Center for Brains, Minds and Machines Leads Progress in Artificial Intelligence

  • Feature Story

Center for Brains, Minds and Machines Leads Progress in Artificial Intelligence

by 

Sara Cody
Members of the CBMM management team include Mandana Sassanfar, Joshua Tenenbaum, Kenneth Blum (Harvard), Kathleen Sullivan, Tomaso Poggio, Matt Wilson, Boris Katz, Gabriel Kreiman, Ellen Hildreth and Nancy Kanwisher. Photo credit: Kris Brewer.

When it comes to brain teasers, understanding how the human mind produces intelligent behavior is one of the most puzzling. Especially when it comes to programming machines with human-level intelligence. Figuring out how humans see, hear, move, reason, communicate and learn is complicated. But as a society, we have a lot to gain by succeeding.

“I think the problem of intelligence is the greatest problem in science because it means understanding ourselves and understanding the tool we use to understand all other problems,” said Prof. Tomaso Poggio, Director of the Center for Brains, Minds and Machines (CBMM) and Investigator in MIBR and CSAIL. “It means that if we can make progress in intelligence then we can make ourselves a bit smarter. It means that we can build machines that help us think better so we can solve all other problems more easily. Though we’ve made progress, we are still very far from solving intelligence.”

BCS has embraced computation as a core impact area of research; in order to develop a deeper understanding of the brain, it is vital to develop a comprehensive understanding of the human mind by building links across multiple levels of analysis, from molecules to synapses to neurons to circuits to algorithms to human behavior and cognition. Moreover, neuroscience has an important role to play to move the engineering of intelligent machines forward. Taking inspiration from neuroscience has yielded major breakthroughs in artificial intelligence in the past 20 years, such as deep learning and reinforcement learning, biological phenomena that have been replicated in machines like Alpha Go Zero and Mobileye systems for autonomous driving. Since it was founded in 2013, CBMM has been central to this approach.

“The goal of CBMM is to foster collaboration between engineers, neuroscientists, computer scientist and cognitive scientists to make progress in understanding intelligence by understanding how the brain makes the mind, how the brain works and how to build intelligent machines,” said Poggio. “We believe the science of intelligence will enable better engineering of intelligence. It is a good bet that in order to reach the next breakthrough in artificial intelligence, we have to look at our own brain.”

In 2016, IBM Research announced a multiyear collaboration with BCS to advance the scientific field of machine vision, a core aspect of artificial intelligence. IBM was particularly attracted to collaborate with BCS and MIT because of the CBMM hub. The collaboration has brought together leading brain, cognitive, and computer scientists to conduct research in the field of unsupervised machine understanding of audio-visual streams of data, using insights from next-generation models of the brain to inform advances in machine vision. In 2017, MIT announced the institute-wide MIT-IBM Watson AI Lab, bringing together a multidisciplinary team from across IBM and MIT to work together advancing AI hardware, software and algorithms. BCS will continue to play a key role in this new initiative while continuing to make progress through efforts like CBMM.

“Looking forward at the next five years, we will be focusing on developing a novel architecture of intelligence beyond deep learning networks, and explore a number of very basic questions,” said Poggio. “There are very interesting mathematical questions in deep networks spanning approximation theory, optimization, and learning theory. We are already getting quite some interesting results. Since the science of intelligence is, like biology, not just one problem, but many problems requiring separate research breakthroughs, it will take time to solve it.”

To learn more about CBMM, visit cbmm.mit.edu.