Training Tech Teams for Inclusive AI

Event graphic titled "Training Tech Teams for Inclusive AI" for World Usability Day. The image features two circular headshots: James Thurston (Head of Accessibility & Digital Inclusion for North America, Atos US) and Michaël Vanderheyden (Lead UX Engineer Central Europe, Atos Germany). The slide includes IAAP certification badges (CPACC, CPWA, KB-BFL), the World Usability Day logo, and the Atos logo against a blue and white background.

Event: Texas DIR World Usability Day
Organizer: Texas Department of Information Resources (DIR)
Date:
Location: Austin, Texas / Virtual

I am thrilled to be (virtually) in Austin today for the Texas Department of Information Resources (DIR) World Usability Day celebration!

It was a true honor to step in for my colleague Beatriz González Mellídez today and join James Thurston on stage to discuss one of the most pressing challenges in our industry: Training Tech Teams for Inclusive AI and Better Accessibility.

Accessibility Level Setting: A Must for Some, Good for EVERYONE

Inclusive AI isn’t just a buzzword; it’s a design imperative. Before diving into training strategies, we grounded the conversation in a shared understanding of accessibility — not as a compliance checkbox, but as a fundamental quality attribute that benefits every user in every situation.

Trainings and Good Practices: Embedding Accessibility — Shared Roles & Responsibilities

Real inclusion happens when accessibility becomes a shared responsibility across the entire team, not just a task delegated to a specialist. In this section, James and I explored how to embed accessibility into daily workflows through targeted training, clear role ownership, and a culture of continuous improvement. The goal: shift teams from reactive remediation to proactive, empathy-driven design.

10 Game-Changing Steps Towards Accessible GenAI

When it comes to Generative AI, accessibility doesn’t happen by default — it has to be designed in. We walked through a practical checklist of 10 actionable steps every team can take to build more inclusive AI-powered products:

  1. Inclusive Sources: Train models against bias by using diverse, representative datasets.
  2. Include laws & policies in training: Embed legal and regulatory accessibility requirements directly into model training.
  3. Ensure full keyboard interaction: Every AI-generated interface or interaction must be fully operable without a mouse.
  4. Use descriptive titles: Clear, meaningful headings and labels help all users navigate AI-generated content.
  5. Ensure authentic representation in generated media: Avoid stereotypes and ensure generated images and videos reflect real human diversity.
  6. Tag and label elements properly: Semantic markup and ARIA attributes are just as important in AI-generated output as in hand-crafted code.
  7. Simplify language and tag it correctly: Plain language benefits everyone; pair it with proper lang attributes for screen reader accuracy.
  8. Follow WCAG’s color contrast standards: AI-generated UIs must meet minimum contrast ratios to be readable for users with low vision.
  9. Provide text alternatives for audiovisual media: Captions, transcripts, and audio descriptions are non-negotiable for AI-generated video and audio content.
  10. User testing with people who have lived experience: No checklist replaces feedback from real users with disabilities — involve them early and often.

Conclusion and Q&A

We wrapped up with an open Q&A, fielding questions from the audience on everything from practical tooling to organisational change management.

A huge thank you to the Texas DIR for hosting such a vital conversation and to James for being a fantastic co-presenter.

Let’s keep making the digital world accessible for everyone!