We talked to Erik Duhaime of Centaur Labs about medical data labelling and the impact of COVID-19.
First of all, how are you and your family doing in these COVID-19 times?
Erik Duhaime: We’re doing well. It was a little scary in the beginning, especially because my wife is a doctor and her hospital was at capacity during the surge in Boston. Now that things have calmed down, we are feeling good, and the team has done a great job adjusting to the Covid-era.
Tell us about you, your career, how you founded Centaur Labs.
Erik Duhaime: Most of my career I’ve been focused on one question, ‘how can a group of people be more effective than the sum of its parts’
I researched the evolution of human cooperation in my undergrad and master’s theses at Brown and Cambridge University. Once I started my PhD the MIT Center for Collective Intelligence, I became more interested in how information technology could enable new ways of working together.
In some of my PhD research, I found that combining the opinions of people with each other and with Artificial Intelligence algorithms could achieve superior accuracy for analyzing images of skin lesions for cancer. Not long after that, I decided to found Centaur Labs in 2017, which is focused on harnessing collective intelligence to improve healthcare for all. Our current focus is to offer highly accurate medical data labeling for AI companies.
How does Centaur Labs innovate?
Erik Duhaime: Today, most medical decisions are made by individuals who are trusted by their credentials. Diving into the research literature, it is actually quite scary how often medical experts make mistakes and disagree with each other.
Part of the problem is that experts don’t necessarily know their true comparative advantage at different tasks. For instance, if you ask a breast radiologist if they are good at reading a mammogram, they will say yes, but don’t really know how they perform compared to their peers and what their blind spots are. In contrast, we leverage a network of medical experts to gather multiple opinions per case and intelligently aggregate them–based on past performance–to yield a more accurate result. This is the first of many approaches that tie to our vision of creating a global network of people and machines that are trusted based on performance metrics to solve medical problems.
How the coronavirus pandemic affects your business, and how are you coping?
Erik Duhaime: The team definitely misses being together in the office, but all things considered we feel relatively fortunate as a business. This is not only because some industries have been hit terribly hard, but also because we think that in the long run, COVID will accelerate trends like the acceptance of AI and new ways of decision making in healthcare.
Did you have to make difficult choices, and what are the lessons learned?
Erik Duhaime: We were in the process of hiring for multiple positions right when COVID hit and the uncertainty at the time made us question if it was the right decision. In the end, we pushed ahead and brought on two fantastic employees during the quarantine. While onboarding people remotely was a bit tricky, these employees have integrated well into the team, and we’ve continued to make great progress as a result.
How do you deal with stress and anxiety? How do you project yourself and Centaur Labs in the future?
Erik Duhaime: Startups can certainly be stressful from time to time. I find one of the best ways to deal with it is by having a dedicated team all aligned around the same core mission. That way, even through the ups and downs, we have this confidence that as a team, we can overcome any challenge in our path. COVID has actually helped with this in that it has instilled a greater sense of responsibility to change healthcare for the better.
Who are your competitors? And how do you plan to stay in the game?
Erik Duhaime: Data labeling is a very hot space right now, driven by the explosion in AI adoption across virtually every industry. As such, many competitors offer data labeling as a service. However, unlike most of our competitors, we have focused on medical data labeling. Medical data labeling has a unique suite of challenges compared to labeling for other industries. For example, accuracy is of the utmost importance, and the labelers often need to have significant medical expertise. Through the use of collective intelligence and performance-based incentives, whereby we only pay labelers who perform well on specific tasks, we are able to offer superior accuracy and scale vs any competitor.