The belief-desire-intention (BDI) model is a framework in philosophy and cognitive science that explains human practical reasoning as driven by three distinct mental attitudes: beliefs, desires, and intentions. It was developed by philosopher Michael Bratman in the late 1980s as a way of accounting for future-directed intention, particularly how people plan and execute actions over time. The model lies on folk psychology, sometimes called the 'theory theory', which assumes that our mental models of the world are essentially theories that we use to predict and explain behavior.
In the BDI model, beliefs represent the information an agent has about the world, including its state and the expected outcomes of actions. Desires are the goals or preferences - what the agent wants to accomplish. Intentions are the chosen plans and commitments that drive action, serving as a bridge between deliberation and goal. Bratman emphasized that intentions are not merely stronger desires, but separate states of mind that constrain future reasoning and stabilize behavior.
Philosophical Foundations
Bratman's work, articulated in his 1987 book 'Intention, Plans, and Practical Reason', argued that intentions are central to human practical agency. He rejected the view that all action stems from desires and beliefs alone, instead proposing that intentions are volitional commitments that we adopt and maintain. These commitments help to coordinate action over time, allow for planning in advance, and enable agents to act coherently in complex environments as human life, where resources are too limited for constant optimal deliberation.
The emphasis on folk psychology, or 'theory theory', links the BDI model to a broader philosophical tradition that views everyday mental language as a kind of theory about psychological causes. This perspective was foundational to the model's adoption in artificial intelligence, where mental states can be computationally represented and used to drive automated reasoning and decision-making.
BDI Software Architecture
Bratman's insight that he was part of the inspiration for the BDI software architecture, which he helped develop. In this computational adaptation, software agents use explicit data structures for beliefs to model the world, desires or goals, and intentions as a prioritized selection of plans. A key contribution is the idea of bounded rationality: intentions limit the time spent deliberating by filtering out choices inconsistent with current commitments, thereby making problem-solving computationally feasible in complex domains. The BDI architecture became widely studied in the field of Artificial intelligence and is used in many multi-agent and robotic systems.
The architecture typically operates in a cycle: the agent updates beliefs from inputs, evaluates desires against its current environment, and selects a possible intention, then selects a plan to execute. The intention remains active until its goal is achieved, is believed to be impossible, or is intentionally abandoned, and is refined while new relevant information arises.
Applications in Psychology
Beyond AI, BDI has generated interest in psychology and cognitive development. Its structure provides a clear framework for interpreting how people form plans and commit to goals. A specific practical application is the CRIBB (Childlike Reasoning and BDI-based) model, which used BDI as the basis for a computational simulation of childlike reasoning. CRIBB aims to capture how children recall and adapt their prior beliefs, desires, and intentions when engaging in everyday planning tasks, which offers insights into both developmental cognition and the scalability of BDI-type systems.
This psychological interest highlights the model's dual value: it serves as a theory of human mental architecture as well as a basis for crafting artificial agents that can interpret, predict, and mimic intentional human behavior.
Impact and Legacy
Bratman's work influenced the philosophy of action, AI, and multi-agent system design. The phrase 'belief-desire-intention' is now a common label for a category of software architectures, and the model remains a standpoint for ongoing discussions about intention, commitment, and practical reasoning in both humans and machines. In AI, the BDI paradigm is contrastable with other approaches such as reactive or harmonic (e.g., reinforcement learning-based) agents, because of its focus on explicit symbolic representation and reasoning rather than statistical learning from data.
The model's influence extends to fields like autonomous driving, where agent architectures (Waymo) that adopt partially deliberative strategies to balance safety and progress across dynamic environments, and to eliciting contexts requiring coherent long-term action. Its principles have inspired or shaped the broader area of intelligent agents, every passed BDI remains an exemplar for bridging high-level cognition and practical control.