AI Platform Resurrects Ancient Rome and Greece Using Scholarly Sources (www.forbes.com)

Researchers at the University of Zurich — ancient historian Felix K. Maier and computational linguist Phillip Ströbel — unveiled "Re-Experiencing History," an interactive platform that generates visually rich, historically informed scenes of ancient Rome and Greece. Rather than producing definitive reconstructions, the tool creates visual hypotheses by steering existing image generators (examples include OpenAI’s DALL·E 3 and a fine-tuned Flux Dev model) with a carefully curated corpus: roughly 300 annotated images and captions plus a retrieval-augmented database of about 70 scholarly books and articles. The system enriches prompts with specific clothing, ritual actions and architectural details to avoid generic “ancient” cliches, producing impressively plausible outputs for well-documented events like triumphal processions while drifting into conjecture for poorly attested rituals. For the AI/ML community this is a compact case study in using small, domain-specific datasets and retrieval-augmented prompt engineering to bias large generative models toward scholarly plausibility. It highlights practical gains for education, museums and research—making gaps and uncertainties in evidence visible—while surfacing persistent technical and ethical limits: model artifacts (glossy textures, missing appendages), contextual hallucinations (smartphones in Roman crowds), difficulty rendering age and labor, and broader risks of bias or misuse. The creators position the platform as a tool to provoke dialogue between evidence and imagination, not to replace scholarly judgment; access is currently limited to University of Zurich accounts pending wider release.