AIkademiaDebate panel of the literature

Legal

About & Methodology

Last updated: 2 September 2026

AIkademia is an experiment in reading the literature through disagreement. Instead of summarising a topic, it stages a debate between positions that scholars actually hold, each represented by an AI agent bound to real, citable publications. It is built by Arcano Zero – AI Lab S.r.l., an AI venture studio in Rome, Italy.

How a debate is built

  1. Topic. You describe a question. A language model proposes four to six points of view in tension, each with search queries written for a bibliographic index.
  2. Sources. For every point of view the app queries OpenAlex, ranks the authors who recur among the most relevant works, and retrieves their titles, abstracts, venues and DOIs. Works of an author that share no vocabulary with the topic are discarded to limit the effect of merged homonyms.
  3. Panellists. The model chooses the scholar who best embodies the position and writes a profile (bio, stance, style, key arguments), constrained to cite only the retrieved works and to declare when it reasons beyond them. You can override the choice: pick another candidate, or search any researcher by name or ORCID in OpenAlex or PubMed and seat them for an existing or a new point of view.
  4. Debate. Panellists speak in rounds. You can ask questions or challenge them, addressing everyone or one person. Each contribution is generated with the panellist’s profile and works in context.
  5. Document. The transcript is turned into a paper with thesis, antithesis, synthesis and a reference list restricted to the retrieved works. PDF and Word exports include a disclaimer, a footer on every page and an appendix listing every panellist and source.

What the simulations are, and are not

Responsible use

Do not quote a panellist as if it were the scholar. Do not submit generated documents as original work where your institution forbids it. Verify every citation. If you are a scholar and want a debate that features you removed or corrected, write to [contact email].

Technology

Next.js on Vercel, Anthropic’s Claude models, the OpenAlex API, Supabase (Frankfurt) for the archive. The source code is kept in a private repository of the Operator.

← Back to AIkademia