Barriers to Effective Agent Communication Position Statement Michael Uschold The Boeing Company P.O. Box 3707, MS 7L-40 Seattle, WA 98124-2207 +1 425 865-3605 michael.f.uschold@boeing.com ABSTRACT 2. different ontology languages are sometimes based on In this position statement, we describe the role of ontologies in different underlying paradigms (e.g. description logic, first- agent communication. We describe a number of key barriers to order logic, frame-based, taxonomy, semantic net, thesaurus) effective agent communication. We believe that these barriers are 3. some ontology languages are very expressive, some are not. a major issue, holding back widespread uptake of agent technology. We outline a strategy for how to better understand 4. some ontology languages have a formally defined semantics, and overcome these barriers. some do not. 5. some ontology languages have inference support, some do Categories and Subject Descriptors not. ?? Even if the exact same language is used, two different people will likely build two different ontologies in the same subject area. General Terms These ontologies can be incompatible in various ways: Standardization, Languages, Theory, Design, Reliability 6. People may use different terms for the same thing. Keywords 7. People may use the same term for different things. Ontology, inter-operation, integration 8. A given notion, or concept may be modeled at different levels of detail. Position Statement When agents communicate with each other, there needs to be 9. A given notion or concept may be modeled using different some way to ensure that the meaning of what one agent 'says' is primitives in the language. E.g. is the notion of being red accurately conveyed to the other agent. There are two extremes, in modeled by having the attribute color, with value red? Or is principal, for handling this problem. The simplest (and perhaps modeled as a class called something like RedThings? Or is it the most common?) approach, it to ignore the problem altogether. both, where either 1) they are independent or 2) RedThings That is, just assume that all agents are using the same terms to is a derived class defined in terms of the attribute color and mean the same things. In practice, this will usually be an the value red. assumption built into the application. This only works, however, 10. A given notion or concept may be modeled with a very when one has full control over what agents exist, and what they different fundamental underlying primitives. For example, might communicate. In reality, agents need to interact in a much when modeling time, one might a time interval as a wider world, where it will cannot be assumed that other agents primitive, another might use a time point as a basic primitive. will use the same terms, or if they do, it cannot be assumed that There is no reason to expect these two ontologies to be the terms will mean the same thing. compatible. The moment we accept the problem, and grant that agents may not Even if the exact same language is used, and if there is use the same terms to mean the same things, we need a way for an substantial similarity in the underlying models and assumptions, agent to discover what another agent means when it the inference required to determine whether two terms actually communicates. The basic idea for how to proceed, in theory, mean the same thing, is in general, intractable. anyway, is to encode the terms and their semantics in ontologies. If agent 1 sends a message to agent 2, then along with this Many of these problems are inherent, and will never go away. The message is an indicator of, or a pointer to what ontology agent 1 challenge to the agent community, and the ontology community is is using. The theory goes, that agent 2 can look in agent 1's to discover where progress is possible, and to move forward. At ontology to see what the terms mean, the message is successfully least two main approaches exist. First, a lot of benefit can be made communicated, the service is performed to specification, and they by increasing the degree of standardization, both in the languages all live happily ever after. The holy grail is for this to be able to and in the content of the actual ontologies. Second, where happen despite the following plethora of difficulties. standardization is not possible, technologies need to be developed for mapping and translating between and among ontologies. 1. there are many different ontology languages Finally, when problems are known to be impossible, in general, the challenge is to find ways to make simplifying assumptions which enable agents to do useful things in practical situations.