I use AI tools in my writing. This post, like several others on this blog, was written with LLM assistance — research, structure, drafting, revision. If you run any of these posts through an AI writing detector, you will likely receive a high probability-of-AI score. The detector will be picking up something real.
It will also be wrong about what that means.
The Constitution Problem
In 2023, as universities began deploying AI detection tools at scale, educators started testing them on texts that were definitively not AI-generated. The results were instructive. The United States Constitution received high AI-probability scores from multiple commercial detectors. GPTZero returned a rating of 92% likely AI-written. The Federalist Papers fared similarly. So did sections of the King James Bible and Kant’s Critique of Pure Reason. Historical documents, written by humans, for human purposes, in an era when no AI existed — flagged as machine-generated.
This was not a marginal edge case. It was consistent across tools and across documents. And while it was widely reported as evidence that the detectors were broken, there is a more precise reading available: the detectors were working correctly, and we had misunderstood what they were measuring.
What the Detectors Actually Measure
Most commercial AI detectors — GPTZero, Turnitin’s detection layer, Copyleaks — use some combination of two statistical signals.
Perplexity. A language model assigns a probability to each token given the preceding tokens. Low perplexity means the text was, token by token, what the model expected — it sits close to the centre of the probability distribution. AI-generated text tends to have low perplexity because that is precisely what generation does: it samples from the high-probability region of the distribution [1]. Human text, on average, has higher perplexity, because humans write for specific contexts with idiosyncratic word choices, rhetorical effects that require the unexpected, and the accumulated noise of composing for a real reader.
Burstiness. A term introduced by Edward Tian, GPTZero’s creator: human writing has high burstiness — sentence lengths vary widely, vocabulary density shifts, complex constructions alternate with simple ones. AI writing is more uniform. The statistical distribution of sentence lengths in LLM output is narrower than in most human prose [2].
The underlying assumption these tools share: human writing is variable, contextually messy, idiosyncratic. AI writing is smooth and predictable.
This is accurate for a large class of human writing — casual prose, personal essays, social media, student writing in informal registers. It is wrong about a different and well-defined class of human writing. The Constitution sits in that class. So does a lot of other text.
The Systemising Brain
Simon Baron-Cohen’s empathising–systemising (E-S) theory distinguishes two cognitive orientations. Empathising involves attending to social and emotional cues, inferring mental states, navigating the pragmatic, implicit layer of communication — what is meant rather than what is said. Systemising involves attending to rules, patterns, and underlying regularities — the drive to understand how things work and to represent them in explicit, transferable, internally consistent terms [3].
Both orientations are distributed across the human population. They are not exclusive, and neither is pathological. But autism spectrum conditions are robustly associated with high systemising and relatively lower empathising — not because autistic people lack emotions or care about others, but because the cognitive mode that comes naturally to them is one of rules, structures, and explicit representation rather than social inference and pragmatic implication. The intense world theory [4] adds a complementary perspective: autistic brains may be characterised by hyper-reactivity and hyper-plasticity, with pattern-seeking and systematising serving partly as a way of making a too-intense world navigable. The systematicity is not a deficit. It is an adaptation.
This has direct consequences for writing.
High-systemising writing tends toward:
Consistent vocabulary. The same term is used for the same concept throughout, because substituting a synonym introduces ambiguity about whether the referent is actually the same. Neurotypical writing freely uses synonyms for stylistic variety; systemising writing resists this on principle.
Explicit logical structure. Claims are supported by stated reasons rather than left to pragmatic inference. If there are three conditions, all three are named. Nothing is “needless to say.”
Low social hedging. Phrases like “as everyone knows” or “obviously” are avoided, because they perform social alignment rather than convey information — and they depend on shared assumptions the writer is not confident are actually shared. (This connects to a point I made in the car-wash-walk post about Gricean pragmatics: autistic communication often violates the maxim of quantity in the direction of over-informing, because nothing is assumed implicit.)
Grammatical parallelism. Parallel logical content takes parallel grammatical form. This is not stylistic affectation; it is a natural consequence of representing structure explicitly.
Minimal rhetorical noise. The prose does not meander, warm up, or perform relatability. It states what needs to be stated.
Now run text with these properties through an AI detector. Consistent vocabulary reads as low lexical diversity. Explicit structure reads as low burstiness. Minimal rhetorical noise reads as smooth, generated output. The detector is measuring these properties accurately. The attribution to machine generation is where it goes wrong.
Liang et al. [5] demonstrated a closely related failure empirically: AI detectors are significantly more likely to flag writing by non-native English speakers as AI-generated. Non-native writers at advanced levels of formal English tend to write more carefully, more consistently, and more in accordance with explicit grammar rules — because they learned the language as a system of explicit rules rather than acquiring it through immersive social exposure. More systematic writing: higher AI probability score. The mechanism is the same. The population is different.
The Physicist Brain
Physics writing has its own conventions, independently developed but pointing in the same direction.
Scientific prose requires defined terms used consistently: in a paper about quantum error correction, “logical qubit,” “physical qubit,” and “syndrome” each mean exactly one thing, used identically in section 2 and section 5. It requires explicit assumptions: “We assume the noise is Markovian.” “In the limit of large N.” These are not vague hedges; they are precise statements about the domain of validity of the results. It requires logical derivation over rhetorical persuasion: the connectives are “since,” “therefore,” “it follows that” — explicit logical operators, not narrative bridges. And the passive construction of “the signal was measured” rather than “I measured the signal” removes the individual from the result, because the result should be reproducible regardless of who performs the measurement.
The outcome is prose that is systematic, consistent, and structurally predictable. From the outside — and from the vantage point of an AI detector — it looks machine-generated.
Paul Dirac is the physicist who comes to mind first here. His 1928 paper deriving the relativistic wave equation for the electron contains almost no rhetorical apparatus. Motivation, equation, consequence: each stated once, clearly, with no warm-up and no elaboration beyond what the argument requires. It is not warm. It is not discursive. It is beautiful in the way that a proof is beautiful: every element earns its place. Run it through GPTZero and see what you get.
This connection between the physicist’s prose style and the autistic cognitive mode is not accidental. Baron-Cohen et al. [6] surveyed Cambridge students by academic discipline and found that physical scientists and mathematicians scored consistently higher on the Autism Quotient (AQ) than humanities students and controls, with mathematicians scoring highest of all. The systemising orientation associated with autism spectrum conditions is also associated with — and presumably selected for — in quantitative scientific disciplines. The physicist’s prose reflects this. So does the writing of a high-systemising person who has never studied physics.
The categories overlap without being identical. What they share is a cognitive preference for explicit structure, consistent vocabulary, and logical transparency over social performance and rhetorical persuasion. The writing that emerges from that preference looks, to an AI detector, like it was generated by a machine.
It was not.
The Category Error
The error AI detectors make is not a measurement error. It is a category error.
They are trained to distinguish two things: output generated by a contemporary LLM, and a specific subset of human writing — typically casual, personal, or student prose collected from online sources. When they encounter text outside either of those training categories — systematic and precise but human-generated — the classifier has no good option. The text does not match the “AI” training data exactly, and it does not match the “human” baseline either. It gets assigned to the bin it fits least badly.
What is happening when the Constitution is flagged: it is systematic, definitional, prescriptive, and internally consistent. It was written by lawyers and statesmen who understood that ambiguity in foundational documents creates legal chaos. They wrote to be unambiguous. The result is text with low perplexity and low burstiness — the statistical signature the detector associates with AI.
GPTZero’s creator Edward Tian acknowledged this problem when it was reported: the Constitution appears so frequently in LLM training data that it registers as “already known” to the model, which artificially lowers its perplexity score. That is a real and specific issue. But it is secondary. The deeper issue is that the Constitution would score low-perplexity even without the training-data contamination effect, because systematic, definitional prose is intrinsically low-perplexity. Precise language is predictable language. That is partly the point of precise language.
The baseline assumption — that human writing is variable and idiosyncratic — holds for much human writing. It does not hold for legal drafting, technical documentation, scientific papers, sacred and historical texts written to be durable and precise, writing by people with high systemising orientation, or writing by non-native speakers at formal registers. That is not a small population of edge cases. It is a substantial fraction of all written material that exists.
Right and Wrong at the Same Time
So: if you think these posts are AI-generated, you are right and wrong at the same time.
Right, in two ways. First: yes, I use AI tools. LLM assistance is part of my writing process — not an occasional aid, but a regular part of how research notes and half-formed arguments become structured posts. Second: the writing style of these posts is systematic and precise in ways that detectors register as machine-generated. That systematicity is real, and if a detector picks it up, it is measuring something.
Wrong, also in two ways. First: the ideas, judgments, and connections in these posts are mine. The decisions about what to include and what to leave out, which papers to cite and how to frame their implications, where the interesting tension lies between neurodiversity research and the assumptions baked into AI detection tools — those are not outputs of a language model working in isolation. They are the product of someone who works at the intersection of these fields and has thought about them for a while. An LLM cannot generate these posts without a human who has already decided what to say.
Second, and more important for the argument here: the systematic, precise character of this writing is not evidence of machine generation. It is a cognitive signature — one associated with physics training, with high systemising orientation, with the overlap between those two things that I have written about elsewhere in the context of neurodiversity more broadly.
The detector is measuring a real property of the text. It is misattributing the origin of that property.
The interesting question this opens is not “did AI write this?” That question is increasingly poorly posed in an era where thinking and writing are already deeply entangled with machine assistance, in ways that differ sharply from person to person and task to task. The better question is: whose judgment is in the text? Whose choices about what to include, what to connect, what to leave out?
The systematicity in this writing is mine. The recognition that AI detectors systematically disadvantage autistic writers, physicist writers, and non-native speakers is a judgment I made, not one a language model was prompted to produce. The connection to the Constitution — a document written to be maximally unambiguous, flagged as maximally AI-like — is a connection I found worth drawing.
Whether that makes this text “human” is a philosophical question I am happy to leave open. What it is not is AI hallucination.
References
[1] Mitchell, E., Lee, Y., Khazatsky, A., Manning, C. D., & Finn, C. (2023). DetectGPT: Zero-shot machine-generated text detection using probability curvature. Proceedings of the 40th International Conference on Machine Learning (ICML 2023). https://arxiv.org/abs/2301.11305
[2] Gehrmann, S., Strobelt, H., & Rush, A. M. (2019). GLTR: Statistical detection and visualization of generated text. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, 111–116. https://doi.org/10.18653/v1/P19-3019
[3] Baron-Cohen, S. (2009). Autism: The empathising–systemising (E-S) theory. Annals of the New York Academy of Sciences, 1156(1), 68–80. https://doi.org/10.1111/j.1749-6632.2009.04467.x
[4] Markram, K., & Markram, H. (2010). The intense world theory — a unifying theory of the neurobiology of autism. Frontiers in Human Neuroscience, 4, 224. https://doi.org/10.3389/fnhum.2010.00224
[5] Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
[6] Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The autism-spectrum quotient (AQ): Evidence from Asperger syndrome/high-functioning autism, males and females, scientists and mathematicians. Journal of Autism and Developmental Disorders, 31(1), 5–17. https://doi.org/10.1023/A:1005653411471