A rigorous guide to separating brain, mind, intelligence, computation, and consciousness – and to asking what AI performance actually proves.
A system receives a question about heat transfer and produces a correct explanation. It distinguishes temperature from heat flow, notes the role of thermal conductivity, and predicts why metal can feel colder than wood even when both are at the same room temperature.
What has the result established?
At minimum, it has established an observable event: under those conditions, the system generated an answer that met a task criterion. If the performance is repeated across carefully varied cases, we may have evidence for a capability. But the same output does not, by itself, tell us that the system has general intelligence, implements the same mechanisms as a biological brain, possesses a mind, or has conscious experience.
Week 11 argued that scientific creativity becomes testable when constraints, assumptions, evidence, and failure criteria are made explicit. Week 12 applies the same discipline to our vocabulary. The central rule is simple: brain, mind, intelligence, and computation are related concepts, but they are not interchangeable measurements.
Four words, four explanatory targets
A brain is a biological system. Neuroscience studies cells, circuits, signals, connectivity, chemistry, anatomy, and their causal relationships to behavior and experience. Computational neuroscience can model these processes, but the usefulness of a computational model does not erase the distinction between a model and the biological mechanism it is intended to explain. Kriegeskorte and Douglas argue that progress requires computational models to be tested against both brain and behavioral data.[5]
Intelligence is different. It is not a tissue that can be directly pointed to. In human research, intelligence is a construct inferred from structured patterns of performance, including reasoning, problem solving, learning, and related abilities.[3][4] Different theories partition those abilities differently. A score can therefore be informative only relative to the tasks, population, measurement model, and interpretation that justify it.
Computation is different again. In mathematics and computer science, computation can be characterized abstractly; in physical science, we also ask how a concrete system implements a computation. Contemporary philosophy of computation emphasizes that computational modeling of a physical system is not automatically the same claim as saying that the physical system itself is a computing device.[8] A storm can be simulated by a computer without the storm thereby executing the simulation.
Finally, mind is the most theory-laden term in this set. Here I use it operationally to refer to mental phenomena and capacities such as perception, memory, thought, affect, intentionality, agency, and conscious experience. That definition deliberately avoids deciding whether minds are identical to brain states, functional organizations, computational processes, or something else. The computational theory of mind is a serious family of positions, but it remains a theory that must explain what “computation” means and how computational description relates to neurophysiological and intentional description.[7]
Consciousness should be separated once more. It concerns subjective experience: whether there is something it is like for a system to be in a state. Intelligence and consciousness may interact, but neither term is a synonym for the other.
One successful output can support several different hypotheses
David Marr’s classic framework helps expose the category problem. He distinguished the computational level – what problem is being solved and why – from the representational and algorithmic level – how the transformation is carried out – and from the implementation level – how that process is physically realized.[2]
Marr’s framework does not solve the mind-body problem. Its value here is methodological: a result at one level does not automatically identify what is true at another.
Suppose an AI system correctly classifies an image. We can ask at least four questions:
- Performance: Did it classify this and related images correctly under specified conditions?
- Capability: Does the performance generalize robustly enough to justify a defined competence claim?
- Mechanism: What representations, algorithms, dynamics, and physical implementation produced the result, and are they meaningfully similar to biological mechanisms?
- Mental status: Does the system instantiate states that a defensible theory would count as understanding, intentionality, or conscious experience?
The questions can inform one another, but answering the first does not settle the fourth.
Construct validity is the missing bridge
The practical danger appears when an operational measure is treated as the construct itself. A benchmark may contain problems labeled “reasoning.” A model obtains a high score. It is tempting to say, “The model has reasoning.”
But the scientifically stronger question is: what evidence makes this benchmark a valid measure of the construct we call reasoning?
A 2025 systematic review of 445 large-language-model benchmarks found recurring patterns in task design and scoring that can undermine the construct validity of resulting claims.[11] This does not mean benchmarks are useless. It means benchmark evidence should be named at the level it directly supports, then extended only through an explicit validation argument.
The same caution applies to the word intelligence. The OECD’s current definition of an AI system is operational: machine-based systems infer from inputs how to generate outputs such as predictions, content, recommendations, or decisions.[6] The definition does not require consciousness, humanlike cognition, or a brain.
Why language becomes especially dangerous around AI
Human beings readily use social language for systems that produce fluent responses. That tendency can be useful for interaction, but it can also blur scientific categories.
Ibrahim and Cheng’s 2026 analysis of more than 200,000 research abstracts reports growth in anthropomorphic terminology in computer science and LLM research, and argues that anthropomorphic assumptions can shape training, evaluation, interpretation, and human interaction.[12] Separately, behavioral research shows that people can distinguish intelligence-like attributes from experience-like attributes when judging LLMs. In a preregistered study of 410 participants, intelligence attributions and experience/consciousness attributions related differently to willingness to accept AI advice.[14]
That matters for scientific communication: “intelligent,” “wants,” “believes,” and “feels” introduce different claims and can shape trust.
The responsible response is to mark when metaphor stops being shorthand and starts carrying an evidential burden.
Consciousness is the clearest case for restraint
In humans, conscious experience depends on brain activity, yet explaining how neural activity relates to subjective experience remains an open scientific and philosophical problem.[18] The field contains competing theories, and recent high-profile evidence has not produced a single settled winner.
In 2025, a large adversarial collaboration tested preregistered predictions from integrated information theory and global neuronal workspace theory using multiple neural measurement modalities. The results included both support and important challenges for each theory, and the authors explicitly resisted a simple final verdict.[15] That is a strong reminder that even for human consciousness – where we have brains, reports, behavior, and physiology – inference remains difficult.
For AI, the uncertainty is greater. Butlin and colleagues therefore propose theory-derived indicators rather than a simple behavioral test, explicitly noting risks of both over-attributing and under-attributing consciousness.[16] A 2026 analysis of biological naturalism similarly argues that biology should inform the problem but cannot currently rule out artificial consciousness by itself.[17]
So the defensible statement is neither “today’s AI is obviously conscious” nor “silicon can never be conscious.” It is narrower: fluent behavior alone is not a validated scientific test of subjective experience, and current assessment remains theory-dependent and uncertain.
Competing philosophies should remain visible
This caution does not imply that computation and mentality must be fundamentally separate. Computational functionalists argue that the right functional or computational organization may realize mental states across different physical substrates.[7] Searle’s famous Chinese Room argument pushes in the opposite direction, arguing that executing a program is not, by itself, sufficient for intentionality.[9]
Those are serious competing positions, not empirical verdicts.
Turing’s own 1950 paper is instructive. Rather than assuming that the vague question “Can machines think?” could be settled by definition, he replaced it with a more operational imitation-game question.[1] Whatever one concludes about the Turing Test, the methodological move remains valuable: when a word carries too much philosophical weight, replace it with a testable question whenever possible.
A claim discipline for intelligent systems
Before writing “the system thinks,” “the model understands,” or “the robot is intelligent,” I recommend four checks:
Observation. What exactly happened, under what conditions?
Construct. Which defined capability or mental concept are we trying to infer?
Evidence bridge. What validation, transfer test, causal evidence, or theory connects the observation to that construct?
Boundary. What alternative explanation or unresolved uncertainty prevents a stronger claim?
This is not linguistic pedantry. It changes experimental design, benchmark design, engineering decisions, educational interpretation, and public trust.
So, this article therefore ends with a deliberate conclusion: a system’s successful performance is evidence, not a universal diagnosis. Brain, mind, intelligence, and computation can be related in powerful theories, but the relationships must be argued and tested rather than smuggled in as synonyms.
That discipline creates the foundation for the incoming week. If STEAM Intelligence is to become a useful organizing idea for this series, it should be defined transparently as an editorial and research framework – not announced as a newly discovered cognitive faculty or established scientific field.
Internal-link suggestions
- Simulation as a Third Mode of Scientific Inquiry
- Physics Education Beyond Formula Memorization
- Creative Boundaries
- A Working Definition of STEAM Intelligence
References
1. Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59(236), 433-460. https://doi.org/10.1093/mind/LIX.236.433
2. Marr, D. (2010). Vision: A computational investigation into the human representation and processing of visual information. MIT Press. (Original work published 1982). https://doi.org/10.7551/mitpress/9780262514620.001.0001
3. Deary, I. J. (2012). Intelligence. Annual Review of Psychology, 63, 453-482. https://doi.org/10.1146/annurev-psych-120710-100353
4. Colom, R., Karama, S., Jung, R. E., & Haier, R. J. (2010). Human intelligence and brain networks. Dialogues in Clinical Neuroscience, 12(4), 489-501. https://doi.org/10.31887/DCNS.2010.12.4/rcolom
5. Kriegeskorte, N., & Douglas, P. K. (2018). Cognitive computational neuroscience. Nature Neuroscience, 21, 1148-1160. https://doi.org/10.1038/s41593-018-0210-5
6. OECD. (2024). Explanatory memorandum on the updated OECD definition of an AI system (OECD Artificial Intelligence Papers No. 8). OECD Publishing. https://doi.org/10.1787/623da898-en
7. Rescorla, M. (2026). The computational theory of mind. In E. N. Zalta & U. Nodelman (Eds.), The Stanford Encyclopedia of Philosophy (Summer 2026 ed.). https://plato.stanford.edu/archives/sum2026/entries/computational-mind/
8. Piccinini, G. (2025). Computation in physical systems. In E. N. Zalta & U. Nodelman (Eds.), The Stanford Encyclopedia of Philosophy (Fall 2025 ed.). https://plato.stanford.edu/archives/fall2025/entries/computation-physicalsystems/
9. Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3), 417-424. https://doi.org/10.1017/S0140525X00005756
10. Mitchell, M., & Krakauer, D. C. (2023). The debate over understanding in AI’s large language models. Proceedings of the National Academy of Sciences, 120(13), e2215907120. https://doi.org/10.1073/pnas.2215907120
11. Bean, A. M., et al. (2025). Measuring what matters: Construct validity in large language model benchmarks. Advances in Neural Information Processing Systems, 38. https://doi.org/10.52202/085713-0590
12. Ibrahim, L., & Cheng, M. (2026). Thinking beyond the anthropomorphic paradigm benefits LLM research. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics, 2551-2563. https://doi.org/10.18653/v1/2026.acl-long.118
13. Colombatto, C., & Fleming, S. M. (2024). Folk psychological attributions of consciousness to large language models. Neuroscience of Consciousness, 2024(1), niae013. https://doi.org/10.1093/nc/niae013
14. Colombatto, C., Birch, J., & Fleming, S. M. (2025). The influence of mental state attributions on trust in large language models. Communications Psychology, 3, 84. https://doi.org/10.1038/s44271-025-00262-1
15. Cogitate Consortium, Ferrante, O., Gorska-Klimowska, U., et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature, 642, 133-142. https://doi.org/10.1038/s41586-025-08888-1
16. Butlin, P., Long, R., Bayne, T., et al. (2026). Identifying indicators of consciousness in AI systems. Trends in Cognitive Sciences, 30(6), 488-501. https://doi.org/10.1016/j.tics.2025.10.011
17. Klatzmann, U., & Doerig, A. (2026). What biology can and cannot tell us about conscious AI. Trends Open. Advance online publication. https://doi.org/10.1016/j.treopn.2026.07.012
18. Wu, W., & Morales, J. (2025). The neuroscience of consciousness. In E. N. Zalta & U. Nodelman (Eds.), The Stanford Encyclopedia of Philosophy (Summer 2025 ed.). https://plato.stanford.edu/archives/sum2025/entries/consciousness-neuroscience/