In the discussion so far presented, the distinguishing feature of embodied AVAs is characterized by . Here we will show that carrying a physical body has further implications.
Let denote an embodied AVA (Fig. 6). Then, first of all, carrying a physical body implies that (some part of) can be observed from other AVAs;
Figure 6: Model of an embodied AVA.
Thus, in equation (1) may include . To estimate from , we define the following new function:
Note that in order for to function meaningfully, must first identify the existence of and then should have the knowledge about how is reflected onto its physical body and/or the world.
While how to obtain such object identification and knowledge is an important problem to be studied, here we assume they are given; and share the knowledge about the physical reflection of . Then, can inform of its state without sending a message:
When is also an embodied AVA, the mutual exchange of and can be realized without exchanging messages. That is, the communication without message exchange can be realized between embodied AVAs. This property is the most crucial feature discriminating embodied from vacuous AVAs. Note that all software agents are vacuous and lack this property.
As mentioned before, to realize the communication without message exchange, the object identification should be established and the knowledge about how the physical state reflects the internal state should be shared between AVAs. These two factors can be considered as defining the essence of Languages. In case of talking with hands, for example, the object identification implies the recognition of a ``speaker'' and the shared knowledge the specific chirology used. We believe the language is an abstract information structure used for communication and can be represented by various different media: physical bodies, sounds, visual patterns, and symbolic messages. From an information processing point of view, the complexity and efficiency of communication may depend on the media; symbolic messages seem to be more efficient and can represent more complex information than physical body actions. If so, we should use the message exchange for communication even if all AVAs are embodied.
Figure 7: Multi-channel communication between embodied AVAs.
When an embodied AVA has the message exchange capability, it has two different ways of communication: action-perception and message exchange (Fig. 7). Two pieces of information carried by an action and a message should be consistent since a sender AVA is a rational agent. This means that the former constrains the latter and vice versa. In other words, constrains and simulates and/or . From equation (5), for example, can encode into such parts of that represent 's physical body state and/or its ``future'' action plan. Then, using defined in equation (6), the receiver changes its internal state to form ``quasi-percept'' of and/or ``expectation'' about 's body state. The action by the sender , on the other hand, is constrained so that its physical body action may be consistent with the quasi-percept of and/or the expectation formed in . This means that in some cases, the message exchange plays the same role as the action-perception. This phenomenon has been know as speech act in linguistics[3].
In summary, the communication between embodied AVAs is best characterized by multi-channel communication links formed by versatile combinations of perception, action and message exchange processes (Fig. 7).