When Professors Become Avatars: Elite Universities Test AI-Powered Teaching Beyond the Classroom
A student rehearses a startup pitch with a digital version of a Harvard professor. An online learner watches a Georgia Tech instructor who was generated from a short recording. At Imperial College Business School, students can ask faculty avatars for course guidance beyond scheduled office hours. These are early examples of a new use of generative AI in higher education: extending an instructor’s teaching presence through a synthetic voice, face, or conversational assistant.
The deployments are not all the same. Some avatars present faculty-written lessons; others answer questions using course materials; still others act as simulated investors or executives so learners can practise before meeting a person. The distinction matters: an avatar that delivers a reviewed lecture is not the same as an autonomous tutor that responds to any question. Across the examples, the most credible model is augmentation—more opportunities to practise and get timely help, with educators retaining responsibility for content and human teaching.
From recorded lecture to always-available teaching presence
Imperial Business School began piloting Davatar, a digital replica of Professor David Shrier, in January 2024. The school later tested additional faculty avatars in online and executive MBA modules. Its stated aim is to support students across time zones and make course guidance available when instructors are not online. The institution describes the work as an experiment, not a replacement for faculty, and says it is continuing to study student and staff responses. 1
The design goes beyond replaying a lecture. Imperial says its avatars are intended to guide dialogue: rather than simply supplying a yes-or-no answer, they can prompt a learner to connect a question to course concepts and consider how those concepts apply. That approach could help a student prepare for an assignment late at night or revisit an idea at their own pace. Yet the school’s positive early feedback is not, by itself, proof that students learn more; further evaluation is needed to establish effects on understanding and achievement. 1
Georgia Tech’s Foundations of Generative AI course offers a different model. Professor David Joyner created DAI-vid, a video avatar based on a short recording of himself, and uses it to present lessons from scripts he writes. Because a revised script can generate a new video without another studio session, the format makes it easier to update material in a fast-changing subject. Joyner remains responsible for the course content; the avatar handles much of the on-screen presentation. 2 3
At Harvard Business School, HBS Foundry uses seven faculty avatars in an eight-week online startup bootcamp. Participants can rehearse investor pitches, sales calls, and board meetings with simulated counterparts before a live meeting. Fortune reported that the program had served 760 founders by August 2026. The bootcamp is not an HBS degree and does not award academic credit, so it is better understood as a practice and mentoring product than as a conventional course. 4
These cases reveal three distinct jobs for an avatar: delivering prepared instruction, helping learners navigate course material, and staging realistic practice conversations. Universities should be clear about which job a particular system performs. A face that resembles a real professor can make an interaction feel personal, but appearance alone does not make its answers accurate, current, or equivalent to that professor’s judgment.
What the early evidence can—and cannot—show
A small study at Emory University’s Goizueta Business School compared learning conditions among 40 students. Students who studied human-generated course content delivered with an AI voice or avatar performed best on the final exam. The fully human-created course came second, while courses with AI-generated content performed less well. The result is suggestive, not decisive: the sample was small, and the work was described as an early experiment. It supports a practical hypothesis worth testing—that AI may help deliver expert-designed material, while the expert should remain responsible for what is taught. 6
Student responses also show why avatars are more likely to work as supplemental tools than as stand-ins for classroom relationships. At George Mason University, some students initially described AI-avatar videos as robotic and said they preferred a human instructor. When asked whether the videos should be removed, however, students wanted to keep them as extra material while still expecting their professor to teach. That distinction is central: convenience and access can be valuable without being a substitute for explanation, feedback, encouragement, and personal knowledge of a student. 7
At present, the examples offer stronger evidence of use and perceived convenience than of improved learning outcomes. Harvard’s program is new and has not yet established long-term comparative results, according to reporting by EL PAÍS. Institutions should therefore evaluate more than user satisfaction or viewing counts. Useful measures include whether learners retain concepts, transfer them to new problems, correct misunderstandings, and know when to seek a human instructor. 5
Consent, trust, and academic responsibility
A professor’s digital likeness carries professional authority. If an avatar gives a confident but incorrect answer, a learner may attribute the mistake to the professor rather than to the model. Systems need clear boundaries around what they can answer, source materials that instructors can review, and a direct route to a human when a question falls outside the course. Faculty should also be able to inspect and correct the system’s outputs as materials change.
Consent must be specific and ongoing. A written agreement should state where an avatar may appear, which materials it can use, how long the licence lasts, who can update it, whether the institution can keep using it after a professor leaves, and how the arrangement can be ended. That is more than a technical detail: the face and voice represent an individual’s identity and intellectual contribution. Reporting on Professor Shrier’s earlier experience with an institution that continued using his recorded image after he left illustrates why permission and control should not be assumed. 5
Students should be told when they are seeing or speaking with an AI avatar, what it can and cannot do, and whether their conversations are stored. UNESCO’s guidance on generative AI in education calls for a human-centred approach that protects privacy and requires ethical and pedagogical validation. That principle applies especially when the system is built to imitate a specific educator. 8
Access also deserves attention. An on-demand assistant could help learners studying across time zones or those who need to revisit a concept several times. At the same time, the institution should provide alternatives for students who cannot or do not want to use an avatar, and should test accessibility, language quality, and performance across different student groups. Personalization is useful only when it does not create a lower-quality track for students who need human contact most.
Closing Thoughts
My view is that the strongest case for professor avatars is not that they can imitate a person perfectly. It is that they can make useful learning activities available at moments when a human instructor cannot be present. Rehearsing a difficult conversation, revisiting a lesson, or asking for a guided prompt can reduce the distance between “I am stuck” and “I can try again.”
But a likeness is not a relationship, and a fluent response is not proof of expertise. Universities should treat the avatar as a faculty-governed teaching instrument: clearly disclosed, limited to a defined role, grounded in reviewed material, and supported by the real educator behind it. The technology earns a place in education only when students learn better and teachers retain authority over how their knowledge is represented.
What does this look like in practice?
A responsible deployment begins with a defined learning problem, not with the desire to make a digital copy of a professor. A faculty member might choose one bounded task: give students repeated pitch practice, explain a stable concept in short on-demand videos, or point learners back to approved readings. The instructor writes or approves the source material, tests common questions and likely edge cases, and decides which topics must be referred to a person.
The avatar should identify itself clearly, cite or point to course sources where possible, and say when it is uncertain. Student conversations should not become an invisible assessment or surveillance channel. Institutions need clear rules for data retention and access, and instructors should have a practical way to review errors and suspend the system. An initial pilot should compare learning outcomes with an appropriate human-supported alternative, not only count how many students clicked the avatar.
Most importantly, the time saved should return to teaching. If an avatar handles repetitive explanations or practice rounds, faculty can use the freed capacity for feedback, mentoring, discussion, and support for students who are struggling. That is a more defensible goal than simply serving more learners with fewer educators.
And how does this impact us?
For students, these tools could mean more flexible study and safer opportunities to practise before high-stakes conversations. They also bring a new responsibility: check important claims, understand when a response comes from AI, and ask a human when context or judgment matters.
For educators, avatars may reduce the burden of recording repetitive lessons and widen access to their expertise. They also require new work: reviewing outputs, setting boundaries, and negotiating control over one’s image, voice, and teaching materials. For universities, the real measure of success is not how lifelike the clone looks. It is whether the system improves learning without weakening trust, privacy, or the human connections that make education more than the delivery of information.
Reference videos
Digi-Dean: “AI and the Future: Transforming Ideas into Impact” — a short Emory Goizueta demonstration of a dean’s AI avatar.
“I Let an AI Avatar Teach My University Course” — a brief comparison case featuring Acadia University; it is not one of the Harvard, Imperial, or Georgia Tech deployments discussed above.
Georgia Tech’s Foundations of Generative AI on edX — course reference featuring DAI-vid as the scripted video presenter.
References
Comments