When Many Intelligent Agents are Better than One – A Conversation with Dr. Mehdi Dastani

Daniel Faggella

Daniel Faggella is Head of Research at Emerj. Called upon by the United Nations, World Bank, INTERPOL, and leading enterprises, Daniel is a globally sought-after expert on the competitive strategy implications of AI for business and government leaders.

When Many Intelligent Agents are Better than One - A Conversation with Dr. Mehdi Dastani

Episode SummaryThe beauty of a platform like eBay is that you can set a price that you’re willing to spend and let eBay do the bidding long after you’ve left the site. What if, in similar fashion, your washing machine could turn on and serve up clean clothes once it had found the cheapest rate and time of day by autonomously communicating with local electricity providers?

In this episode, we discuss multi-agent intelligent systems with Computer Scientists Dr. Mehdi Dastani, who provides a perspective on how this emerging dynamic technology is changing the landscape of how and how companies and governments operate, allowing for greater systemic change that might not be possible otherwise.

GuestDr. Mehdi Dastani

Expertise:  Computer Science and Artificial Intelligence

Recognition in BriefDr. Mehdi Dastani joined Utrecht University in 2002 after earning Masters of Science degrees in Philosophy and Computer Science, as well as a Doctorate in Humanities, from the University of Amsterdam. Dastani has co-authored multiple texts on multi-agent systems and related topics, as well as authored an extensive number of journal publications and book chapters. He is currently working on a number of research projects, including Programming Cognitive Robotics.

Current AffiliationsAssociate Professor at Utrecht University; Member of Editorial Review Board of International Journal of Agent Technologies and Systems (IJATS)

Building Intelligent Systems

Multi-agent systems are a kind of distributed intelligent system, with sets of software that embody certain properties like autonomy; the ability to perceive environment and other agents; the ability to communicate, collaborate and compete to get resources, etc., and work together to achieve a certain outcome. Dr. Mehdi Dastani sums the idea up succinctly as “a set of individual software that interacts with one another.” Mail servers on the Internet, which connect to each other and send messages, is one simple form of a multi-agent system.

The intersection of multi-agent systems and AI seems a natural fit in today’s “smart” landscape. In the beginning, explains Dastani, intelligence was a property of one kind of entity, of software that could understand human language, perceive visual inputs, etc., in order to perceive its environment, make decisions, and interact. What we consider multi-agent, he says, is a relatively recent development. Single intelligent agents that interact with others become, in a sense, more powerful.

These agents may be more intelligent as an individual entity due to its ability to request and receive information, and may also have improved functionality based on capabilities of other agents. At the same time, this interaction might result in the evolution of global intelligence, a more efficient and more knowledgeable system of entities.

Mehdi names automated financial market trading as an example, defining interactions in terms of the auction; this method of interaction between individual entities might make the market more efficient, such as yielding products at a better market price and reducing the amount of time it takes to exchange.

Another example is eBay, in which humans create their own agent and configure it by selecting the desired item for purchase, setting in place a maximum price, and then leaving the software, which becomes active and bids on a person’s behalf as the auction approaches a close.

Potential applications for other sectors, such as energy, are promising. Dastani describes a recent paper about an experiment in the manufacturing of intelligent, distributed control system in washing machines. In simple terms, when you want to make a wash, the machine will contact local energy providers to find out where and when it can buy energy for the task for the lowest price. “It’s possible that if washes in the middle of the night, it could be much cheaper…in this way, we can make more efficient balance in energy use,” says Mehdi.

There are other examples of real-world simulations in areas like transportation, economics, and government. Simulating various traffic situations, in which each individual car acts as its own single piece of software, and observing various cause and effects can help governments make more informed investment decisions in building roadways and technologies that help reduce traffic jams. Multi-agent simulations that allow us to replicate intelligent decisions to help institutions make better policy decisions benefits a much greater segment of society.

Game theory may be the best concept with which to understand multi-agent systems, states Mehdi. In the game theory branch of economics, a fundamental idea is that optimal interactions can be designed. “This doesn’t mean multi-agent systems by definition are efficient; the point is to design a mechanism that governs interactions of agents, “ says Dastani. “How can we understand and analyze actions and how can we decide which interaction is better, or more efficient, or more productive, or giving us more knowledge, or giving us more functionality with services?”

Technology: Always a Double-Edged Sword

With all of the benefits that come with this emerging technology, might there be an inherent risk in multiple intelligent agents coming together to create ever more powerful agents? Is this a similar path to what Steven Hawking and others reference when they describe a singularity point at which an intelligent system can make decisions over human beings? In one sense, we are right to wonder, says Dastani.

“We don’t know what kind of emergent property will come out of all these interactions of combined intelligence, but this is the same with everything else.”

When we design an energy grid, we make decisions and set in a place a system that governs our interaction with the environment for the rest of our foreseeable existence. At some point, we essentially don’t have direct control over all of the elements in a system, whether it be an energy grid or a food distribution system. These systems become essential to our existence, and we should be careful about considering the implications of implementation and level of control, says Mehdi.

While Dastani is not so sure that multi-agent systems will ever turn their intelligent aims on overthrowing their human creators, he does believe that we will continue to become more dependent on intelligent systems, those that work on behalf of us and make decisions for us.

“We cannot ignore these kinds of things…they are services we appreciate and try to develop more of…the question is, will these services turn against us – I’ve no reason not to think that this will happen because it assumes that this kind of general, global intelligence will create a kind of consciousness.”

In the end, Dastani sees artificial intelligence in the same light as he does other tools and systems. AI is a technology and a tool, neither inherently good or evil, and it’s our responsibility to maintain as good an understanding of the benefits and potential consequences as is humanly possible.

Image credit: Utrecht University

Stay Ahead of the AI Curve

Discover the critical AI trends and applications that separate winners from losers in the future of business.

Sign up for the 'AI Advantage' newsletter: