Harvard Business School and Imperial College London are using artificial intelligence to create digital replicas of professors who can interact with students beyond the limits of conventional teaching. The technology allows learners to practise business presentations, sales calls and boardroom discussions with virtual versions of real academics, but its expansion is bringing questions about intellectual property, accountability and the authority attached to a professor’s face and voice into sharper focus.
David Shrier, a professor at Imperial Business School in London, has first-hand experience of the risks involved in allowing an institution to use an academic’s identity. After he left a previous university, the institution continued using his image in videos promoting an online entrepreneurship course that he had recorded, without paying him. Shrier hired a lawyer to have the material removed from the internet. Today, however, he uses an artificial intelligence replica of himself, known as Davatar, to answer students’ questions at any hour. “What I know is that now my name, my image and my appearance belong to me,” he said.
The distinction, for Shrier, is control over how his digital identity is used. His experience illustrates the importance of consent and ownership as universities develop systems that reproduce the appearance and knowledge of their teaching staff.
Since April, Harvard Business School has offered HBS Foundry, an eight-week online programme costing $699 and aimed at founders of start-up businesses. The platform uses video avatars of seven professors, created by the US company HeyGen, to simulate professional situations in which participants can rehearse investor pitches, sales calls and meetings of a company’s board. The course is designed for entrepreneurs, including postgraduate and doctoral students, and is open to the public, although prospective participants must apply for access in writing. According to The New York Times, Harvard has tested the programme at more than 100 universities.
Katharina Rings, who directs the initiative, argues that the human appearance of the avatars helps overcome the sense of anonymity often associated with artificial intelligence. “We repeatedly hear that AI can seem anonymous, almost like a black box,” she told EL PAÍS by email. With an avatar, she said, participants know there is a real person behind the system, which encourages interaction.
Foteini Giannopoulou, an instructional designer working on the programme, sees the technology as a way to extend the practical training central to business education. “With AI, we can move beyond access to information and towards broader access to practice,” she said. Rather than simply receiving information, students can rehearse situations in which they must apply their knowledge and communicate under pressure.
Harvard does not yet have long-term evidence demonstrating that the approach produces better learning outcomes. Rings acknowledged that HBS Foundry had been launched too recently to permit a longitudinal comparison between founders who practised with avatars and those who did not. Surveys have nevertheless indicated that participants using the programme feel calmer when attending business meetings, while investors have reported that some founders appear more precise in their presentations.
Enrique Dans, a professor of innovation at IE University, cautions against interpreting such feedback as proof of educational effectiveness. The technology could deliver equivalent results at lower cost or help students who currently receive insufficient support, he said. But without independent evaluation, he added, “we only have a marketing tool with the appearance of a teacher”.
Imperial College provides another example of how digital replicas are being integrated into teaching. Shrier introduced his avatar in January 2024 for students taking his AI Ventures course. He estimates that enrolment has increased from 78 to about 200 students. The replica draws on more than half a million words from his writings and teaching materials, providing a substantial body of source material for its responses.
One central concern is whether these systems can generate statements that their human counterparts never authorised. AI tutoring tools have made mistakes before. In February 2024, The Wall Street Journal tested Khanmigo, the tutoring system developed by the educational organisation Khan Academy, and found that it struggled with elementary subtraction and endorsed an incorrect answer to a problem involving Pythagoras’s theorem with an enthusiastic “Excellent!”.
Rings says Harvard professors can review their avatars’ responses and change them at any time. Jeff Bussgang, a Harvard professor and co-founder of venture capital fund Flybridge Capital, initially monitored conversations involving his digital replica through an online portal. He no longer does so regularly. “I simply trust Katharina, her team and the algorithm,” he said.
Other academics take a more active approach. Shikhar Ghosh, another Harvard professor with a virtual replica, updates the system as his thinking changes and receives an alert whenever a user attempts to push the simulation beyond its designated subject area. One user even asked the avatar to draft a prenuptial agreement. Shrier, meanwhile, works with Star Echo, a company founded by one of his former students, to establish boundaries on what his digital counterpart can generate. “I don’t want my official university avatar to be forced into saying anything,” he said.
These safeguards have not resolved broader concerns about how universities introduce AI into academic work. In a July 2025 report based on a survey of 500 academics, the American Association of University Professors found that 71% believed the introduction of AI was overwhelmingly being led by administrators, with limited faculty participation. The report also warned of implications for intellectual property.
At Imperial College, concerns among teaching staff centre on ownership and the possibility that an AI replica could attribute inaccurate, outdated or objectionable statements to a professor, according to O’Grady, chief learning designer at the university’s Idea Lab. Dans argues that primary responsibility should rest with the university offering the system, alongside the technology provider and the professor where the latter has authorised use of their identity. The risk, he said, is that a lecturer’s face and voice lend an AI-generated response borrowed authority: “The student does not perceive an algorithmic hallucination, but a statement by a professor.”
Regulation is developing alongside the technology. The United States has no federal law specifically governing digital replicas, although states including California and Tennessee regulate aspects of their use. The proposed No Fakes Act, intended to establish a national framework, passed the Senate Judiciary Committee in June and remains under consideration. In the European Union, Article 50 of the AI Act requires companies, from 2 August 2026, to notify users when they are interacting with AI and to label content that imitates real people. Spain’s Council of Ministers approved a draft organic law on 7 July that would treat the unauthorised use of a person’s voice or image for advertising or commercial purposes through AI as illegitimate.
Bussgang remains convinced that virtual professors can be used across higher education. The format has “universal applicability”, he said. “I can’t think of any setting in which it couldn’t work.”
