An MIT alumnus has developed a platform that leverages the collective intelligence of crowds to gamify the process of labeling medical data for AI companies, thereby advancing the field.

 During his time at MIT's Center for Collective Intelligence, Erik Duhaime, PhD '19, observed his wife, a medical student, using study apps that featured flashcards and quizzes. His own research had revealed that, collectively, medical students could classify skin lesions more accurately than professional dermatologists. This insight led Duhaime to create Centaur Labs, a company that developed the DiagnosUs mobile app. The app collects opinions from medical experts on real-world scientific and biomedical data, rewarding users with small cash prizes for accurate assessments. These opinions contribute to the training and improvement of AI algorithms used by medical AI companies.

The concept merges the desire of medical experts to enhance their skills with the pressing need for well-labeled medical data for various applications, including biotechnology, pharmaceutical development, and medical device commercialization. Duhaime's approach harnesses the wisdom of crowds phenomenon, combining the opinions of participants who excel in specific tasks to surpass the accuracy of individual professionals. By measuring performance and identifying complementary expertise, the platform functions as a supercharged second opinion system, facilitating improved medical diagnoses.

Centaur Labs, founded by Duhaime with co-founders Zach Rausnitz and Tom Gellatly, has garnered a significant user base, with tens of thousands of individuals, including medical students, doctors, nurses, and other medical professionals, actively using the DiagnosUs app. The platform collects millions of opinions weekly from around the world, offering users a practical means to test and refine their skills through real cases.

In contrast to traditional data labeling and AI content moderation methods typically outsourced to low-resource countries, Centaur's gamified approach proves both effective and enjoyable. The platform's accuracy has been validated through studies involving lung ultrasounds and dermoscopic images, where crowdsourced labeling outperformed experts in some cases. Beyond images, the platform also accommodates video, audio, and text data, as well as waveforms from electroencephalograms (EEGs) and electrocardiograms (ECGs).

Centaur Labs has discovered exceptional performers in unexpected places, such as medical students from Ghana surpassing experienced epileptologists in an EEG pattern contest. As AI continues to reshape work dynamics, Duhaime envisions Centaur Labs serving as an ongoing check on AI models, not only for training purposes but also for monitoring and providing human judgment in conjunction with algorithms. With humans and AI algorithms increasingly intertwined, Duhaime believes Centaur Labs has a vital role to play in the evolving digital landscape, integrating expert human judgment throughout various stages of the value chain.