Skip to main content

A brief history of formal argumentation

Formal argumentation didn't appear fully-formed. It assembled itself over ~40 years out of several parallel strands of AI research. This is a narrative overview — no numbered steps, no tables, just context to ground the rest of the library.

1980s — default logic and non-monotonic reasoning​

The modern story begins with Reiter (1980) on default logic and McCarthy's circumscription work. The puzzle: classical logic can't handle "birds generally fly, but penguins don't." Tweety the penguin flies or doesn't depending on which default you apply last. Reiter's default rules introduced the concept of defeasible inference — conclusions that hold unless contradicted. This is the soil argumentation grew from.

1990s — abstract argumentation is founded​

Dung (1995) is the hinge. Dung abstracted away what an argument is — treating each argument as a node in a directed graph, with attack edges between them — and asked what's logically defensible. His conflict-free, admissible, preferred, stable, and grounded semantics gave the field a formal core that 30 years of subsequent work has built on.

Parallel work (Gordon, Verheij, Vreeswijk) tackled structured arguments — arguments as derivations from premises via rules. These two threads (abstract vs structured) stayed somewhat separate until ASPIC+ unified them.

2000s — structure, values, bipolarity, schemes​

Four major contributions this decade:

  • ASPIC and early structured-argumentation frameworks (Prakken's group) — formalizing how structured arguments reduce to Dung frameworks.
  • Bench-Capon (2003) — value-based argumentation, the Hal & Carla insulin example, and the idea that different audiences rationally reach different conclusions from the same framework.
  • Cayrol & Lagasquie-Schiex (2005) — adding support relations alongside attacks (bipolar frameworks). Arguments reinforce each other, not just contradict.
  • Walton, Reed, Macagno (2008) — the 60+ argument schemes catalogue. Walton had been cataloguing schemes since the early 90s; this book became the canonical reference.

2010s — weighted semantics and ASPIC+ matures​

Dunne, Hunter, McBurney, Parsons & Wooldridge (2011) introduced weighted argument systems — attacks carry real-valued weights, and a budget β tolerates a total attack weight being dropped. The continuous parameter this adds to argumentation is the same parameter we use for scene-intensity modulation.

Modgil & Prakken (2014) published the ASPIC+ tutorial — an approachable entry point to structured argumentation, complete with penguin examples and undercutting semantics.

Baroni, Caminada, Giacomin (2011) surveyed the field's semantics — complete, preferred, stable, grounded, plus ideal, eager, CF2, stage, and more. If Dung opened the question "which sets stand?", this paper shows how the field has answered it in multiple non-equivalent ways.

2020s — applications branch out​

Argumentation has moved into:

  • Legal informatics — evidence reasoning, case-based argumentation.
  • Deliberative systems — multi-agent negotiation and policy deliberation.
  • AI explainability — using argument structure as a scaffold for interpretable reasoning.
  • Scene AI — which is what this library is about.

This library is an attempt to apply that formal machinery to scene AI: β as scene intensity, schemes as action templates, acceptance as beat outcome. None of the individual pieces are new research — the work here is integration, and we owe every primitive to the papers linked throughout this site. It is one way to put this theory to use, not the only one.