Skip to content
Ndugelo Trevar PhophiAnthropology · Human Behaviour · AI

Projects

A research programme, not a portfolio.

Each project is documented the way a laboratory documents its work: the question it answers, why it matters, how it is studied, what it is finding and where it leads next.

Browse by theme

3 of 6 projects

Project 01

Current — Honours Dissertation, 2026

Algorithms of Trust

A desktop-based qualitative study examining how South Africa's national AI policy discourse imagines its users, and how that imagined user compares to the narratives South Africans actually produce about AI in public digital spaces.

How do the assumptions about "the user" embedded in South Africa's AI policy discourse compare with the lived narratives and concerns South Africans express about artificial intelligence online — and what happens when those two accounts diverge?
Why it matters
Policy that misjudges who its users are and what they can do risks excluding the very people it claims to serve. This project surfaces that gap directly, using South Africa's own recent AI policy process as the case.
Methods
Critical discourse analysis (Norman Fairclough)Comparative policy document analysis: 2024 Framework and 2026 Draft PolicyPublic digital narrative collection from South African online spaces
Findings
Institutional assumptions about AI users grew more attentive to inequality between the 2024 Framework and the 2026 Draft Policy, but both still presuppose a level of digital literacy and verification capacity that most South Africans do not have. This gap was underscored when the Draft AI Policy itself was withdrawn in April 2026 after it was found to contain AI-hallucinated citations.
Impact
A grounded, current case study of the distance between policy imagination and public experience in South African AI governance, with direct relevance to how digital-literacy assumptions shape technology policy more broadly.
Future implications
Lays groundwork for the author's planned Master's research extending this line of inquiry into trust between banks, humans, and AI systems.

Project 02

Active · Honours research

Negotiating Trust in Artificial Intelligence

An ethnographic study of how people arrive at — and withdraw — trust in AI-mediated services, treating trust as an ongoing negotiation with institutions rather than an attitude towards a tool.

How do people negotiate trust in artificial intelligence when they cannot inspect, contest or fully understand the system deciding on their behalf?
Why it matters
Adoption strategies routinely assume that better performance produces confidence. In practice people calibrate trust against institutional history, perceived recourse and the conduct of the person in front of them.
Methods
Semi-structured interviewsParticipant observationThematic and narrative analysisReflexive field notes
Findings
Emerging: trust attaches to accountability, not accuracy. Participants ask who answers for the outcome long before they ask how the system works.
Impact
Gives institutions a language for trust that survives contact with real users, and a way to diagnose failures that no performance metric surfaces.
Future implications
Extension into a comparative Masters study across health, education and financial services.

Project 04

In development

Digital Health and University Students

A study of how young adults use digital health tools, symptom checkers and AI chat interfaces as a first, informal layer of care.

What do students do with digital health information before, instead of, or alongside seeing a clinician?
Why it matters
This population is treated as digitally fluent and therefore unproblematic. Fluency and safety are not the same thing.
Methods
Focus groupsDigital diary methodsSurvey instrumentWalkthrough interviews
Findings
Planned.
Impact
Evidence for campus health services and digital health providers designing for actual help-seeking behaviour.
Future implications
Longitudinal follow-up across an academic year.