Understanding networked systems

By developing mathematical models, the researchers described the flow of traffic in Atlanta as a graph, noting that changes in flow are driven by people's decisions about traveling. For example, in the snowfall situation, if city officials tell workers to leave their jobs before 4 p.m. because of the snow, workers will start to leave around 2 p.m. to avoid the rush, creating traffic congestion much earlier than officials intended.

"If we figured out which people could stay on at their jobs longer and incentivize them to do that, we could avoid a huge congestion problem," Dahleh said. "However, we are still learning how to create this 'incentive architecture.' We examine how much slack the system has and how much perturbation it can take—we seek to move the system closer to stability by changing people's behavior."

Preventing failure

Dahleh focuses his research not only on transportation networks, but also on other networked infrastructures such as the electrical grid and financial markets.

Munther Dahleh, director at the MIT Institute for Data, Systems, and Society, spoke on his group's research work in network systems at the KAUST 2018 Winter Enrichment Program. Photo by Asharaf K. AbdulRahman.

 ​

"All these networks have a common theme: things move around in them," he said. "People are in the highest layer of these systems, where they interact with the networks by using them. There is a middle layer, which is people like me—we design interactions between people and the system, and we call these control systems. There is also an additional layer, which is trying to control people. You can give people rules, but you cannot tell them what to do. That is the hardest problem."

In Dahleh's middle layer, he and his team want to "get the maximum effectiveness out of the system and prevent 'bad' behavior like congestion or a cascade failure," he said.

Real-time interactions

"Ten or 15 years ago, researchers like me had our work divided into silos," Dahleh noted. "One silo thought about physical systems and analyzing the systems and the other thought about theories. The two silos didn't talk to each other because they didn't need to. Today, however, the real-time aspect of our interactions with these systems brings the two silos together. You aren't able to understand the system without understanding the interactions of people with it."

Through data, Dahleh noted, researchers can better understand human interactions with networked systems—but gathering the data can be problematic.

"Data is everywhere in the world now, and we can measure almost anything we want with it. Data is a positive enabler, but in order to understand it, we have to build models with it. Theories are needed to develop models, and then we have to realize an abstract model in real life," he said.

Munther Dahleh (second from right), director at the MIT Institute for Data, Systems, and Society, receives thanks for his keynote lecture during the KAUST Winter Enrichment Program from the Enrichment Programs team. Photo by Asharaf K. AbdulRahman.

 

"To understand interconnected networks and their fragilities—or possible failures—you still need engineers and economists, for example, but today you also need anthropologists and data scientists," Dahleh added. "Our ability to collect and aggregate data is already well beyond our ability to understand what it can tell us—and no single discipline will hold the key to solving the problem. Abstraction will be critical for the new dimension of research at the intersection of physical systems and people."

Related stories:

• Art as a window into sight
• Hackers and the Internet of Things
• The day the earth smiled
• Leaders—be the impact!
• A career with purpose