Banning the use of artificial intelligence (AI) in higher education risks shutting down conversations about how to innovate in pedagogy, according to a learning expert at Google Deepmind.

Miriam Schneider, director of learning initiatives in the technology firm’s main AI research division, said that AI does not have to “change pedagogy” but instead can “reinforce it”.

As many universities grapple with how best to approach the use of AI in assessments and learning, the “confrontation” between “what should remain uniquely human” and what can be done by AI “can have a meaningful role to play”, she told ����Vlog.

“It basically gives us this opportunity for reflective pause and to be able to grapple with some of those bigger questions that we mostly just sort of assumed.

����Vlog

“I think it will accelerate conversations around what the design of school systems will look like to best embrace more holistic learning for students, opposed to just knowledge transfer,” said Schneider, who has worked across Google for over 20 years, moving to Deepmind, the company’s main AI research laboratory that funds various scholarships and funding for university research, last September.

“I think those are important conversations to have. I think the risk is we don’t end up having them if we make it all or nothing.”

����Vlog

Conversations are increasingly focusing on “what do future models of learning and education look like as the technology becomes more advanced”, and Schneider said “interestingly it has almost less and less to do with the technology itself, instead thinking about the roles between teachers and students”. 

“How do you centre humanity? Centre connections and relationships? How do you increase motivation? There are some places where technology will be great at that, and some places where it will not.”

Following criticisms that large language models (LLMs) were being misused by students to cheat in academic settings, tech companies have launched various “study modes” to counter this. Google’s version, “guided learning” mode on Gemini, is powered by its LearnLM methodology which underpins Google’s education products, including Google Classroom.

Speaking to THE at last month’s World Education Forum, Schneider said the question for edtech makers is “how you bring the outside in” to “inform” how AI is optimised for learning. 

����Vlog

But in order to do that, “it requires you to treat learning as a science and that means there are core fundamental truths around learning science that we must anchor on and build the technology around. Not the other way.”

Expecting an LLM to “be good at teaching and learning” without any prior training is like “taking any stranger off the street, putting them in front of a classroom and saying, ‘great, now you’re a teacher’”.

“It requires intentional building of LLMs to basically be able to tune it to be good at pedagogy to reinforce what we know about learning science and not assume that will happen without some kind of intervention.”

juliette.rowsell@timeshighereducation.com