Collective AI can amplify tiny perturbations into divergent decisions
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913003669028864 |
|---|---|
| author | Shimao, Hajime Khern-am-nuai, Warut Kim, Sung Joo |
| author_facet | Shimao, Hajime Khern-am-nuai, Warut Kim, Sung Joo |
| contents | Large language models are increasingly deployed not as single assistants but as committees whose members deliberate and then vote or synthesize a decision. Such systems are often expected to be more robust than individual models. We show that iterative multi-LLM deliberation can instead amplify tiny perturbations into divergent conversational trajectories and different final decisions. In a fully deterministic self-hosted benchmark, exact reruns are identical, yet small meaning-preserving changes to the scenario text still separate over time and often alter the final recommendation. In deployed black-box API systems, nominally identical committee runs likewise remain unstable even at temperature 0, where many users expect near-determinism. Across 12 policy scenarios, these findings indicate that instability in collective AI is not only a consequence of residual platform-side stochasticity, but can arise from sensitivity to nearby initial conditions under repeated interaction itself. Additional deployed experiments show that committee architecture modulates this instability: role structure, model composition, and feedback memory can each alter the degree of divergence. Collective AI therefore faces a stability problem, not only an accuracy problem: deterministic execution alone does not guarantee predictable or auditable deliberative outcomes. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_09127 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Collective AI can amplify tiny perturbations into divergent decisions Shimao, Hajime Khern-am-nuai, Warut Kim, Sung Joo Artificial Intelligence Multiagent Systems Large language models are increasingly deployed not as single assistants but as committees whose members deliberate and then vote or synthesize a decision. Such systems are often expected to be more robust than individual models. We show that iterative multi-LLM deliberation can instead amplify tiny perturbations into divergent conversational trajectories and different final decisions. In a fully deterministic self-hosted benchmark, exact reruns are identical, yet small meaning-preserving changes to the scenario text still separate over time and often alter the final recommendation. In deployed black-box API systems, nominally identical committee runs likewise remain unstable even at temperature 0, where many users expect near-determinism. Across 12 policy scenarios, these findings indicate that instability in collective AI is not only a consequence of residual platform-side stochasticity, but can arise from sensitivity to nearby initial conditions under repeated interaction itself. Additional deployed experiments show that committee architecture modulates this instability: role structure, model composition, and feedback memory can each alter the degree of divergence. Collective AI therefore faces a stability problem, not only an accuracy problem: deterministic execution alone does not guarantee predictable or auditable deliberative outcomes. |
| title | Collective AI can amplify tiny perturbations into divergent decisions |
| topic | Artificial Intelligence Multiagent Systems |
| url | https://arxiv.org/abs/2603.09127 |