The Myth of the “Dumbing Down” Machine
How AI Studies Erase Neurodivergent Minds and Why Assistive Intelligence is an Act of Cognitive Justice
💡 If you enjoy deep dives into neurodivergence, cognition, and the ethics of technology, consider becoming a paid subscriber to support future essays like this one.
1. The Unexamined Bias Behind “AI Makes You Dumber”
The claim that “AI makes people dumber” is catchy — but it rests on unexamined assumptions about how thinking works, what counts as effort, and who the “people” actually are.
Cognition is not uniform. Not everyone begins with the same baseline of working memory, executive function, or sensory bandwidth. Yet most research treats human thought as a single-track, linear system that AI might either strengthen or weaken.
For autistic and neurodivergent individuals — especially Gestalt Language Processors (GLPs) — that simplification is deeply flawed. AI can, in fact, reduce the energy cost of communication and participation. When designed well, it functions as an accessibility tool, not a cognitive crutch.
This essay explores how current research misunderstands cognitive load, ignores neurodivergent realities, and assumes an impossible baseline of “equal mental effort.” It also reframes AI not as a threat to intelligence, but as a technology of liberation — one that can reduce invisible strain and expand access to expression.
2. What the Studies Actually Measure — and What They Miss
Most “AI harm” studies take two forms:
Self-report surveys (e.g., Microsoft or Carnegie Mellon): asking knowledge workers if they feel less focused when using AI.
Task-based experiments (e.g., MIT): comparing participants’ essay writing or reasoning with and without AI assistance, sometimes paired with EEG scans to measure brain activity.
Researchers typically conclude that AI users exert less effort or display lower neural activation. That’s taken to mean “AI reduces engagement,” which morphs into “AI makes you dumber.”
But here’s the problem: less activation does not equal less intelligence. It can mean efficiency — or that participants are offloading unnecessary strain. Experienced chess players show less brain activation than beginners because their brains have learned to optimize the task. Efficiency is not decay.
Even more critically, these studies assume that everyone begins with the same cognitive load. They ignore how neurodivergence reshapes sensory processing, working memory, and executive function. That oversight makes the research inherently biased.
3. Cognitive Load Is Not Equal for Every Brain
Cognitive load theory divides mental effort into three parts:
Intrinsic load — the inherent complexity of a task
Extraneous load — distractions, poor design, or sensory clutter
Germane load — the effort devoted to learning and meaning-making
Most studies only consider the first two. But autistic individuals often carry a baseline extraneous load before any task begins: managing sensory overwhelm, filtering input, self-monitoring, or masking.
In one computational neuroscience study, autistic participants exhibited roughly 42% greater information gain (a measure of internal signal complexity) than neurotypical controls — a possible reflection of heightened sensory and cognitive processing at rest.
If we all start at different baselines, demanding that autistic people demonstrate equal output with equal effort is not scientific fairness — it’s systemic bias disguised as rigor.
Autistic thinkers shouldn’t have to exhaust themselves to prove they’ve worked hard enough to deserve intellectual legitimacy.
4. The Hidden Energy Economy of Neurodivergence
Cognitive effort is not just abstract; it’s physical and metabolic.
Autistic and ADHD individuals often speak of energy management — the cost of “ordinary” tasks like organizing, transitioning, or filtering background noise. These invisible expenditures accumulate and lead to what’s called autistic burnout: profound depletion after prolonged overextension.
When AI tools help reduce this load — automating transcription, suggesting structure, simplifying wording — they aren’t “making people lazier.” They’re redistributing energy toward creativity, connection, or meaning.
That redistribution is not a moral failure. It’s accessibility.
5. Gestalt vs. Linear: Why Language Processing Differences Matter
Much of the AI-as-cognitive-aid debate assumes that everyone writes and thinks linearly: word by word, outline first, argument next.
But many autistic people are Gestalt Language Processors (GLPs) — they think and speak in chunks, scripts, or intuitive wholes. Their brains store and reuse full phrases (“gestalts”) rather than building every sentence from scratch. Over time, these are unpacked and remixed into unique expression.
Researcher Marge Blanc’s Natural Language Acquisition framework describes how GLPs progress from echolalic wholes to analytic flexibility. It’s not a deficit model — it’s a different architecture of language.
When writing or communication tasks demand purely analytic, linear construction, GLPs face double work: translating holistic thought into fragmented syntax. That added friction inflates their cognitive load — which most AI studies never account for.
6. How AI Can Reduce Cognitive Load and Expand Access
Instead of amplifying inequality, AI can neutralize it — if designed with awareness.
AI can reduce several kinds of mental strain:
Executive load: planning, sequencing, outlining
Mechanical load: spelling, punctuation, formatting
Retrieval load: word-finding, recall under pressure
Social load: interpreting tone or emotion in communication
AI tools like TwIPS (Texting with Interpret, Preview, Suggest) already demonstrate this. In trials with autistic users, TwIPS helped clarify tone and intention in texts while preserving autonomy — users remained the authors, simply aided by structure.
Similarly, Autiverse, an AI-guided multimodal journaling app, helps autistic teens narrate their daily lives using visuals and conversational scaffolds, improving coherence without draining energy.
For Gestalt processors, AI can:
Recognize and respect gestalt phrases
Offer remixing suggestions instead of forced corrections
Preserve prosody and rhythm of scripted language
Help unpack gestalts gradually, on the user’s terms
Used well, AI becomes not a replacement for thought but a translator — a bridge between internal gestalt cognition and the external world.
7. When “Less Effort” Is Actually Liberation
To a neurotypical observer, AI might look like a shortcut.
To a neurodivergent user, it can feel like finally breathing.
Efficiency is not decline. It’s adaptation.
And accessibility is not indulgence. It’s equity.
Every brain uses tools — from eyeglasses to spellcheckers — to interface with the world. AI simply extends that lineage. What matters is whether we build it with empathy for difference or with suspicion toward it.
8. Toward a More Inclusive Cognitive Future
If we want honest science, our studies must evolve.
Researchers must:
Recruit neurodivergent participants and measure sensory/energy load.
Analyze performance through multiple processing styles (linear and gestalt).
Track fatigue and persistence, not just task accuracy.
Abandon the assumption that identical effort equals fairness.
Educators and policymakers must:
Accept AI tools as valid accommodations.
Evaluate ideas and synthesis, not mechanical labor.
Design flexible formats for writing, testing, and communication.
Technologists must:
Involve neurodivergent users in co-design.
Make AI features modular and user-controlled.
Ensure privacy, consent, and dignity in every interaction.
When we design for difference, we create tools that make everyone freer.
9. Conclusion: Cognitive Justice as the Next Frontier
The real danger isn’t that AI makes us dumber.
It’s that we keep designing research and policy as if all minds were the same.
Autistic and neurodivergent thinkers should not have to prove intelligence through exhaustion. They should not have to sacrifice their energy to satisfy an experiment’s idea of “effort.”
AI, when used intentionally, can redistribute that energy — away from survival, toward creation and contribution. It doesn’t cheapen thought; it frees it.
Let’s stop asking whether AI makes us less human, and start asking how it can make humanity broader.
✨ If this essay resonated, share it with someone who believes intelligence should have only one shape.
References / Further Reading
Velázquez, R., & Galán, R. F. (2013). Increased information processing in autistic brains: A computational neuroscience study. Frontiers in Neuroinformatics
Blanc, M. (2023). Using the Natural Language Acquisition Protocol to Support Gestalt Language Development. ASHA Perspectives
Rukhshan, M. et al. (2023). TwIPS: Texting with Interpret, Preview, Suggest — AI for Autistic Communication Support. PDF Link
ArXiv. (2025). Autiverse: AI-Guided Multimodal Journaling for Neurodivergent Adolescents. ArXiv Link
Frontiers in Psychiatry (2025). Processing differences in gestalt vs analytic language learners. Frontiers in Psychiatry