Automating chaos: the rise of the software agent

Managing chaos is tough for even the most talented managers; leaving it to computers requires an act of faith few are willing to make. But, increasingly, perhaps they should.

Managing chaos is tough for even the most talented managers; leaving it to computers requires an act of faith few are willing to make. But, increasingly, perhaps they should.

Defence contractor BAE Systems has developed a set of algorithms that help managers, individuals and machines respond effectively to dynamic situations such as battlefield actions, high-rise fires or earthquakes, where information is incomplete, often wrong or contradictory, and where wrong decisions will cost lives and money.

The Aladdin project (for Autonomous Learning Agents for Decentralised Data and Information Networks) was driven by the need to respond better to terrorist attacks such as 9/11 and 7/7, as well as natural disasters such as Hurricane Katrina and the Boxing Day tidal wave that devastated much of south-east Asia.

"There are many disaster recovery methodologies, but you cannot programme for a 'Black Swan' event," says Aladdin project director Simon Jewell. "You have to understand the limits of your model."

Coping with Black Swans

Aladdin aims specifically to allow people to cope with Black Swan events - unexpected or hard-to-predict events that have significant consequences.

Aladdin deputy director Simon Case describes the target environment as one where control is distributed, uncertainty and ambiguity are endemic, multiple self-interested stakeholders have different objectives, resources are limited and may vary over time, and timeliness of correct action is important. In short, chaos.

Case says the goal is to make an ecosystem where autonomous agents can come together and produce a preferred or at least highly desirable outcome without any one agent being in control. "It must be able to decide who will do what where and when even though no-one has the complete picture," he says.

That would comprise a tall order at the best of times. The Aladdin algorithms so far are the result of a five-year, £5.5m joint industry-academic research project. It was funded by BAE Systems and the Engineering and Physical Sciences Research Council (EPSRC). Defence systems contractors Selex Galileo and MBDA Missile Systems have also supported the project.

Around 60 research academics from the universities of Southampton, Oxford and Bristol, and Imperial College, London have worked on Aladdin. The project has already yielded more than 150 top flight academic papers, several international industrial research prizes and two awards from the British Computer Society.

Agent science

The practical goal of the project has been to develop software "agents" that act in concert with other agents under the conditions listed above. This has led to scientific advances and at least eight patents in fields such as game theory, coalition formation, sequential decision-making, efficient recursive filters to handle missing, delayed and out of sequence data, a more practical Gaussian process to infer future sensor readings and to recover from sensor failures, as well as generalised covariance union (GCU) to handle sensor data whose trustworthiness cannot be confirmed.

The algorithms have not yet been packaged into a directly usable format, and there is some debate as to whether they can be so reduced. The most concurrent active agents so far is nine. But the researchers have built several "demonstrators" that combine information from a network of sensors intelligently to produce actionable information for both humans and machines in real time.

These include an automated escape information system for burning buildings or ships; a multi-agency disaster response system; a high-resolution battlefield surveillance and management system; a naval force threat evaluation and weapons assignment system; and a military logistics system.

Nick Jennings, head of Southampton University's computer and electronics school, said that in terms of output, Aladdin had been the most successful of the many the military research projects he had worked on in the past 20 years. He ascribed this to critical mass, time enough, the right level of funding and close interworking between stakeholders.

This success has led to a £5.5m follow-on project, Orchid, which is now ramping up. Orchid, led by Nick Jennings, will explore the new science needed for "human-agent collectives" (HACs) to meet their individual and collective goals. BAE Systems is once again the lead industrial sponsor.

Smart grid will need smart agents
The UK's smart energy grid will have to rely on a network of autonomous agents that cooperate to keep homes, offices and factories supplied with gas and electrical power, believes Nick Jennings, professor of computer science at the School of Electronics and Computer Science at the University of Southampton.
The smart grid, which will connect almost 50 million meters to energy suppliers' delivery and information systems, will be too complex for centralised control, Jennings told Computer Weekly at the public launch of Aladdin, an agent-based technology research project.
"The company we are working with, Secure Meters, and other industry partners, believe the centralised model is just not going to happen," he said.
He said the system that would emerge would allow autonomy inside the house with an agent in the smart meter to control household consumption. The agent would learn how to do that by understanding household energy usage patterns. It would also incorporate information about energy generated by the house itself from solar panels, wind and other sources, and decide whether to use it, store it, trade with neighbours or sell it to the grid.
Jennings said billing required only information flows about who was using what energy and when, and for these to be properly authenticated. "This will happen," he said.

Click here to hear the full interview >>

Computer Weekly says:

The idea of an agent or golem that would do your bidding literally goes back to Adam. So although the idea is not new, getting one to work properly, or at all, is revolutionary.

The Aladdin project is conceptually similar to that of the internet, in that because there is no central controller, the system can continue to work despite unpredictable or even locally catastrophic events. But Aladdin takes the concept from the network to the application layer, because the agents must behave as if they were intelligent, communicating entities.

It is early days for Aladdin, but its follow-up, Orchid, will move agent technology closer to passing the Turing test. How satisfying it would be for UK scientists to close that circle.

Read more on IT risk management