AIDG: A Formal Decomposition of Information Extraction and Containment Asymmetries in Multi-Turn LLM Dialogue
Fuente:
arXiv
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917532721479680 |
|---|---|
| author | Sakhawat, Adib Sadab, Fardeen Shahriar, Rakin |
| author_facet | Sakhawat, Adib Sadab, Fardeen Shahriar, Rakin |
| contents | Multi-turn LLM evaluation is typically reported as a single win-rate scalar, conflating distinct capabilities. We introduce AIDG (Adversarial Information Deduction Game), formalizing multi-turn adversarial dialogue as a two-player partially observable stochastic game (POSG) and decomposing performance along Seeker (extraction) and Holder (containment) roles. The decomposition isolates three failure modes: cooperative-prior leakage, constraint-reasoning interference, and inefficient hypothesis-space traversal. Across 439 games over six frontier LLMs, defensive performance is tightly clustered (sigma = 1.9 ELO) while offensive performance varies substantially (sigma = 53.3 ELO); confirmation framing increases extraction odds 7.75x over uninformed deduction (p < 0.00001); and constraint violations account for 41.3% of deductive failures, uncorrelated with scale (rho = 0.0). We position the containment-over-extraction gap not as a surprising finding but as a measurable consequence of locally resolvable defensive decisions versus globally coupled offensive planning, and use the decomposition to attribute the gap per model. All design choices, including turn-decay weighting and the Bradley-Terry rating model, are derived from explicit assumptions. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_17443 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | AIDG: A Formal Decomposition of Information Extraction and Containment Asymmetries in Multi-Turn LLM Dialogue Sakhawat, Adib Sadab, Fardeen Shahriar, Rakin Computation and Language Multi-turn LLM evaluation is typically reported as a single win-rate scalar, conflating distinct capabilities. We introduce AIDG (Adversarial Information Deduction Game), formalizing multi-turn adversarial dialogue as a two-player partially observable stochastic game (POSG) and decomposing performance along Seeker (extraction) and Holder (containment) roles. The decomposition isolates three failure modes: cooperative-prior leakage, constraint-reasoning interference, and inefficient hypothesis-space traversal. Across 439 games over six frontier LLMs, defensive performance is tightly clustered (sigma = 1.9 ELO) while offensive performance varies substantially (sigma = 53.3 ELO); confirmation framing increases extraction odds 7.75x over uninformed deduction (p < 0.00001); and constraint violations account for 41.3% of deductive failures, uncorrelated with scale (rho = 0.0). We position the containment-over-extraction gap not as a surprising finding but as a measurable consequence of locally resolvable defensive decisions versus globally coupled offensive planning, and use the decomposition to attribute the gap per model. All design choices, including turn-decay weighting and the Bradley-Terry rating model, are derived from explicit assumptions. |
| title | AIDG: A Formal Decomposition of Information Extraction and Containment Asymmetries in Multi-Turn LLM Dialogue |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2602.17443 |