Agents as webs of beliefs - by Richard Ngo

Unifying beliefs, goals, and actions

Agents as webs of beliefs Unifying beliefs, goals, and actions Richard Ngo Jun 27, 2026 29 21 3 Share In this post I’ll sketch out an informal model of intelligent agents as webs of beliefs (or belief webs for short). The belief webs framework pulls together ideas from active inference, agent foundations and machine learning. In doing so it aims to unify beliefs, goals and actions as three facets of a single phenomenon. Few of these ideas are original to me, but I haven't seen anyone tie them together in a single place before (though after posting this I learned that Leverage Research arrived

related reading

• Towards a scale-free theory of intelligent agencymindthefuture.info
• Towards a scale-free theory of intelligent agency — AI Alignment Forumalignmentforum.org
• Beliefs are Chosen to Serve Goals — LessWronglesswrong.com
• The behavioral selection model for predicting AI motivations — LessWronglesswrong.com
• Embedded Agency (full-text version) — LessWronglesswrong.com
• Understanding Agency through Markov Blankets — LessWronglesswrong.com
• [1902.09469] Embedded Agencyarxiv.org
• pdfopenreview.net
• Open-minded updatelessness — LessWronglesswrong.com
• Irrationality as a Defense Mechanism for Reward-hacking — LessWronglesswrong.com
• [2206.12765] Generalized Beliefs for Cooperative AIarxiv.org
• [2608.03958] A game theory for foundation models shows new paths to rational cooperation through similarity inferencearxiv.org