic considerations are highly relevant to evaluating and generating models of cognition. However, they have understated the relevance of their analysis for the central philosophical question of the relation between mind and matter. I will describe how metabolic reasoning about energy consumption should be combined with other constraints of speed, space, and history. The result provides an alternative to multiple realizability and medium independence: coherent materialism.
The key question concerns the relation between mental processes and physical processes. The mind–brain identity theory says that all mental processes, such as problem solving, are brain processes, but multiple realizability claims that the mind could be realized by a wide range of physical processes, such as digital computers and space aliens. Medium independence narrows the range by allowing that some physical media may be relevant to restrictions on realization but is vague about what these restrictions might be. Haueis and Colaço astutely notice that metabolism is highly relevant to restricting the range of possible realizations, because mental systems operating in the real world must acquire and use energy economically and efficiently to enable them to operate in their environments.
Metabolism, however, is not the only constraint on cognitive effectiveness in real-world situations. One advantage of neural networks is their massive parallelism, which allows them to perform many computations simultaneously. The brains of humans and other organisms use this parallelism to work rapidly in the world in pursuits where speed matters, in finding food, avoiding predators, and acquiring mates. Speed is a property of physical systems, as we see in the contrast between relatively slow neuronal firing and incredibly fast (trillions of times per second) operation of silicon chips.
The constraint of speed interacts with the metabolic constraint of energy consumption. If greater speed comes at the cost of energy expenditure, then it may not improve an entity’s ability to survive and reproduce in a relatively non-threatening world.
A third restriction on the physical realization of mental processes is space, because thinking must be an appropriate size to operate in their environments. Human brains could not be much bigger because they initially grow in and are birthed from mothers. Realizability and independence are therefore limited by the size of the thinking substrate. The data centers that run current artificial intelligence models are enormous, covering multiple square miles, but they cannot be unlimited because of the availability of land. The constraints of metabolism and space interact because large systems require more energy to enable them to operate.
The fourth constraint on physical realization is history, which operates because both biological evolution and technical design cannot pull new arrangements out of the air but must work with prior physical properties. Biology never managed to produce brains capable of extrasensory perception, and computer design also has to build on top of currently available technologies. Metabolism interacts with history because any new entity, evolved or designed, is limited by the energy capabilities of its predecessors.
These four constraints show the implausibility of multiple realizability and make even medium independence suspect. But we should not swing back to mind–brain identity that assumes zero independence, not because of fanciful thought experiments, but because generative AI has brought computers capable of many intelligent operations, including sophisticated language use, problem solving, learning, and many kinds of reasoning. The rise of powerful AI beginning with ChatGPT in 2022 shows that minds are realizable in special computers as well as brains.
Consideration of these four interacting constraints might be described as scant realizability or minimal independence, but I call the resulting view of mental-physical arrangements coherent materialism (Thagard, Reference Thagard2025, Reference Thagardin press). Here coherence is not just the vague idea of fitting together, but rather the precise idea, computable by neural networks and other algorithms, that coherence is satisfaction of multiple conflicting constraints (Thagard, Reference Thagard2000). According to coherent materialism, only rare physical processes will sufficiently satisfy constraints of energy, time, space, and history in their environments to achieve mental accomplishments such as intelligent problem solving. At the moment, we have only two realizations of such minds that we know about, using biological brains and engineered computer chips using GPU and similar processors. Others may turn up through technological innovation or biological evolution on other planets, but constraints that include metabolism show that we should not expect many ways of physical production of minds.
The new AI systems illustrate the interactions of the four constraints of energy, time, space, and history, sharply in contrast with how humans satisfy these constraints. Compared to humans, AI data centers are massive energy hogs, requiring hundreds of megawatts to operate compared to the 20 watts of human brains. On the other hand, they are capable of enormous speed, because brains are limited by the relative slowness of neural firing and neurotransmitter propagation. AI data centers are becoming ridiculously large, but they can be built that way without having to worry about being born or carried around. AI models are free of the historical constraints of having to evolve slowly through genetic inheritance but are becoming capable of major leaps in part because they can help to design themselves, as happens when programming models made by OpenAI and Anthropic contribute to their next generation.
The new AI and considerations of the metabolism of brains together point the way to a new solution to the mind–body problem, coherent materialism. The energy constraint of metabolism should be viewed as interacting with the additional constraints of time, space, and history to greatly restrict realizability and independence. Hence what Haueis and Colaço call the “metabolic turn” in philosophy is even more consequential than they surmise.
Acknowledgements
None.
Financial support
This project has no funding.
Competing interests
The author has no competing interests.