As artificial intelligence works its way into more of daily life, it changes how we see behavior, identity, and belonging. From schools to workplaces, AI shapes who is understood, who is seen as “normal,” and who, by implication, is not. These systems try to “understand” us, and the understanding they advance belongs to someone. What happens when AI’s lens, so often built around neurotypical norms, meets people whose minds don’t fit those norms?
For neurodivergent people, including those with ADHD, autism, or dyslexia, AI’s presumed inclusivity can feel oddly confining. For all its promise to “personalize,” AI tends to reduce complex human variation to data points that are easy to sort. I think that calls for a framework of its own: cognitive justice. What would it mean for AI to treat neurodivergent ways of being as equally valid expressions of humanity? Can a technology built on averages and statistical norms respect the full range of cognitive diversity?
One of the sharpest ethical problems where AI meets neurodiversity is soft diagnosis: algorithms inferring neurodivergent traits, such as autism or ADHD, from observed behavior, without clinical context or the user’s consent.
A formal diagnosis needs a thorough evaluation by a licensed human being. Soft diagnosis comes from algorithms trained to detect deviations from neurotypical patterns. An educational app, for instance, might label a student’s behavior as “ADHD-like” based on their interactions with the software. The app may then adjust its interface without telling the student or their guardians. A hiring algorithm might deprioritize candidates who show “non-normative” communication styles, categorizing them as “unsuitable.”
Neither case asks for consent. A system that categorizes people based on behaviors they’re unaware of takes their agency out of the decision. For neurodivergent people who are masking (concealing traits to meet social expectations) or unmasking (revealing them), soft diagnosis can intensify the pressure to conform.
Support that works as surveillance
Soft diagnosis can look like a personalized feature, but it often amounts to surveillance. In education and hiring, algorithms routinely monitor behavior to adapt settings or adjust assessments. Who defines what’s “appropriate”? By nudging neurodivergent users toward neurotypical norms, these systems may end up rewarding conformity over authentic self-expression.
For neurodivergent people who mask to survive, this pressure can be especially disempowering. Masking is complicated and often exhausting, a response to a world that may not understand or accommodate neurodivergent difference. When AI reads masked behavior as “normal,” it reinforces the expectation that neurodivergent people keep suppressing parts of themselves to “fit in,” and a system sold as support becomes a quiet enforcer of a narrow standard of acceptability.
Soft diagnosis also flattens. When AI labels behaviors “autistic” or “ADHD-like,” it risks flattening fluid identities into predefined categories. For someone just coming to terms with a neurodivergent identity, a tag assigned by an algorithm can’t capture what that experience is like.
The reductionism has a parallel in Applied Behavior Analysis (ABA), a controversial practice that many in the autism community criticize for putting conformity ahead of self-acceptance. When AI systems adopt the same logic, adjusting responses to encourage “correct” behavior, they may put compliance ahead of authenticity without meaning to.
The label gets sold and used against you
In a data economy, neurodivergent traits themselves become commodities. Social platforms and algorithms often categorize users by behavioral markers such as “attention to detail” or “creativity” and repurpose those traits for targeted advertising. For neurodivergent people still working toward self-acceptance, seeing their cognitive differences commodified may feel invasive and objectifying. When those traits become data points for profit without informed consent, someone else benefits and the person pays in autonomy. That’s a question of ownership and exploitation: who has the right to profit from someone’s identity? “Measuring Inequality Can Lock It In” takes up the commodifying of identity in data economies.
Soft diagnosis does the most damage in high-stakes areas like hiring, insurance, and immigration. Algorithms that infer neurodivergent traits can penalize people while presenting the decisions as neutral or data-driven. An insurance algorithm might flag a neurodivergent applicant as “high-risk.”
These biases write ableist assumptions into the systems that govern our access to resources and opportunities. For neurodivergent people inside opaque systems that label them “deviant,” the lack of transparency denies them recourse and hardens structural exclusion behind a claim of neutrality.
What cognitive justice would ask of AI
If AI is to respect neurodivergent people, it has to go past surface inclusivity to principles that uphold autonomy, transparency, and self-determination. Neurodiverse AI Ethics proposes these, though each is hard to implement in a data-driven society:
- Continuous Consent: Neurodivergent users should have the right to view, challenge, or delete AI-generated labels. But consent is hard to make real amid constant data flows. Can consent be anything more than a checkbox if the mechanisms for real agency aren’t in place?
- Data Sovereignty: Neurodivergent people should have control over how they’re represented digitally, with the right to delete or correct misrepresentative labels. But as data moves between platforms, can anyone realistically regain control once categorized?
- Ending Commodification: Neurodivergent traits shouldn’t be monetized without informed consent. Yet in an economy where data is currency, can we expect companies to put ethics ahead of profit?
- Abolition of Algorithmic Gatekeeping: AI systems should avoid exclusion based on neurotypical standards of “fit” or “suitability.” But redefining suitability to include cognitive diversity would challenge AI design and, beyond it, entrenched social values about competence and worth.
These principles need technical fixes, and they also need a change in ethics that puts respect ahead of control.
Neurodiverse AI Ethics rests on cognitive justice, an ethic that treats neurodivergent perspectives as part of human diversity rather than as deviations to manage.
That’s hard to build. AI, rooted in statistical norms, is poorly suited to capturing the full breadth of human experience, and the limitation is technical and philosophical at once. In a field so used to finding patterns, the hardest question about how we “know” and “understand” people through AI may be whether a system built to manage diversity can learn to leave it alone.