Semantical Considerations on Dialectical and Practical Commitments

This paper studies commitments in multiagent systems. A dialectical commitment corresponds to an agent taking a position about a putative fact, including for the sake of argument. A practical commitment corresponds to an agent being obliged to another to bring about a condition. Although commitments have been used in many works, an adequate formal semantics and axiomatization for them does not yet exist. This paper presents a logic of commitments that illustrates the commonalities and differences of the two kinds of commitments. In this manner, it generalizes the developments of previous papers, precisely delineates the meanings of commitments, and identifies important postulates used informally or semiformally in previous work. This paper considers “social” commitments as introduced in (Singh 1991): by one agent to another, not of an agent to itself. Commitments help formalize a variety of interactive, loosely contractual, settings especially including argumentation and business protocols. Despite several formalizations that use commitments, there is surprisingly little work treating them as an abstraction in their own right. With few exceptions (reviewed in the last section), existing work has generally not emphasized the model-theoretic semantics of commitments as such, concentrating on ways of reasoning with or using them. It was a sensible research strategy to first establish that commitments were a useful concept. However, now that the case for commitments has been made well, further progress is hampered by the lack of a clear modeltheoretic semantics. For example, tools for designing correct protocols or verifying the interoperability or compliance of agents would rely upon a precise notion of what it means for an agent to be committed, which unfortunately is lacking. Analyses of commitments range in complexity from obligations to extensive conglomerates of social expectations and obligations. Following (Singh 1999), this paper takes a middle ground, erring perhaps toward simplicity. A commitment here is somewhat like a directed obligation, but one that arises in a context, and which can be manipulated in standardized ways. This paper doesn’t discuss context and manipulation, but its semantics provides a basis for specifying them precisely. Richer notions are readily accommodated in this approach. For example, Castelfranchi (1995) Copyright c © 2008, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. requires that a commitment be explicitly accepted by its creditor. This is reasonable in some applications but not others. We can easily define additional concepts that combine multiple instances of the more basic kinds of commitments studied here to achieve the various intuitive requirements of Castelfranchi and other researchers. Dialectical and Practical Commitments Commitments fall into two main varieties. Dialectical or dialogical commitments (Norman et al. 2004) reflect positions taken in dialogue or argumentation. By contrast, practical commitments reflect promises made during negotiation or trade. Thus dialectical commitments are about what holds and practical commitments about what is to be done. For example, a stock quote may be a dialectical commitment about the price; or a practical commitment to sell at the specified price. The two commitments may go together but not necessarily so. This paper lays the groundwork for formulating any such constraints as needed for different applications. Some key patterns of reasoning arise in dealing with commitments. Examples of natural reasoning patterns during modeling include (Ex 1) if a pharmacy commits to delivering medicines if the customer pays and shows a prescription, then once the customer shows the prescription, the pharmacy is committed to delivering medicines if the customer pays; (Ex 2) if a merchant commits to a customer to ship goods and commits to the same customer to send warranty paperwork, then the merchant commits to the customer to ship goods and send warranty paperwork; (Ex 3) a commitment that if the light is on, the light will be on would not be meaningful. We capture these patterns below as postulates B2, B5, and B8, respectively (along with several other postulates). How commitments relate to time is important. Dialectical commitments are about claims staked now (even if about future conditions) whereas practical commitments are about actions to be performed or conditions to be brought about in the future. Thus, postulates that hold for one kind of commitment can fail for the other kind. Examples Ex 1 and Ex 2 above would hold for dialectical but fail for practical commitments unless we impose additional constraints: because conjunction means that the conditions must hold simultaneously. Clearly, the above examples illustrate practical commitments. Therefore, we identify constraints under which various postulates hold for both kinds of commitments. Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence (2008)