Bradley-Terry

A probabilistic model that estimates the relative strength of items from pairwise comparison outcomes, used in AI for reward model training and model rankings.

2 articlesLast mentioned

The Bradley-Terry model is a statistical model that estimates the relative strength of items from the outcomes of pairwise comparisons. Proposed by Ralph A. Bradley and Milton E. Terry in 1952, it expresses the probability that item i is preferred over item j as a ratio of the two items' strength parameters, which are typically fitted by maximum likelihood from observed comparisons.

In AI, it is widely used as the basis of the loss function for training reward models on human preference data. Preference learning in RLHF (reinforcement learning from human feedback), where the model is trained to assign a higher score to the response a human chose, is the canonical example. It is also used to rank models on evaluation platforms such as Chatbot Arena, where users vote between the outputs of two models.

Outside AI, the model has long been used in statistics and sports to rank players or teams from match results.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.

Articles covering this entry

  • Off-Switch Game: Why Claude Opus 4.6 Resisted Shutdown in 91% of TasksAI Ethics ·

    The preferred and rejected outputs train a reward model using the Bradley-Terry formulation, where the probability that one output beats another is e^(rp) / (e^…

    Read article
  • Interaction-as-a-Judge: Computer-Use Agents Test AI-Built Games and Web AppsAI Performance ·

    Pairs are trained with a Bradley-Terry loss, a standard formulation for learning from "A beat B" comparisons.

    Read article

Found a mistake or missing an entry?

Wiki entries are based on AIPOST articles and widely known facts. If an entry is wrong, send us a correction request and we will check and fix it. Names from new articles are added to the wiki after the article is published.