Agents on the Web: Cognitive Agents

agents get smarter and our expectations of their capabilities rise correspondingly , we will no doubt treat them more anthropomorphically. It is already common for both system designers and users to attribute beliefs and intentions to agents— much the way people attribute such cognitions to their pets. For example, we might say that our cat appears aloof because it believes it is superior to its owners. We also use such terms for machines, especially when we are annoyed with them. For example, if an ATM gives us the wrong foreign exchange, we might wonder if it intends to cheat us—a possibility that would be replaced by more benign attributions if the ATM actually gave more money than we expected. In the case of agents, however, the basis for ascribing beliefs and intentions to their actions is not just sentimental or frivolous. Even though computational agents are pieces of machinery, their designs must be specified and behaviors commanded by humans. And because humans think and speak using cognitive terms such as beliefs, knowledge, desires, and intentions, it is more natural to use the same cognitive concepts when constructing agents and assigning tasks to them. Cognitive Reflections of Environmental States An agent operates in some physical or computational environment. An agent is itself a physical system of some sort. Even a pure software agent is embodied on a computer that gives a home to the agent's internal structures (data structures, if you will) and enacts its program. For an agent to act properly in a changing environment, some combination of its data structures and program must reflect the information it has about its environment. Because this information would reflect the state of the environment according to the agent, it can be termed its knowledge or a set of its beliefs. (The distinction between knowledge and beliefs is stronger in ordinary language than it is in the literature about agents, where knowledge is usually treated simply as true belief. Some researchers represent the relationship between the two using additional attributes, such as justifications, but we'll just accept the simpler definition.) Desires correspond to the state of the environment the agent prefers. Intentions correspond to the state of the environment the agent is trying to achieve, which should be a consistent subset of the agent's desires and directly connected to the agent's actions. Notice that it is the human designer who determines the agent's …