MIThinker: A Plug-and-Play Policy-Optimized Thinker For Motivational Interviewing Counseling

Yizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, Ee-Peng Lim · 2026

Read MIThinker: A Plug-and-Play Policy-Optimized Thinker For Motivational Interviewing Counseling by Yizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, Ee-Peng Lim (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-29265.txt.

Description

Yang and four co-authors add a clear thinking step to AI counsellors that use motivational interviewing. This is a counselling style that helps clients find their own reasons to change. Current language-model counsellors reply straight away, without tying their inner thought to the techniques of the style, which limits how well they work.

The team's model, MIThinker, first writes out guiding thoughts, then uses them to guide strategy choice and the reply. Since no tagged thought data exists, their AugR1-MI tool works backwards, building likely counsellor thoughts from real replies. Training mixes fine-tuning on examples with reinforcement learning. In tests, MIThinker raised its theory-of-mind scores and strategy fit. Their full agent, MindfulMI, matched the counselling skill of leading systems while using about a tenth as much compute.

Catalog record

CREATORYizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, Ee-Peng Lim
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE122 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.29265

Browse the full library · About the Tower of Babel