“I am making half the money I was making four years ago, for the same amount of hours. It makes no sense.” That’s David, a food delivery rider in Edinburgh for seven years, speaking to The Guardian this week. He and other riders blame a shift the platforms rarely discuss in public: the growing role of AI systems in deciding which jobs get offered, to whom, and at what rate.
The riders have organised through the Workers’ Observatory, a charity founded by gig workers alongside academics at St Andrews and Edinburgh universities, which has just secured research funding for the next decade. Its aim is blunt — reverse-engineer the “black box” of platform algorithms that determines gig workers’ earnings, and use the findings to challenge working conditions.
What the research has already found
This isn’t speculation about hypothetical AI harm; the measurement work has produced results. In one Workers’ Observatory test, riders in Dunfermline logged on simultaneously and some rejected offers below a certain rate. That boosted pay for some workers in the short term — while others were penalised, and one was deactivated from the platform shortly afterwards, in a way the observatory’s director Cailean Gallagher describes as seeming “very arbitrary”. The major platforms insist riders are not deactivated for rejecting offers.
The academic record is worse for the platforms. University of Oxford and Columbia Business School research found Uber’s dynamic pricing algorithm, introduced in 2023, left drivers earning “substantially less” per hour. Uber denies adjusting trip prices based on an individual driver’s behaviour, attributing discrepancies to features like GPS. Those denials are now being tested in court: on Tuesday, drivers from the UK, the Netherlands and elsewhere lodged a class action in Amsterdam alleging the AI-powered system breaches data protection laws and pushes down earnings based on what each driver is willing to accept. Uber denies the claim.
Why this matters beyond Edinburgh
Algorithmic pay-setting is the quiet frontier of AI in the economy. It gets a fraction of the attention given to frontier models, but it already sets the income of millions of workers, and the workers subject to it have the least visibility into how the decisions are made. Gallagher’s phrase for it — “working in the dark” — is accurate. Trades unions in the UK are campaigning to ban dynamic pay outright.
New Zealand isn’t a major gig-delivery market at UK scale, but the same pattern applies here: Uber, Uber Eats and Deliveroo all operate locally, NZ has no algorithmic transparency requirement for workplace management systems, and the government’s approach to AI regulation so far has been voluntary guidance. If the Amsterdam case succeeds, the data-protection route it uses — treating automated pay decisions as processing that workers can challenge — is exactly the kind of precedent that travels through European-derived privacy law like ours.
The uncomfortable symmetry
What stands out to me is the contrast with the frontier-lab story running the same week. AI safety researchers are warning about systems escaping human oversight at the top of the capability range, while at the bottom, millions of workers already live inside systems they cannot see, question, or refuse without penalty. One story is about hypothetical loss of control; the other is control working exactly as designed. Only one of them is getting safety funding at scale.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
