RebuttalAgent: ИИ научился «читать мысли» рецензентов (и спасать ваши статьи)

The publication process in top-tier scientific journals and ICLR-level conferences has long become an intricate psychological duel, which researchers themselves call the "dance in shackles." You spend years developing an architecture, weeks training models, and months polishing your text, only to face the verdict of anonymous Reviewer #2, who skimmed your work diagonally. The rebuttal stage—the response to criticism—becomes the moment of truth: either you convince your opponent of your correctness, or your work goes to the archive until next year.

The problem is that reviewers rarely write directly what troubles them, leaving authors to read between the lines. Researchers from Hong Kong University of Science and Technology decided it was time to change the rules of the game and presented RebuttalAgent. It's not another ChatGPT wrapper for grammar correction, but a complex system built on the principles of "Theory of Mind."

In psychology, this term describes the ability to understand that another person has their own beliefs, intentions, and knowledge that may differ from yours. RebuttalAgent attempts to reconstruct the reviewer's mental state: it analyzes the review text to understand whether the criticism stems from genuine misunderstanding, lack of data, or fundamental disagreement with the methodology.

The system's operation is built on multi-stage analysis. First, the agent deconstructs each reviewer comment, identifying hidden cognitive biases and expectations. Then it models possible dialogue trajectories: what happens if we respond aggressively, and what if we admit a mistake. This resembles a chess engine calculating game variations, but instead of pieces on a board—arguments and scientific facts. RebuttalAgent selects formulations that don't merely answer the question but "close" the psychological gestalt of criticism, making the reviewer's position more vulnerable to changing their assessment upward.

Why is this important right now? Given the exponential growth in submissions to AI conferences, the quality of peer review inevitably declines. Reviewers are overwhelmed, they make mistakes, and often miss the point.

In such an environment, it's not the smartest but the most persuasive who win. RebuttalAgent essentially democratizes access to the "art of persuasion," which is especially critical for scientists for whom English is not a native language. On the other hand, we're entering dangerous territory where scientific discourse becomes an algorithmic battle.

If an article is written by AI, reviewed (often) by AI, and now defended by AI, where in this chain is there room for actual human knowledge advancement? The industry has seen attempts to automate paper writing, but the work of Hong Kong researchers goes deeper. They've taken on the holy of holies—academic reputation and the peer review process.

If RebuttalAgent proves its effectiveness over the ICLR 2026 distance, we may face a crisis of confidence in the very institution of peer review. After all, if any criticism can be "persuaded" with the right script, such criticism is worthless. The main thing: the scientific community has obtained a legal cheat code for getting through the gatekeeping of top conferences.

Will this become a tool of justice or finally bury the objectivity of peer review?