Ideas Made to Matter
Artificial Intelligence
AI Expert Spotlight: Dimitris Bertsimas
By
We asked several MIT experts about their latest projects and what they see as the most exciting—and concerning—aspects of the AI boom.
Dimitris Bertsimas, Vice Provost of MIT Open Learning, Associate Dean for Business Analytics, and Professor of Operations Research
has worked in artificial intelligence for more than 20 years, spanning fields from health care to climate. He's optimistic that AI will be multimodal and multitasking, but he also sees challenges ahead.
How would you describe your work in the artificial intelligence space?
One specific area I’m working on is called Holistic AI for Medicine, or HAIM. Medicine is largely multimodal: people use images, radiology reports, nurses’ notes, sometimes genomics, x-rays, and sometimes multiples of these to make a diagnosis. Humans look at a variety of inputs that capture different things. But holistic AI for medicine takes a multimodal approach. Take the prediction of whether somebody has aortic stenosis, which is a prelude to very significant heart disease and potentially a heart attack that mostly goes undetected. If you have information from all of the above, and you have built a system that uses machine learning, you can make predictions about a person’s heart.
You can also use a multimodal approach for climate, including for the prediction of hurricanes. Current models are based only on mathematical data. They’re not bad models, but they only have one type of input. We also have data regarding atmospheric conditions like air pressure, temperature, and precipitation. We also have historical data and photographs of what hurricanes look like at various points in time. Combining this data and the current models, we can improve the predication of the direction and magnitude of hurricanes by 20 percent.
A third project is surveillance for wildfires and enabling early action. I come from Greece. A year ago, the largest forest in Europe disappeared from fire, and the whole city where I was born was threatened. As the planet is warming, the phenomenon of wildfires has become more intense. What can we do about it?
We can use machine learning, artificial intelligence, and optimization to select where to put sensors for temperature and for smoke, where to base drones and how to move drones during the day to maximize the probability of the detection of fires. We have data from satellites, the geography of the environment, and data we collect online from sensors that are in the ground, as well as drones that we fly to get information to update and mobilize resources—humans, trucks, and airplanes. If you detect the fire early enough, you have a chance of controlling it and saving human lives.
An area that I’m also exploring further is using artificial intelligence for global education. I believe AI is for all fields—geology, the social sciences, medicine, law. In my new role as vice provost for Open Learning, I have been preparing an MIT-led global initiative for educating people on AI and its applications.
What is the biggest opportunity in working with AI?
AI is multimodal and multitasking. Not only can it make medical diagnoses based on a variety of inputs, but it can also simultaneously try to predict multiple different diseases. The output is not just one thing, but multiple things. A generalized exam could save the lives of many, many people because you can detect all these things at once.
What do you see as the biggest challenges or areas for caution?
AI, especially generative AI, is an oligopoly. It is in the hands of very, very few companies, and in the wrong hands, it could be a problem in the world. It’s like nuclear energy, which can meet a large portion of the energy needs of a country, but at the same time can be used to create atomic bombs. There is a need for some regulation. And I think it’s coming.
READ MORE IN OUR AI EXPERT SPOTLIGHT SERIES
