IIT-Bombay proposal on AI-based proctoring raises concern among faculty

Citing approval letter of the Institute Ethics Committee, the mail stated: “If the instructor has the recorded videos already, we can directly take it from him/her. Or if the student’s video was not captured using any video conferencing service and has instead only been locally captured by the student, she/he can be asked to fill a form and attach her/his video along with it.”

Faculties in the institute have objected to the use of this data without seeking the consent of the students. Speaking to The Indian Express on condition of anonymity, a professor said, “When the students were given a long set of instructions on how proctoring will happen, nowhere was it mentioned that they are signing up to allow their videos to be used for post-processing work. Whereas the proposal to work on this kind of an initiative was in place when that email to students was drafted. The ethics committee was told by those conducting this project that if consent is taken, the purpose of the study will be defeated. Why is purpose important than consent?”

Nearly 2,000-4,000 students of the institute have appeared for the online exams until now, said sources.

Another faculty highlighted that the Institute Ethics Committee does not include any member who specialises in data privacy. “Some of us privately raised concerns that recording wasn’t given consent. But it has been brushed aside by the institute,” the professor said.
Speaking to The Indian Express, deputy director (academic and infrastructural affairs) Professor S Sudarshan said, “It is still a research project and it is still in its initial stages. The long term focus is to give privacy to students, since once build, it will not upload entire students’ data, but only the parts where students’ actions are deemed as ‘copying’. Care will be taken to ensure that the data is protected.”

When contacted, professor Ganesh Ramakrishnan, heading the activity, said, “(Semi) automated anomaly detection using machine learning can aid proctoring and serve as a deterrent to malpractices etc. To this end, based on a request from the institute authorities, I along with a couple of my students had volunteered to design a machine learning model and an associated process that can be used for anomaly detection as part of remote proctoring. The model has been currently trained by a suitable adaptation of a publicly available action recognition dataset from NTU (Nanyang Technological University).”

He added, “The code is currently open sourced and is built on a system that won us a best demo award at a conference in 2019. More details, including code, training data and modality, etc can be found at https://www.cse.iitb.ac.in/~ganesh/anomaly. We are interacting with the Institute Ethics Committee to figure out the appropriate modality by which student and faculty consent is taken prior to testing or evaluating our models on any real-world proctoring data.”