University of Michigan denies selling student data to train AI
The University of Michigan has debunked claims that it's selling student data to AI companies. To recall, yesterday, Google DeepMind employee Susan Zhang, reported getting a sponsored LinkedIn message. It offered lecture recordings, student discussions, office hours, as well as essays for a licensing fee. The university now claims it was a misunderstanding, and that students had consented to the data usage.
Here's the message received by Zhang
Details of the sales message
For $15,595 (around Rs. 13 lakh), the sales message offered 85 hours of lectures, discussions, and interviews. Meanwhile, for $12,595 (roughly Rs. 10.4 lakh), a set of 829 papers by students of the University of Michigan, across various disciplines, was available. A package for both data sets was priced at $25,000 (approximately Rs. 20.7 lakh). The message stated that the data could be useful for tuning or training large language models (LLMs) or natural language processing.
Third-party vendor involved
University spokesperson Colleen Mastony explained that a new third-party vendor sent the message with incorrect information. They have since been asked to stop their work. She added that no transactions or content sharing occurred by the vendor and that student data has never been up for sale by the university. The vendor's identity and the inaccuracies in their information were not revealed.
Catalyst Research Alliance's role
Catalyst Research Alliance reportedly offered the data for sale, claiming to partner with the University of Michigan and North Carolina State University. Mastony said that student volunteers contributed the recordings and papers in 20-year-old research studies without any personally identifiable information. These materials have been freely available for a long time to academics as a tool for writing improvement, and articulation in education.