The Belief–Desire–Intention (BDI) software model is a software architecture developed for programming intelligent agents. It is superficially characterized by the implementation of an agent's beliefs, desires, and intentions, but it actually uses these concepts to solve a particular problem in agent programming: separating the activity of selecting a plan (from a plan library or an external planner) from the execution of currently active plans. Consequently, BDI agents can balance time spent deliberating about plans (choosing what to do) and executing those plans (doing it). A third activity, creating the plans in the first place (planning), is not within the scope of the model and is left to the system designer and programmer.
The BDI model is inspired by Michael Bratman's theory of human practical reasoning, which also uses the terms belief, desire, and intention. For Bratman, desire and intention are both pro-attitudes (mental attitudes concerned with action), but commitment distinguishes intention from desire, leading to temporal persistence in plans and further plans being made on the basis of those to which one is already committed. The BDI software model partially addresses these issues: temporal persistence in the sense of explicit reference to time is not explored, but the hierarchical nature of plans is more easily implemented, as a plan consists of steps some of which may invoke other plans, implying a kind of temporal persistence.
BDI Agents
A BDI agent is a particular type of bounded rational software agent imbued with mental attitudes: beliefs, desires, and intentions. These attitudes are implemented in an architecture that includes components such as beliefs, desires, goals, intentions, plans, and events.
- Beliefs represent the informational state of the agent - its beliefs about the world, including itself and other agents. Beliefs can include inference rules, allowing forward chaining to lead to new beliefs. The term "belief" rather than "knowledge" recognizes that what an agent believes may not necessarily be true. Beliefs are stored in a database, sometimes called a belief base or belief set.
- Desires represent the motivational state of the agent - objectives or situations the agent would like to accomplish, such as finding the best price or becoming rich.
- Goals are desires that have been adopted for active pursuit, with the restriction that the set of active desires must be consistent (e.g., one should not have concurrent goals to go to a party and to stay at home).
- Intentions represent the deliberative state - what the agent has chosen to do. Intentions are desires to which the agent has committed, meaning in implemented systems that the agent has begun executing a plan.
- Plans are sequences of actions (recipes or knowledge areas) that an agent can perform to achieve one or more of its intentions. Plans may include other plans, reflecting Bratman's idea that plans are initially only partially conceived.
- Events are triggers for reactive activity, updating beliefs, triggering plans, or modifying goals. Events may be generated externally (via sensors) or internally.
The BDI model has also been extended with an obligations component, giving rise to the BOID agent architecture to incorporate obligations, norms, and commitments in social environments.
BDI Interpreter
An idealized BDI interpreter, based on SRI's PRS lineage, operates as follows:
- Initialize state.
- Repeat:
- Options: option-generator (event-queue)
- Selected options: deliberate(options)
- Update intentions(selected-options)
- Execute()
- Get new external events()
- Drop unsuccessful attitudes()
- Drop impossible attitudes()
This interpreter cycles through generating options from events, deliberating to select options, updating intentions, executing plans, and handling new events, while dropping attitudes that are unsuccessful or impossible.
Formal Logics and Reasoning
An important aspect of the BDI software model is the existence of logical models through which BDI agents can be defined and reasoned about. Research has led to axiomatizations of some BDI implementations and formal logical descriptions such as Anand Rao and Michael Georgeff's BDICTL, which combines a multiple-modal logic (with modalities for beliefs, desires, and intentions) with the temporal logic CTL*. More recently, Michael Wooldridge extended BDICTL to define LORA (Logic Of Rational Agents), incorporating an action logic to reason about individual agents and interactions in multi-agent systems.
Limitations and Criticisms
The BDI software model is one example of a reasoning architecture for a single rational agent and a concern in broader multi-agent systems. Known limitations include:
- Learning: BDI agents lack specific mechanisms to learn from past behavior and adapt to new situations.
- Three attitudes: Classical decision theorists and planning researchers question the necessity of having all three attitudes, while distributed AI research questions whether the three attitudes are sufficient.
- Logics: The multi-modal logics underlying BDI, which do not have complete axiomatizations and are not efficiently computable, have little relevance in practice.
Despite these limitations, the BDI model remains influential in agent-based systems. It has been proposed by Steven Umbrello and Roman Yampolskiy as a means of designing autonomous vehicles for human values, and it continues to inform research in Artificial intelligence and multi-agent systems. The model's separation of deliberation and execution has parallels in other areas of AI, such as Reinforcement learning and planning, and its concepts are relevant to the design of autonomous-vehicles and Robotics.
Applications and Relevance
The BDI architecture has been applied in various domains, including autonomous-vehicles, Robotics, and video-games. It provides a framework for implementing rational agents that can balance reactive and deliberative behavior. While it does not ensure all characteristics associated with intelligent agents, such as private beliefs or communication, it offers a structured approach to agent programming. The model's influence extends to multi-agent-systems and agent-based-modelling, and it remains a topic of research in Artificial intelligence and cognitive-science.
As of the early 2020s, BDI continues to be studied and extended, with efforts to integrate learning mechanisms and address computational challenges. Its formal foundations, such as BDICTL and LORA, provide a basis for verifying agent behavior, though practical implementations often simplify these logics. The model's emphasis on mental attitudes offers a human-inspired approach to designing intelligent systems, distinguishing it from purely reactive or purely deliberative architectures.