Neurodiversity & AI at Work: Risks, Bias, and Opportunity (A Practical Guide for Employers)
- Divergent Thinking
- May 15
- 3 min read
AI is already shaping recruitment, performance, learning, productivity tools, and day-to-day decision-making.
For neurodivergent people, that creates a double reality:
AI can be a powerful cognitive support tool.
AI can also hard-code bias and amplify barriers—especially when organisations treat “average user” as the default.
This post is a practical employer guide: where risk shows up, what to ask vendors, and how to use AI in ways that improve accessibility and performance rather than quietly excluding people.
If you want training for managers and teams that includes practical AI considerations (and avoids hype), start here:
The core risk: AI often optimises for “typical” behaviour
Many AI systems assume:
fast processing
consistent communication style
typical language patterns
typical attention/working memory
typical social signalling
But neurodivergent behaviour can be:
uneven (variable energy and focus)
non-linear (strong in depth, slower in switching)
direct, literal, or less “polished” socially
less predictable in speed (not capability)
If systems reward “polish” over substance, bias becomes structural.
Where bias shows up most in workplaces
1) Recruitment and selection
Risk patterns:
CV screening that rewards conventional wording
automated video interview analysis (voice, facial expression, “confidence” proxies)
timed online tests that measure coping speed more than job skill
chatbots that penalise direct or atypical responses
What to do:
avoid automated emotion/face-scoring as a decision tool
offer alternative formats and extra time where appropriate
validate tools for accessibility, not just “accuracy”
ensure a human review step exists for adverse decisions
2) Performance and productivity analytics
Risk patterns:
measuring “activity” (messages, meetings, camera-on time) as performance
tracking speed of response as “engagement”
penalising different work rhythms
What to do:
measure outcomes, not social visibility
separate communication style from competence
ensure performance standards are explicit and observable
3) Learning and knowledge systems
Risk patterns:
content that is too dense, too fast, too text-heavy
AI summaries that miss nuance or context
tools that privilege one reading style
What to do:
provide multiple formats (summary + detail + audio)
use clear structure, headings, and examples
treat accessibility as a core requirement, not an “add-on”
4) Internal comms and “helpful” automation
Risk patterns:
AI rewriting that removes meaning or tone
over-automation that increases ambiguity (“it sounds nice but says nothing”)
reliance on chat-based tools that create constant interruptions
What to do:
adopt a clarity standard (deliverable / deadline / definition of done)
use AI to make communication more checkable—not more generic
protect focus time; don’t let AI increase message volume
The opportunity: AI as a cognitive support tool (when used well)
AI can help neurodivergent staff reduce cognitive load, especially in:
summarising long documents or meetings
converting notes into action lists
drafting first versions (emails, outlines, reports)
creating step-by-step plans from vague tasks
simplifying language without losing meaning
generating checklists and templates
translating information into different formats
Key principle:
AI should reduce friction and externalise memory—not replace judgement.
A practical “neuroinclusive AI” checklist for employers
Use this when adopting AI tools or policies.
1) Accessibility by design
Can users adjust pace, format, and outputs?
Are there alternatives to time pressure?
Is the UI usable for different sensory and cognitive needs?
2) Avoid proxy metrics
Does it infer “engagement” from behaviour (camera, response speed, sentiment)?
Does it score “professionalism” or “confidence” without job relevance?
If yes, high risk.
3) Transparency and control
Can users see what data is captured?
Can they correct or contest outcomes?
Is there a human review route?
4) Fairness testing
Has it been tested for disability/neurodiversity impacts?
What evidence exists beyond vendor claims?
5) Clear boundaries
Where is AI advisory vs decision-making?
Who is accountable for outcomes?
If you can’t answer these, you’re not ready to deploy at scale.
What to ask vendors (copy/paste questions)
“What accessibility testing have you done, specifically around disability and neurodiversity?”
“What behaviours does your model treat as ‘high quality’—and are those job-relevant?”
“Do you use proxies like facial expression, tone, sentiment, response speed, or camera time?”
“How do users request adjustments or alternative assessment formats?”
“What is the appeal process for an adverse decision?”
“What data is stored, for how long, and who can access it?”
“Can you provide documentation of bias evaluations and limitations?”
A simple internal policy that reduces risk fast
If you want an immediate improvement, adopt these five rules:
AI must not be the sole basis for hiring or performance decisions.
Any automated screening has a human review pathway.
Employees can request alternative formats and reasonable adjustments for AI-mediated processes.
Performance is measured by outcomes, not digital visibility.
AI tools should reduce cognitive load (summaries, checklists, templates), not increase message volume.
These are practical, not ideological. They’re risk management.
Want training that covers AI risks and neuroinclusion?
If you want a session for leaders, HR, product teams, or managers on AI risks, bias, and practical opportunity (with clear actions), explore training options here:
Comments