The Oxford student side hustle reshaping social science research

As a doctoral student at the University of Oxford, Phelim Bradley’s research focused on genomic sequencing, but he soon began to notice the difficulties faced by his social science peers.

Not only were DPhil students struggling to find adequate cohorts of survey respondents, but it was tricky for them to send follow-up questions or track shifting opinions over time, explained the Irish scientist-turned-entrepreneur, who founded Prolific in 2014 to tackle this problem.

Almost a decade later, the London-based company has established a 120,000-strong pool of respondents who are paid at least £6 an hour, but typically more, to answer questions – with the company shelling out more than £75 million to date. Last month the academic tech start-up announced it has raised £25 million of investment to expand its activities around artificial intelligence, with Google and Meta, as well as several top US and European universities, already among its clients.

“I was shocked by the quality of data tools that were available to academics and how challenging they were to use,” said Dr Bradley, who took an undergraduate degree in physics at University College Cork before a master’s in computational biology at the University of Cambridge. “I started Prolific in my first year at Oxford. It was a side project, but we built the first platform in three months and it was used mostly by colleagues.”

With 22,000 researchers using the site last month and a study launched on the platform every three minutes, Prolific has proved a huge hit with scholars, partly because it has addressed the so-called weird problem faced by European and US researchers, whose respondents are often entirely drawn from “Western, educated, industrialised, rich and democratic” societies and sometimes from the student population.

“This is the initial problem that we were trying to solve, to allow researchers to tap into a much wider audience and avoid these ‘weird’ samples,” said Dr Bradley. “We also wanted to make it really easy for researchers to access different samples but keep the costs low enough that a PhD researcher could still afford to use it,” he said.

Having a wider range of demographics within Prolific’s 120,000-strong respondent pool should help produce more robust results, but it is now an issue for AI models if they are trained using unrepresentative population samples whose feedback can bake biases into technology, continued Dr Bradley.

“We have helped to fine-tune a mental health chatbot, making sure it was safe, but also able to deal with a variety of different users,” he said.

By casting its net so wide, Prolific has also captured a pool of respondents with specialised knowledge that researchers would struggle to access, Dr Bradley added.

“If you’re doing research with lawyers or certain professionals, it can be incredibly difficult to reach these groups,” said Dr Bradley, who also hopes the platform will enable academics to participate more fully in AI-linked research, where the cost of technology and accessing users has made it impossible for academia to keep pace with the likes of Microsoft-backed OpenAI, the firm behind ChatGPT.

“It’s not obvious that the people going to win [in AI] are those with huge pre-existing research,” he predicted.

jack.grove@timeshighereducation.com