Back in the 1950s and ’60s, vigorous intellectual activity developed around the new engineering and physics of electronics and information processing. MIT was a major Institution contributing to this expanding whirlwind of ideas. Building 20 was at its center, the focus of such diverse fields as information theory, cybernetics, neural networks, linguistics, neuroscience, and computer science. Researchers roaming the building included Claude Shannon, Norbert Wiener, Warren McCulloch, Walter Pitts, Noam Chomsky, Jerry Lettvin, and Marvin Minsky. The intellectual ferment eventually ignited an ambitious attempt to understand intelligence and replicate it in machines.
Marvin Minsky and Seymour Papert were key in launching the field of artificial intelligence, while Noam Chomsky and David Marr pioneered cognitive science and computational neuroscience.
Over the past five decades the field of machine learning has quietly led to a host of technologies that have changed our lives, beginning with computers such as Deep Blue grandmasters of chess, and leading into Google search, speech recognition, face detection in digital cameras, cars that see, and computers (like IBM’s Watson) that win at Jeopardy.
The next several years will be a golden age for the development of similar intelligent “apps” for our phones, computers, and cars that will both help us work and entertain us. Cars equipped with Mobileye cameras (an Israeli technology) already drive themselves. Orcam is working on a system to help visually impaired people see obstacles and read. Each of these systems will be at human level performance in a narrow domain of intelligence, yet none can really be called intelligent. That is, none of them could be mistaken for a person; they would not pass the Turing test. It seems that the problem of intelligence, of how the brain generates it, and of how to make truly intelligent machines, is still wide open.
Intelligence presents one of the greatest problems in science today. I would argue that this problem is greater than solving the questions of the origin of the universe, the structure of matter, and the origin of life, because the brain is the very tool we use to understand all other great problems. Understanding intelligence requires an understanding of how the brain works, of how to build machines as intelligent as we are, and of how to develop more intelligent organizations. Even a partial solution to the problem of intelligence has great potential benefits for our society, technology, and the economy.
Imagine a world where intelligence and its emergence from brain activity is truly understood. Instead of narrow systems, such as Watson and Apple’s Siri, imagine really smart systems that could change the world. Smart systems could revolutionize the education of people with special needs and unusual learning styles, and provide compassionate care for the aged and challenged with robots that anticipate their needs. Systems that recognize how culture influences thinking could help avoid social conflict. The work of scientists and engineers could be amplified to help solve the world’s most pressing technical problems. Mental health could be understood on a deeper level to find better ways to intervene. These accomplishments will take decades. But the time has come to actively and energetically pursue research focused on the scientific foundations and practical applications of this ambitious goal.
The prospect of understanding intelligence is tantalizing and timeless. Yet, now is a uniquely appropriate time to pursue this vision at the level of basic research and of technology development.
Why now? Because today, the key fields of cognitive science, neuroscience, and computer science/artificial intelligence are re-converging. This convergence is driven by powerful new tools that allow studies of the brain and mind that inform the design of intelligent artifacts and vice versa – a positive loop from science to engineering and from engineering to science. These new tools include a dramatic increase in computing power and storage, the availability of massive datasets, the development of mathematical frameworks for learning, and the broad progress in neuroscience over the course of the past five decades.
Integrating intelligence across vision, language, planning, and other domains is possible now because of advances in understanding the cognitive, neural, and computational bases of each domain. A developmental approach to intelligence is now possible because of revolutionary work in infant cognition based on looking time, reaching, and increasingly large-scale video data collection, and physiological and neuroimaging methods. Neuroimaging, plus genetic methods of dissecting the neural circuits underlying learning and decision, now allows the study of intelligence and the brain at both large and small scales.
Recent discoveries about brain systems and innate behavioral capacities for social cognition, and the development of robotic and AI systems that for the first time interact intelligently with humans, make it clear that we must study intelligence in a social context. (I will be discussing these discoveries in great detail during my masterclass session at the upcoming Israeli Presidential Conference in Jerusalem.)
The broader impact of this research effort will be the establishment of a new field: the Science and Engineering of Intelligence. This new area of research and technology will leverage the progress in computer science, neuroscience, and cognitive science. The ability to develop more intelligent machines will influence a technology-based economy in the long term. The dream is to understand, at the computational level, how human intelligence emerges from the integration of nature and nurture — innate capacities and learning — and how we may reproduce this aspect of intelligent behavior in machines.
Success will ultimately enable us to understand ourselves better, to produce more intelligent machines, and perhaps even to make ourselves smarter.