Peer to Peer, from Humans to Agents

It's cloudy outside the MDRS. We haven't seen these kind of cool temperatures in the previous three Mobile Agents field experiments at the MDRS. I even felt rain when we were testing our iMAS system outside with Brent, our crew geologist. Now, back in the hab, while writing this story, I am listening to the HabCom Agent say "The generator output is unstable." It is saying this, because between the last two power and current readings there was a difference larger than normal. I ask it (or should I say him or her?), "What is status of the generator?" It comes back and says, "The generator is not providing power to the hab." I check with my fellow crewmembers, "What is going on with the generator?" I ask. "Oh, Don is turning it off and is trying to reconnect Wendy", one of my mates says. "Wendy's engine is running, but it produces no power. The speculation is that the electrical part of the generator may have a problem."

What is interesting about this short dialogue is that it is both between myself and the agent, and myself and the crew engineer. It's interesting from two perspectives. First, it is amazing that the software agent gave us the message, without us asking for it. Can we call our agent intelligent and autonomous? Has agent and artificial intelligence technology come this far that we can have software agents that are peers amongst us? Second, if our agent is so smart, why did I have to ask the crew engineer, clearly an intelligent carbon life-form (although some might question the intelligent part)? Why could the software agent not answer that question?

The short answer is, Mike and Ron, our Mobile Agents developers did not program the agent with the ability to know why there is no power. The long answer is more complex and detailed, too long to go into in this short story. However, I can say that it has to do with being able to "get the data." It's all about being able to get the telemetry; The data from the generator to make the appropriate inferences. That is how our agents work. But, telemetry is not enough. You need to know the history of what happened with Wendy these past two weeks, and even longer. Don, the Mars Society engineer from "The Sands" in Hanksville, has a special "relationship" with Wendy. You get the feeling that he treats Wendy as a stubborn person. Will our agents every be able to have such a relationship?

What are software agents? There are different kinds. Short, tall, male, female? No, not that kind. Software agents come in the form of objects that have independent behavior. They are the result of a new programming paradigm, like object-oriented programming, but agents are different from objects in that they can act independently, without being called. Agents can be written in an "agent-enriched" Java language. They mostly allow us to easily distribute our programs across many computers and still have them communicate with each other.

Then, there are software agents that have their roots in reactive planning, a sub-field of the artificial intelligence research field. They are called belief-desire-intention or BDI agents. These agents act based on beliefs they get. Beliefs are propositions about some state of the world. You can say it is the agent's interpretation of the world. Using these beliefs (which are first order predicate statements), the agent can make inferences, using logic and an inference engine. Based on the beliefs and their inference rules, agents can form desires and intentions. With these they can take action, like sending a message to the speaker system up stairs in the hab stating that the generator is "off-line."

This is how our agents in the Mobile Agent system work. They are a combination of Java agents and BDI agents. Just as people, our agents communicate with each other in a language. A specific theory about human languages is called speech act theory. Simply, it describes human communication in the form of speech acts. Speech acts are utterances that can be clumped together into, amongst others, a request or an inform utterance. Our agents also use speech acts to communicate with each other. This is how we can decompose the different needed functions in our system. We develop different agents that can have short ask-and-tell conversations with each other. This year we developed a Power System Monitoring agent that gets the telemetry from the One Meter system deriving information from the power grid in the hab. This agent communicates with a crewmember's personal agent, which in turn communicates with the crewmember's dialogue agent. The Dialogue agent translates the speech act into human speech, and vice versa. This is how we can talk to our personal agent.

There are a lot of agents in our current system. Every year more agents appear, which means every year the capability of our system increases. That is part of the beauty of agent technology. We can increase the capability of our system simply by adding more agents. As long as the agents speak in our speech acts protocol, they can "live amongst" us.

Next year, who knows, maybe the Wendy generator will have an agent running on it. Maybe we should call this the "Don Agent." It would be nice, although we would miss the real Don, something we should never forget. I predict that more and more people will get their own agents. Agents that will do things for them and before we know it we will interact with our agents as if they are people. However, they will never be the same. Our interaction will never be the same, as long as we know it is a software agent.

Written by Maarten Sierhuis' Personal Agent