Imperfect Education

Making observations of life is the most fundamental discovery process of biology. Today’s imaging technology shows scientists more detail than ever before, with vast and complex image sets produced in labs around the world every day. Analysing these data takes specialist skills and lots of time, and is subject to the individual analyst’s subjective perspective. Automating the process would speed it up and remove inconsistencies. Machine algorithms must be trained by examples, however, so it’s possible that they might replicate the subjectivity they aim to eradicate if trained by one expert’s examples. To investigate, researchers compared algorithms trained on a single expert’s work to those trained on many experts’ analyses (pictured, with different experts’ marks on a mouse brain image in different colours). Training the algorithm with many people’s input reduced subjectivity and improved outcomes, showing that many teachers are better than one.

Written by Anthony Lewis

Anthony Lewis

Writer

Ant was the web editor in the early days of BPoD, and has since worked with the UK’s top science organisations as a writer and multimedia producer. He now splits his time between freelance work and managing multimedia at The Lancet.

Website

• Image from work by Dennis Segebarth and Matthias Griebel, and colleagues
• Institute of Clinical Neurobiology, University Hospital Würzburg, and Department of Business and Economics, University of Würzburg, Würzburg, Germany
• Image originally published with a Creative Commons Attribution 4.0 International (CC BY 4.0)
• Published in eLife, October 2020