Marvin Minsky Fellowships support promising AI research

Marie Cosindas

Marvin Minsky, a founding faculty member of the MIT Media Lab, was a renowned mathematician and computer scientist widely hailed as “the father of artificial intelligence.” So it’s fitting that after Minsky passed away in January 2016, the Marvin Minsky Foundation decided to honor his legacy by funding students whose research in AI shows great promise.

At a recent event to commemorate Minsky’s work and to announce the fellowship recipients, his daughter Margaret Minsky recalled that her father once received a fellowship that helped him navigate a crucial time in his career. “He was changing direction: from the neural-net models of the brain he had just worked on in his doctoral thesis and the construction of the world’s first neural-net machine, to grasping around for the new ideas which became symbolic AI.” 

Below are summaries of the Minsky Fellows' research:

Ben Bloomberg, Opera of the Future group

Ben’s research is concerned primarily with the orchestration of interactions between humans and machines to create new live experiences and artistic works. While much of the work in AI is concerned with learning and statistical pattern recognition, he believes that we must look beyond these methods to find the best approaches for humans and machines to come together in emotionally meaningful ways.

Marvin Minsky’s analysis of music and emotion is the basis for Ben’s work. His initial efforts have revolved around intelligent musical instruments and performance systems which use sensors to interpret the expression of performers emotionally. Ben’s Hyperproduction system uses a network of agents to process external stimuli modeled loosely on those in Minsky’s book, The Society of Mind. Ben hopes to expand the input and output capability to allow it to connect to more types of infrastructure—from lights to vehicles to HVAC systems—and he plans to expand possible processes to include new, advanced deep learning techniques. This will enable musical, emotional articulation of everyday objects and environments.

Bjarke Felbo, Scalable Cooperation group

Bjarke’s research concentrates on teaching machines to understand human emotion, at both individual and societal levels. This was Marvin Minsky's ultimate goal, as articulated in his 2006 book, The Emotion Machine. During Bjarke’s first year at the Media Lab, he has demonstrated his strong technical competencies by beating state-of-the-art deep learning algorithms across several benchmark datasets related to emotion modeling.

Going beyond the common computer science limitation of only beating benchmarks, Bjarke is also building fruitful collaborations with other research fields, including faculty from the MIT Department of Brain and Cognitive Sciences, in which he examines how machine learning can potentially help our understanding of human emotions. Another collaboration is with social and political scientists, where Bjarke identifies underlying factors affecting our society's opinions on climate change and race issues. He does this by applying his emotional machine learning techniques to analyze millions of posts from social media.

Kfir Schreiber, Molecular Machines group

The focus of Kfir’s research is in applied machine learning and AI for drug development. His premise is to develop novel AI and machine learning approaches alongside existing ideas, to transform the pharmaceuticals industry by reducing the complexity of new drug development in terms of both time and resources. Today, the time-to-market for a new drug is 15 years on average, while its research and development cost is over 2 billion dollars. Kfir hopes to harness the power of AI to reduce the time-to-market and cost significantly, allowing the development of drugs that are out of reach today. Currently, his work centers on the early pre-clinical phases. Current projects include protein folding, protein-protein interactions, and small molecule drug design.

In addition to his core research, Kfir is deeply interested in fundamental AI questions that were heavily inspired by Minsky's books and ideas. Some recent works include interaction between learning agents as seen in adversarial learning scenarios, intention inference in conflicts between mental agents, and ethics in AI.

As all three Media Lab students received their fellowships, Margaret Minsky shared some advice handed down by her father. “Attach yourself to the right people; seek great minds to learn from and make them part of yourself.” And, as she presented the fellows with gifts of hardback copies of The Society of Mind from libraries across the world, she added another quote from Marvin: “One is bound only by a simple oath to seek whatever seems the truth.”

Related Content