Elamur Research⌕
☰
home / papers / 2026 / MA²P: A Meta-Cognitive Autonomous Intelligent Agents Framework for Complex Persuasion

MA²P: A Meta-Cognitive Autonomous Intelligent Agents Framework for Complex Persuasion

Paper · 2026 · Dingyi Zhang, Ziqing Zhuang, Linhai Zhang et al.

Summary

The paper addresses complex persuasive dialogue, where the persuadee's beliefs and desires are hidden and must be inferred. LLM persuaders often detect obstacles such as cost or lack of time but still answer with generic advice. Their success also swings widely between domains. MA$^2$P is a training-free, plug-and-play framework built on LeCun's blueprint for autonomous intelligent agents and on meta-cognition (planning, monitoring, evaluation). It works in three stages:

  • Meta-level judging: a Configurator picks a meta-strategy from a structured knowledge base, using historical success counts per domain, and sets evaluation rules.
  • Task-level persuading: Perception, World Model, Persuader and Short-Term Memory agents turn inferred mental-state cues into concrete, strategy-conditioned utterances.
  • Knowledge updating: an Evaluator judges the outcome and writes successful cases back to the knowledge base.

Tested on CToMPersu with five base LLM persuaders, MA$^2$P raised persuasion success in both LLM-based and human evaluations. It also improved mental-state alignment and planning coherence and reduced cross-domain variance. This matters for counseling, education and other goal-directed dialogue systems.

Key points

  • Motivation: on CToMPersu, gpt-5-mini's success rate ranges from 88.24% in its best domain to 16.67% in its worst, a 71.57-point gap.
  • Stage 2 runs a loop of four agents:
    • Perception extracts beliefs, desires and latent concerns.
    • World Model combines the meta-strategy with memory to choose a turn-specific strategy.
    • Persuader Agent turns that strategy into an utterance.
    • Short-Term Memory keeps the dialogue history and intermediate states.
  • Meta-strategy selection is an argmax over domain-matched candidates, scored by success counts in the knowledge base's Case Layer. Each successful episode increments the relevant count.
  • Limitations:
    • Most automatic metrics rely on an LLM evaluator, and the human study is small.
    • Each new domain needs a warm-up phase to seed the knowledge base.
    • The simulated persuadee models only belief and desire, not traits such as personality.
  • The prompts are released, and the authors say they will also release the code and knowledge base. All experiments use an offline simulation with an LLM persuadee.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓King's College LondonLondon · United Kingdom42 works · 1,617 citations🎓Southeast UniversityNanjing · China28 works · 354 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

💡 Concepts 1Key ideas and methods explained on our concept pages.Artificial intelligence · 49 works

Cite

Cite as Dingyi Zhang, Ziqing Zhuang, Linhai Zhang, Ziyang Gao and Deyu Zhou. 2026. MA²P: A Meta-Cognitive Autonomous Intelligent Agents Framework for Complex Persuasion. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1550

@inproceedings{zhang2026meta,
  title = {{MA²P: A Meta-Cognitive Autonomous Intelligent Agents Framework for Complex Persuasion}},
  author = {Dingyi Zhang, Ziqing Zhuang, Linhai Zhang, Ziyang Gao and Deyu Zhou},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1550},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1550}
}

References

44 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sacha Altay, Anne-Sophie Hacquin, Coralie Chevallier, and Hugo Mercier. 2023. Information delivered by a chatbot has a positive impact on covid-19 vaccines attitudes and intentions. Journal of Experimental Psychology: Applied, 29(1):52. find
  2. Qian Chen, Changqin Yin, and Yeming Gong. 2023. Would an ai chatbot persuade you: an empirical answer from the elaboration likelihood model. Information Technology & People. find
  3. Zirui Cheng and Jiaxuan You. 2025. Towards strategic persuasion with language models. Preprint, arXiv:2509.22989. — [arXiv:2509.22989](https://arxiv.org/abs/2509.22989) find
  4. Robert B. Cialdini. 2021. Influence, New and Expanded: The Psychology of Persuasion. Harper Business. find
  5. Jacob Cohen. 1968. Weighted kappa: Nominal scale agreement with provision for scaled disagreement or partial credit. Psychological Bulletin, 70(4):213– 220. find
  6. Adrian de Wynter and Tangming Yuan. 2025. The thin line between comprehension and persuasion in llms. Preprint, arXiv:2507.01936.
  7. Zhaoyu Fan, Kaihang Pan, Mingze Zhou, Bosheng Qin, Juncheng Li, Shengyu Zhang, Wenqiao Zhang, Siliang Tang, Fei Wu, and Yueting Zhuang. 2025. Towards meta-cognitive knowledge editing for multimodal llms. Preprint, arXiv:2509.05714. — [arXiv:2509.05714](https://arxiv.org/abs/2509.05714) find
  8. Kazuaki Furumai, Roberto Legaspi, Julio Cesar Vizcarra Romero, Yudai Yamazaki, Yasutaka Nishimura, Sina Semnani, Kazushi Ikeda, Weiyan Shi, and Monica Lam. 2024. Zero-shot persuasive chatbots with LLM-generated strategies and information retrieval. In Findings of the Association for Computational Li… find
  9. Jingzhi Gong, Rafail Giavrimis, Paul Brookes, Vardan Voskanyan, Fan Wu, Mari Ashiga, Matthew Truscott, Mike Basios, Leslie Kanthan, Jie Xu, and 1 others. 2025. Tuning llm-based code optimization via meta-prompting: An industrial perspective. arXiv preprint arXiv:2508.01443. — [arXiv:2508.01443](https://arxiv.org/abs/2508.01443) find
  10. Wenhao Gu, Li Gu, Chingyee Yee Suen, and Yang Wang. 2025. Metawriter: Personalized handwritten text recognition using meta-learned prompt tuning. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 23494–23504. find
  11. Rui Ha, Chaozhuo Li, Rui Pu, and Sen Su. 2025. From "aha moments" to controllable thinking: Toward meta-cognitive reasoning in large reasoning models via decoupled reasoning and control. Preprint, arXiv:2508.04460. — [arXiv:2508.04460](https://arxiv.org/abs/2508.04460) find
  12. Zejiang He, Jingyuan Huang, Menglong Lu, Zhen Huang, Shanshan Liu, Zhiliang Tian, and Dongsheng Li. 2025. GCML: Gradient coherence guided meta-learning for cross-domain emerging topic rumor detection. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 19… find
  13. Chuhao Jin, Kening Ren, Lingzhen Kong, Xiting Wang, Ruihua Song, and Huan Chen. 2024. Persuading across diverse domains: a dataset and persuasion large language model. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1678– 1706… find
  14. Chuhao Jin, Yutao Zhu, Lingzhen Kong, Shijie Li, Xiao Zhang, Ruihua Song, Xu Chen, Huan Chen, Yuchong Sun, Yu Chen, and Jun Xu. 2023. Joint semantic and strategy matching for persuasive dialogue. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 4187–4197, Singapore. As… find
  15. Tianjie Ju, Yujia Chen, Hao Fei, Mong-Li Lee, Wynne Hsu, Pengzhou Cheng, Zongru Wu, Zhuosheng Zhang, and Gongshen Liu. 2025. On the adaptive psychological persuasion of large language models. Preprint, arXiv:2506.06800. — [arXiv:2506.06800](https://arxiv.org/abs/2506.06800) find
  16. Shirish Karande, Santhosh V, and Yash Bhatia. 2024. Persuasion games with large language models. In Proceedings of the 21st International Conference on Natural Language Processing (ICON), pages 576– 582, AU-KBC Research Centre, Chennai, India. NLP Association of India (NLPAI). find
  17. Imran Khan. 2025. From literal to liberal: A meta-prompting framework for eliciting human-aligned exception handling in large language models. Preprint, arXiv:2510.12864. — [arXiv:2510.12864](https://arxiv.org/abs/2510.12864) find
  18. Yann LeCun. 2022. A path towards autonomous machine intelligence. https://api.semanticscholar.org/CorpusID:251881108. Version 0.9.2, 2022-06-27. ↗
  19. Mingjin Li, Yu Liu, Huayi Liu, Xiang Ye, Chao Jiang, Hongguang Zhang, and Yu Ruan. 2025a. MADS: Multi-agent dialogue simulation for diverse persuasion data generation. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 399–415, Suzhou (Ch… find
  20. Wenjun Li, Dexun Li, Kuicai Dong, Cong Zhang, Hao Zhang, Weiwen Liu, Yasheng Wang, Ruiming Tang, and Yong Liu. 2025b. Adaptive tool use in large language models with meta-cognition trigger. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pap…
  21. Zhuoqun Li, Hongyu Lin, Yaojie Lu, Hao Xiang, Xianpei Han, and Le Sun. 2024a. Meta-cognitive analysis: Evaluating declarative and procedural knowledge in datasets and large language models. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and… find
  22. Zhuoqun Li, Hongyu Lin, Yaojie Lu, Hao Xiang, Xianpei Han, and Le Sun. 2024b. Meta-cognitive analysis: Evaluating declarative and procedural knowledge in datasets and large language models. Preprint, arXiv:2403.09750. find
  23. Janet Metcalfe and Arthur P. Shimamura, editors. 1994. Metacognition: Knowing about Knowing. The MIT Press, Cambridge, MA. A Bradford Book. find
  24. Yujin Potter, Shiyang Lai, Junsol Kim, James Evans, and Dawn Song. 2024. Hidden persuaders: LLMs’ political leaning and their influence on voters. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 4244– 4275, Miami, Florida, USA. Association for Computa… find
  25. Haiming Qin, Jiwei Zhang, Wei Zhang, KeZhong Lu, Mingyang Zhou, Hao Liao, and Rui Mao. 2025. R-CHAR: A metacognition-driven framework for role-playing in large language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 26984–27002, Suzhou, China… find
  26. Peixin Qin, Chen Huang, Yang Deng, Wenqiang Lei, and Tat-Seng Chua. 2024. Beyond persuasion: Towards conversational recommender system with credible explanations. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 4264–4282, Miami, Florida, USA. Association for Computati… find
  27. Alexander Rogiers, Sander Noels, Maarten Buyl, and Tijl De Bie. 2024. Persuasion with large language models: a survey. Preprint, arXiv:2411.06837. — [arXiv:2411.06837](https://arxiv.org/abs/2411.06837)
  28. Till Raphael Saenger, Musashi Hinck, Justin Grimmer, and Brandon M. Stewart. 2024. AutoPersuade: A framework for evaluating and explaining persuasive arguments. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 16325–16342, Miami, Florida, USA. Associat… find
  29. Eilam Shapira, Omer Madmon, Reut Apel, Moshe Tennenholtz, and Roi Reichart. 2025. Human choice prediction in language-based persuasion games: Simulation-based off-policy evaluation. Transactions of the Association for Computational Linguistics, 13:980–1006. find
  30. Sonali Uttam Singh and Akbar Siami Namin. 2025. The influence of persuasive techniques on large language models: A scenario-based study. Computers in Human Behavior: Artificial Humans, 6:100197. find
  31. Mengzhou Sun, Sendong Zhao, Jianyu Chen, Haochun Wang, and Bing Qin. 2025. Meta-rag: Meta-analysis-inspired evidence-re-ranking method for retrieval-augmented generation in evidence-based medicine. Preprint, arXiv:2510.24003. — [arXiv:2510.24003](https://arxiv.org/abs/2510.24003) find
  32. Ziyu Wan, Yunxiang LI, Xiaoyu Wen, Yan Song, Hanjing Wang, Linyi Yang, Mark Schmidt, Jun Wang, Weinan Zhang, Shuyue Hu, and Ying Wen. 2025. ReMA: Learning to meta-think for LLMs with multi-agent reinforcement learning. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. find
  33. Xinming Wang, Jian Xu, Bin Yu, Sheng Lian, Hongzhu Yi, Yi Chen, Yingjian Zhu, Boran Wang, Hongming Yang, Han Hu, Xu-Yao Zhang, and Cheng-Lin Liu. 2025a. Mr-align: Meta-reasoning informed factuality alignment for large reasoning models. Preprint, arXiv:2510.24794.
  34. Xuewei Wang, Weiyan Shi, Richard Kim, Yoojung Oh, Sijia Yang, Jingwen Zhang, and Zhou Yu. 2019. Persuasion for good: Towards a personalized persuasive dialogue system for social good. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5635–5649, Florenc…
  35. Yaoke Wang, Yun Zhu, XintongBao XintongBao, Wenqiao Zhang, Suyang Dai, Kehan Chen, Wenqiang Li, Gang Huang, Siliang Tang, and Yueting Zhuang. 2025b. Meta-reflection: A feedback-free reflection learning framework. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguist… find
  36. Yiming Wang, Zhuosheng Zhang, Pei Zhang, Baosong Yang, and Rui Wang. 2024. Meta-reasoning: Semantics-symbol deconstruction for large language models. In Findings of the Association for Computational Linguistics: ACL 2024, pages 622–643, Bangkok, Thailand. Association for Computational Linguistics. find
  37. Yuqing Wang and Yun Zhao. 2024. Metacognitive prompting improves understanding in large language models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 1914–1926, Mexico… find
  38. Tianhao Wu, Weizhe Yuan, Olga Golovneva, Jing Xu, Yuandong Tian, Jiantao Jiao, Jason E Weston, and Sainbayar Sukhbaatar. 2025. Meta-rewarding language models: Self-improving alignment with LLM-as-a-meta-judge. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing,…
  39. Emily Xiao, Yixiao Zeng, Ada Chen, Chin-Jou Li, Amanda Bertsch, and Graham Neubig. 2025. Prompt-MII: Meta-learning instruction induction for LLMs. In Submitted to The Fourteenth International Conference on Learning Representations. Under review. find
  40. Weimin Xiong, Yifan Song, Qingxiu Dong, Bingchan Zhao, Feifan Song, XWang, and Sujian Li. 2025. MPO: Boosting LLM agents with meta plan optimization. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 3914– 3935, Suzhou, China. Association for Computational Linguistics. find
  41. Dingyi Zhang and Deyu Zhou. 2025. Persuasion should be double-blind: A multi-domain dialogue dataset with faithfulness based on causal theory of mind. Preprint, arXiv:2502.21297. — [arXiv:2502.21297](https://arxiv.org/abs/2502.21297) find
  42. Geng Zhang, Yizhou Ying, Sihang Jiang, Jiaqing Liang, Guanglei Yue, Yifei Fu, Hailin Hu, and Yanghua Xiao. 2025. From remembering to metacognition: Do existing benchmarks accurately evaluate LLMs? In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 13440–13457, Suzhou, Ch… find
  43. Haodong Zhao, Jidong Li, Zhaomin Wu, Tianjie Ju, Zhuosheng Zhang, Bingsheng He, and Gongshen Liu. 2025. Disagreements in reasoning: How a model’s thinking process dictates persuasion in multi-agent systems. Preprint, arXiv:2509.21054. — [arXiv:2509.21054](https://arxiv.org/abs/2509.21054) find
  44. (a) GPT-4o-mini (Baseline). (b) GPT-4o-mini (MA 2 P). *Figure 5: Domain-wise persuasion success rates for GPT-4o-mini across domains.* *Figure 6: Domain-wise persuasion success rates for deepseek-v3 across domains. Each dot corresponds to one domain; dot size indicates the number of test instances (… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-ma-2-p-meta-cognitive/