A logic of intentions and beliefs (Q689079)

A formal theory of intentions and beliefs (based on Discourse Representation Theory, DRT) is presented. The authors consider the paper as a step towards a unified theory of cognition, action and communication. The presented theory has advantages both over sentential theories and possible-worlds-based theories. Unlike sentential theories, this theory does not connect a belief or intention with a corresponding sentence in an agent's mind. It is just required that the content of the belief or intention approximate the content of the agent's mental state. Unlike the possible-worlds approaches, this theory avoids attributing logical omniscience to agents. In considering the structure of the agent's internal states, the approach differs from possible-worlds approaches. In considering the external anchors to the real world, it differs from the sentential approaches. A formal framework for this approach provides DRT. A formal language \({\mathbf {DRS}}\) is defined. Members of the language are sentences, called DRS's. Some of the DRS's are of the form \(x\) \textit{Believes} \(L\) or \(x\) \textit{Intends} \(L\), where \(L\) is a DRS. The language contains a temporal part inspired by \(\text{CTL}^*\). The semantics of the language is defined by a model containing possible worlds, possible times, a partial order -- temporal precedence, a set of agents, a class of functions assigning cognitive states to the agents at different worlds and times. Cognitive states are represented as sentences in \({\mathbf{DRS}}\). A standard semantics is given for ordinary logical and temporal connectives. Two functions, Content and Honesty, are crucial when defining semantics of intention and belief. Intuitively, Content of an attitude is the set of alternative scenarios (some maximal eternal branches of time points) it selects. For belief those scenarios are selected at whose initial world and time the believed DRS is true. For intention scenarios leading to its fulfillment are selected. Honesty of an attitude depends on whether the attitude matches structurally with the agents' cognitive state. This function is a binary one. The value for Honesty \((x\) \textit{Believes} \(K\), \(CS)\), where \(CS\) is a DRS called cognitive state, is the set of pairs \((\pi,L)\), where \(\pi\) is a renaming function defined on variables, \(L\) is a part of \(CS\), and \((x\) \textit{Believes} \(K)\leq^ B_ \pi L\). Relation \(\leq^ B_ \pi\) reflects some restrictions on the structure of an agent's internal states and the reasoning power they are endowed with. Honesty\((x\) \textit{Intends} \(K)\) is defined in a similar way. To put it together, the presented formalization meets the above-mentioned motivations. The external anchors to the real world are formalized by embedding functions, and they are essentially used in the definition of Content. The definition of Honesty reflects the agent's mental state and its structure (by considering \(CS\) and its approximations). A weak logic for intention and belief is described in the fourth part. It represents the minimal rationality that an agent must exhibit for it to make sense to ascribe beliefs and intentions to them. The logic is sound and complete with respect to the defined semantics. In the fifth part, further axioms are introduced. These axioms allow to model ``more intelligent'' agents or agents whose intentions are connected more tightly to their environment. For each new axiom, soundness and completeness theorem is stated (with respect to models that satisfy some specified constraints).