​A web-based plugin which addresses misleading numerical claims in online media reports earned three Stellenbosch University alumni first place in the Cambridge Misinformation Hackathon recently.

Gregor Feierabend, Josephine Rey and Luca Powell, who are all enrolled for the MPhil in Advanced Computer Science at Cambridge University, are SU alumni in respectively mathematics, computer science and electrical engineering respectively.  All three of them are also fortunate to be on full scholarships: Gregor is a Gates Cambridge Scholar, while Josie is an awardee of the Patrick & Margaret Flanagan scholarship, and Luca is a Skye Foundation scholar.

According to Gregor, the trio have known each other since day one of their undergraduate studies at SU: “We have jointly participated in several competitions already, which probably makes us such a good team. In our first year, Josephine and I were part of the winning team of the 2019 Stellenbosch Hackathon. All three of us are also regular participants in the Harvard Puzzle Day, a fun competition hosted online by the CS50 Computer Science course at Harvard University". 

“Each of us contributes with different skills and different backgrounds to our team. The Cambridge Misinformation Hackathon emphasises the importance of interdisciplinary work - not only within computer science, but also in the humanities and social sciences," Gregor adds.

The Misinformation Hackathon is organised every year by Students Against Pseudoscience at Cambridge and supported by the Centre for Research in the Arts, Social Sciences, and Humanities (CRASSH), also at Cambridge University.

According to the team, a significant source of misinformation lies in how numerical figures are reported on my official media outlets: “Probability principles are often misunderstood by the general public, and this is taken advantage of by the media," they explain in their submission to the competition.

For example, in one report bacon consumption is linked to a 20% increase in risk of pancreatic cancer. “This seemingly alarming increase does not, however, reflect the actual risk of initially getting pancreatic cancer: 1 in 80 people. A 20% increase to this initial rate would be 1.2 out of 80 people, which translates to 6 out of 400 instead of 5 out of 400 people getting pancreatic cancer, which is not as severe," they continue.

The team proposed a web-based plugin which can identify misleading numerical claims in online news articles and highlight these. When a reader then hovers over these highlights, an overlay clarifies why the claims may be misleading and converts any misleading numbers to an easily understandable format. For example:

• Relative claims are converted to a standardised format, e.g., “5 in a 1000" becomes “0.5%";
• Relative changes are flagged as potentially misleading and the user is urged to do further research to find the correct context in which the claim is made, e.g., “50% fewer car accidents" could mean a decrease from 2 to 1 or from 200 to 100 incidents.
• Rankings of absolute quantities are flagged as potentially misleading and the reader is urged to compare these with relative rankings, e.g., “more people die by car accidents than by shark attacks" might be misleading as it does not mention the relative risk of injury".

According to Gregor the MPhil in Advanced Computer Science at Cambridge is half taught, half research-based and covers a wide range of subjects.

 Josephine majored in BSc Computer Science at SU, followed by a BSc Hons in Information Technology at the University of Cape Town. At Cambridge, she is doing research to support Machine Learning and Artificial Intelligence in education in Africa. Luca completed his BEng degree in Electrical and Electronic Engineering at SU and is now implementing federated learning on satellite constellations for his research project. Gregor, who majored in BSc and BSc Hons Mathematics at SU, is doing research on the mathematical foundations of computer science to develop formalisms that allow us to prove software correct. 

​