Imagine having an unlimited budget for individual tutors that offer hyperpersonalized courses that maximize the productivity and development of student skills. This summer I presented this idea with an absurd and solipsist test.
I asked an artificial intelligence tutor agent to play my role, an Oxford teacher about media and AI, and to give me a personalized master’s degree, based entirely on my work.
I configured the agent through a chatgpt tool ready to use, housed on the Nebula One -based in Azure, with the instruction to investigate and supplant my identity, and then create custom material based on what I already thought. I did not tell the extensive linguistic model (LLM) what to read or do anything to improve their abilities, such as giving access to learning materials that are not publicly available online.
The course of the agent on media and was well structured: an original six module trip, of a quarter of duration, through my own collected work, which I had never devised, but I admit that I would have liked.
He was interactive and agile, and demanded mental acuity through frequent format changes. He was intellectually challenging, as Oxford’s good tutorials should be. The agent taught rigorously, responding instantly to everything he asked. He had a deep understanding of the scene in rapid evolution of AI and the media from the same perspective as me, but had investigated more.
Apparently, this was nourished by all my multimedia production: books, speeches, articles, press interviews, even university conferences that did not know if they had been recorded, much less used to train GPT-4 or GPT-5.
The course was a great learning experience, although supposedly I already knew everything. So, in the inevitable survey for students, I gave my agent version a deserved five -star rating.
For example, in a section that addressed the ethics of non -players (PNJ) in video games, he asked:
Q: If the PNJs are generated by AI, who decides their personalities, background or moral? Could this generate biases or stereotypes?
A: If a PNJ of AI can learn and adapt, do you blur the line between character and “entity” (independent actor)?
These are important philosophical questions, which will probably come to light when Grand Theft Auto 6 comes out next May. I am excited that my agent has invented them, although my real self does not.
My agent me also based on what my real self does. In cinema, I knew the standard adobe after, which it already knew (it is used to create moving graphics and visual effects). But Nuke added, a professional tool that is used to combine and manipulate visual effects in Avengers, of which (I am ashamed to say) I had never heard.
We recommend you: the invisible AI or the hidden muscle of the great industries
Hallucinations are still a problem
So where did the knowledge that the agent had of me came from? My editorial, Routledge, signed a training data agreement with Openai, which I suppose I could cover my books on media, AI and live experience.
Unlike some authors, I am willing to do it. My books guide people through a fascinating and constantly evolving theme, and I want them to participate in the global conversation, in all possible formats and territories.
This availability should be extended to what is now potentially the easiest “language” to discover from all, the one that the AI models speak. The priority for any writer who agrees with this should be the optimization of AI: make your work easy to find, process and use for the LLM; similar to search engines optimization, but for AI.
To develop this, I tested my idea even more hiring a Deep Seek agent from China to give a course on my materials. Seeing me less visible in his training corpus, it was difficult not to offend me. There is no greater desaite in the era of AI than an outstanding llm who considers your book about ia irrelevant.
When I experienced with other AI, they had problems obtaining precise information, something very typical of 2024. Thanks to Gemini 2.5 Pro from Google, I learned amazing biographical details about me, as if I had a role at the head of the media company The Runaway Collective.
When I asked Grok, from Elon Musk, what was my best appointment, he said: “Whatever your question, the answer is AI.” It is a great phrase, but Demis Hassabis, winner of Google Deepmind’s Nobel Prize, not me.
Where we head with AI
All this self -centered summer fun was clearly absurd, although not quite. Agent self -learning projects are possibly what university teaching really needs: interactive, analytical, insicious and personalized. And there are emerging research on its value. A German study discovered that the classes taught by AI helped motivate high school students and helped them review their exams.
It will not spend much time before we begin to see this type of real -time layer formally incorporated into school and university education. Anyone who imparts classes to grade students will know that AI already exists. Students use AI transcript to take notes. The content of the classes is extracted in seconds from these transcripts and will have formed a dozen LLM in a year. To help with the drafting of trials, chatgpt, Claude, Gemini and Deep Seek/qwen are the sine qua non condition of the projects of generation Z.
But here is the trick. As the AI becomes increasingly important in education, the human professor becomes more important, no less. It will guide the learning experience, incorporating published works to the conceptual framework of a course and promoting the participation and stimulus of the face -to -face students. They can expand their value as personal tutors of AI, through agents, for each student, depending on their individual learning needs.
Where do the youngest teachers fit, who do not have a catalog of professionals to train LLM? Well, the younger the teacher is, the more likely he is an IA expert. They can use AI to develop their own conceptual vision of a course, expanding research beyond their own work and indicating the agent what should be included.
In AI, two alternative positions are usually valid simultaneously. AI is emotionally intelligent and at the same time insensitive. It is both a predictor of glorified texts and a highly creative partner. It is costing jobs, but it is also creating them. He is making us silly, but he is also driving us.
The same happens in teaching. The AI threatens the learning space, but can release a powerful interaction. A generalized belief is that it will return to the most silly students. But perhaps AI could really be opening for students a new level of customization, challenge and motivation.
*Alex Connock He is the main researcher, Said Business School, University of Oxford.
This text was originally published in The Conversation
Inspy, discover and share. Follow us and find what you are looking for on our Instagram!
