Who is this suitable for?
One or two master's students enrolled in the DESIGN programme with an interest in human-AI collaboration, information design, and internet culture. Technical skills to build and run a working prototype, including orchestration of AI agents through APIs, are a prerequisite.
Project summary
The proposed answer to the third question (see above) is a pipeline of AI agents, one per field of knowledge and artistic disciplines, each arguing in the voice of the most cited scholar or most renowned artist in that field or discipline. The fields come from UNESCO's International Standard Classification of Education and the disciplines from UNESCO Framework for Cultural Statistics. The pipeline runs unattended on a cloud-based virtual machine, in four phases:
- Profile research. An agent compiles a sourced profile of the scholar or artist: publications, interviews, debates, and characteristic lines of argument.
- Nomination. A second agent takes the profile as its persona brief and nominates three to five digital works, arguing in character and marking where it extrapolates beyond the sources.
- Cross-model verification. Claude, GLM, and Mistral independently check profiles and nominations. The panel spans American, Chinese, and European models, so that agreement is built on multicultural and multilingual training data.
- Review by the real scholars and artists. The people whose personas were simulated correct their own profiles and nominations, with an invitation to contribute.
What you would design
- Orchestration and prompt architecture that keeps each persona grounded in sourced material and makes its guesses visible.
- Interfaces for reading profiles, comparing nominations, and showing where the three models disagree.
- A visual account of the canon's bias in language, culture, geography, and citation networks.
What the work involves
The thesis follows a Design Science Research approach: problem framing grounded in the literature on canon formation, provenance, and AI personas; designing and building the artifact, with scope settled with the supervisor early; and evaluation with the scholars and artists whose personas the system generated.
The project takes its own weakness as part of the design problem and academic challenge. An agent speaking as a living scholar is a simulation with ethical implications, and favoring highly cited scholars or renowned artists risks skewing toward Western representatives. Your design decisions determine how visible those limits are and, ultimately, what enters the canon.
Supervision
Sune Dueholm Müller (sunedm@ifi.uio.no)