/ Microsoft Research

Microsoft Research — Inclusive Futures in AI & Mixed Reality

Two independent research projects at Microsoft Research Cambridge exploring the ethical and experiential foundations of inclusive AI — from avatar authenticity in spatial computing to human-AI interaction design for Blind and Low-Vision users.

Jan – Apr 2022 · Cambridge, UK

Organisation
Microsoft Research, Cambridge, UK
Role
Research Lead & Strategist · Research & Co-Design Facilitator
Collaborators
ML Researchers, Engineers, Design Researchers, BLV Community
Scroll

This case study is written in a reflective format due to NDA restrictions. Visuals are being rebuilt separately as conceptual diagrams.

Human-AI Interaction for Find My Things

Research & Co-Design Facilitator · Cross-functional with ML Researchers & Engineers

Assistive AI tools can often provide feedback that is too dense or poorly timed, leading to cognitive overload for Blind and Low-Vision (BLV) users. Working alongside a specialised team of Machine Learning researchers and engineers, the goal was to research how AI-driven object recognition could be communicated through intuitive, non-visual interaction patterns.

I worked between the BLV community and the ML research team, running a co-design series to turn user feedback into requirements engineers could build.

Workshop execution

I co-facilitated three workshops with 10+ participants from the BLV community. We tested how users want to interact with an object-finding AI—balancing speed, certainty, and audio or haptic cues.

Research synthesis

Synthesised high-level qualitative data from workshops into actionable patterns for the technical team:

Key interaction principles defined

Auditory Hierarchy: Spatial orientation (direction and proximity) must be prioritised over verbal description — users need to know where before what.

Interaction Optimisation: Identified the specific cues users required to feel in control during search — helping the team understand how the AI should communicate its confidence levels.

Scalable Accessibility Patterns: Defined non-visual interaction patterns (audio and haptic) designed to reduce cognitive load for power users of assistive tools.

This foundational research was instrumental in the development of the "Find My Things" feature within the Seeing AI ecosystem. By centering the expertise of the BLV community and collaborating across technical disciplines, the research helped move the feature from a technical concept to a functional, human-centric assistive tool.

Inclusive Avatars in Mixed Reality

Research Lead & Strategist · 4-Month Research Project

In spatial computing, avatars often rely on standardised motion libraries that fail to represent the diverse physicalities of people with disabilities. This creates a "digital uncanny valley" where a user's virtual representation does not align with their lived physical identity — leading to exclusion in social Mixed Reality environments.

Core question

How do people with disabilities in Mixed Reality want to perform their avatar's bodily movements to support communication?

Decoding Movement Authenticity — with lived experts from the creative industry

Over four months, I led a deep-dive discovery phase to bridge the gap between human expression and technical architecture. I facilitated collaborative sessions and interviews with movement specialists, choreographers, and performance artists — experts with a professional vocabulary for non-verbal communication that technical teams typically lack.

Critical insight — Movement Authenticity

Authenticity in a digital space is not about "perfect" or "smoothed" motion. Through expert interviews, we defined Movement Authenticity as a qualitative result of how an individual's intent and physical presence are translated into a digital medium.

By working with artists with extensive experience of physical disabilities, we identified Signature Movements — the unique ways individuals occupy space and express identity through motion.

The Inclusive Research Handbook

A comprehensive strategic guide for engineers and designers, synthesising artistic and qualitative insights into a formal Taxonomy of Inclusive Movement. This provided a framework for future ML data collection and model generation — translating human experience into technical requirements.

Vision Videos

Conceptual storytelling assets demonstrating the potential of an "authenticity-first" approach to avatar design. These artefacts were produced to align product teams and senior leadership around a more inclusive vision for the future of Mixed Reality.

Goal

To define the foundational "ground truth" required to inform the ethical development of future machine-learning-generated avatars — ensuring that disabled users can be represented with integrity in spatial computing environments.

Working between BLV participants, choreographers, and ML engineers taught me how to turn lived experience into specifications a model team can build from.

Defining that human ground truth—and making it legible to an engineering system—is a habit I carry into every AI project I lead now.