>> AI, Deliberation & the Resolution Deficit
Tacitus starts from a simple empirical fact: modern organizations and polities can generate far more conflictual communication than any human team can structurally parse. That gap between volume of contention and capacity for resolution is the Resolution Deficit. AI is useful here not because it is “smart”, but because it can methodically map structure where humans mostly see noise.
The research programme behind Tacitus draws on three converging strands: AI for democratic deliberation, constitutional and multi‑agent debate systems, and large‑scale polarization analysis. The platform is our attempt to bring those insights into the messy, high‑stakes world of institutions, companies, and peace processes.
>> Deliberative “Habermas Machine” Experiments
Recent experiments by research labs and universities have shown that large language models can facilitate small‑group deliberation: clarifying options, re‑phrasing hostile statements, and surfacing overlapping proposals that participants can live with. These so‑called “Habermas Machine” systems do not replace judgement; they structure it. Tacitus adapts that intuition to institutional conflicts, with a focus on traceability and mandates.
>> Constitutional & Multi‑Agent Debate Systems
Work on constitutional AI and other rule‑guided agents shows how multiple models can critique, refine, and constrain each other under explicit norms. Instead of a single opaque output, you get a record of arguments, objections, and revisions. Tacitus uses a similar pattern: one agent builds the conflict ontology, another challenges it, a third searches for resolution corridors that respect legal, ethical, and political constraints.
>> Polarization & Common Ground Discovery
Parallel research on polarization uses AI to cluster viewpoints, detect framing asymmetries, and generate summaries that different camps recognise as “fair enough”. Tacitus pushes this further: instead of just describing disagreement, we search for actionable overlaps – clauses, sequences, and guarantees that can be defended in front of multiple audiences at once.
Most AI tools for “mediation” stop at sentiment analysis or generic summarization. Tacitus is built to operate where failure has real cost: boardrooms, ministries, multilateral missions, complex negotiations. Three design choices matter:
We combine vector search with a graph database populated via the Tacitus Ontology: actors, interests, constraints, red lines, guarantees, trust edges, and veto paths. The same graph powers the Deep Analysis and Ontology & Conflict Graph views, so the model reasons over a shared structure rather than free-floating text alone.
Every recommendation is conditioned on who you are and what you are allowed to do: mediator, CEO, envoy, regulator. The same graph can yield very different viable corridors depending on mandate and risk tolerance.
Every step leaves a trail: which emails or notes supported an inferred interest, which edges were added or removed, which trade‑offs are being proposed. That auditability is essential for any serious use in governance or good offices.
Tacitus sits downstream of a fast‑moving research ecosystem. A non‑exhaustive sample of work that informs our design:
- > Experiments with AI facilitators in citizen assemblies and online deliberation platforms, showing that models can help groups converge on overlapping options without erasing disagreement.
- > Constitutional and collective‑oversight approaches to AI, where multiple agents critique and refine outputs under explicit normative constraints.
- > Large‑scale analysis of social media and news ecosystems to detect polarization patterns, framing asymmetries, and potential “bridge narratives”.
- > Work on argumentation schemes, graph‑based reasoning, and computational models of negotiation, which directly inspire the Tacitus conflict ontology.
Our contribution is pragmatic: turning these ideas into operational tooling that lives where conflict actually unfolds – in email threads, meeting minutes, draft texts, and the political space around them.