Skip to main content

Communication

Agents are social: they interact with other agents to cooperate, coordinate or negotiate1. Agent communication builds on speech act theory2, where saying something is doing something: a message is an action meant to change the mental state of its receiver, informing it of a fact or asking it to achieve a goal. Agent communication languages such as KQML3 name these intentions with performatives (tell, achieve, askOne, ...). Since agents are autonomous, though, no sender can change a receiver's mind directly: it is the receiver that decides what a message means to it. Agents can also communicate indirectly, through the environment, by stigmergy.

Messages in JaKtA​

JaKtA keeps the core of direct communication minimal: a message carries a sender, a receiver and a payload, which can be any object. Sending is an action, provided by the MessagingSkill like any other skill: agent.sendTo(receiver, payload) sends a message to an agent, on whichever node it lives, and agent.broadcast(payload) to every agent.

adding.goal {
    takeIf { it == "sendPing" }
} triggers {
    agent.sendTo(bob, "Ping!")
}

The receiver decides what a message means​

A received message is an external event, and the receiver's handlesMessageEvents decides what it means: new beliefs, as an informative message, new goals, as a request, or nothing at all, by returning null:

handlesMessageEvents { message ->
    when (val payload = message.payload) {
        is String -> AgentUpdate.Belief(setOf(ToldBelief(payload, message.sender)), emptySet())
        else -> null
    }
}

Here ToldBelief(text, sender) is a data class of the agent's own: beliefs need not be messages. AgentUpdate.Goal(additions, removals) would accept the message as a request instead. The resulting belief or goal events trigger plans as any other, so the agent reacts to messages as it reacts to what it perceives.

The ping-pong tutorial builds a complete example with these primitives.

KQML messaging

The Prolog incarnation adds Jason-style4 KQML performatives on top of these primitives, with [source(sender)] annotations on what agents receive: see KQML messaging.

Indirect communication​

Agents can also coordinate through the environment: they act on a shared world model through a skill and perceive the changes the others made, without ever addressing each other.

Footnotes​

  1. Michael Wooldridge, Nicholas R. Jennings. Intelligent agents: theory and practice. The Knowledge Engineering Review 10(2), 1995. ↩

  2. John R. Searle. Speech Acts: An Essay in the Philosophy of Language. Cambridge University Press, 1969. ↩

  3. Tim Finin, Richard Fritzson, Don McKay, Robin McEntire. KQML as an Agent Communication Language. CIKM 1994. ↩

  4. Rafael H. Bordini, Jomi F. Hübner, Michael Wooldridge. Programming Multi-Agent Systems in AgentSpeak using Jason. Wiley, 2007. ↩