Entry Date:
October 29, 2010

MIT Intelligence Initiative (I2@MIT)

Principal Investigator Tomaso Poggio

Project Website http://isquared.mit.edu/

The problem of intelligence –the nature of it, how the brain generates it and how it could be replicated in machines – is arguably one of the deepest and important problems in science today. Philosophers have studied intelligence for centuries, but it is only in the last several decades that key developments in a broad range of science and engineering fields have opened up a thriving "intelligence research" enterprise, making questions such as these approachable: How does the mind process sensory information to produce intelligent behavior – and how can we design intelligent computer algorithms that behave similarly? What is the structure and form of human knowledge – how is it stored, represented and organized? How do human minds arise through processes of evolution, development and learning, and what are their roots in genetics? How can we build more intelligent synthetic hardware, at the diverse scales of robots, micro- and nano-devices, or networks? How does collective intelligence arise in social and economic systems? Are there common principles of learning, prediction, decision or planning that span these diverse scales and systems?

Many of us at MIT believe that the time has come for a new, fresh attack on these problems. The launching off point will be a new integration of the fields of cognitive science, which studies the mind, neuroscience, which studies the brain, and computer science and artificial intelligence, which develop intelligent hardware and software. These fields grew up together in the 1950’s but drifted apart as each became more specialized. In the 21st century they are re-converging, driven by new tools that allow studies of the brain and mind to inform the design of intelligent artifacts and vice versa -- a virtuous loop from science to engineering and from engineering to science. These tools include radical increases in computing power, the availability of massive multimodal datasets, and the development of unifying mathematical frameworks for learning, inference and decision that support a common understanding of intelligence in minds, brains and machines. Around the core of these fields, the true scope of the enterprise driven by these new tools is highly interdisciplinary. We envision the mind-brain-computation core as one pillar of a broad initiative spanning many departments and all schools of the Institute. Other complementary pillars will include collective intelligence, with a core built on the social sciences, economics, management, media, and computer science; and physical or molecular intelligence, with a core built on biology and bioengineering, electrical engineering and computing, materials science, chemistry and physics. As a global leader in so many of these fields, MIT is uniquely positioned to lead this synthesis.

To fully exploit MIT’s unique potential we propose an Intelligence Initiative (I?). I? would draw as broadly as possible from faculty across the Institute whose work connects to the topic of intelligence – intelligence in humans or animals, in machines or molecules, in cultural or collective settings. A central goal of I? is to encourage and enable more integrative approaches than do conventional funding sources and institutional structures. For example, intelligence develops in a child from the interaction of genetic priors with learning from the environment, and integrates the growth of knowledge across vision, language, motor and social understanding, yet conventional research rarely focus on these interactions. Likewise, the study of collective and individual intelligence should be much more tightly coupled: understanding when the collective “whole” will be more or less intelligent than the sum of its parts must be grounded in knowledge of the capacities of individual minds; at the same time, the most important societal effects of individual cognition may emerge only in collective settings like financial markets, public health, political conflict, or energy and natural resource decisions. Thus, the “I?” label could also stand for a program of integrative intelligence research.

Beyond being a great intellectual mission, understanding the origins of intelligence, building more intelligent artifacts and systems, and improving mechanisms for collective decisions will be critical to the future prosperity, education, health, and security of the US and the world at large. The transformative power of internet resources like Google, Wikipedia and YouTube prompts us to ask, “How much more could we do if search engines really understood the language and intent behind a user’s questions, or the scene unfolding in a video, the way another human does?” The recent financial crisis forces us to examine where and when our intelligence failed – in the minds of individual traders, executives, or investors? In the mechanisms of our economy or our legal and regulatory system? On what timescale? – and to consider how our financial markets and aspects of our economy could be engineered in more intelligent ways. The National Academy of Engineering recently announced 14 Grand Challenges for the 21st century. Fully half of these challenges represent potential projects for I?. Four are at the core of the mind-brain-computation interface: reverse engineer the brain, advance personalized learning, enhance virtual reality, and engineer the tools of scientific discovery. Three others highlight the application of computational, collective and physical intelligence to problems of global importance: advance health informatics, engineer better medicines, secure cyberspace.

Needs and Opportunities for Funding and Strategic Cooperation.
An Intelligence Initiative could serve as the focus for several kinds of fundraising efforts, addressing different needs and opportunities. Conversations with many faculty suggest that we should target both research and education in an integrated way, with a particular focus on developing graduate students and postdoctoral fellows who can carry out the highly interdisciplinary and collaborative work we envision, hiring faculty into new cross-departmental chairs who can teach and advise these students, and providing stimulating physical spaces, meetings and events where these people can interact.

Graduate education may be the most immediate priority. It is typically graduate students who drive the most groundbreaking work at MIT, who are in the best position to devote their full efforts to collaborative research across existing labs or groups, and who have come here to be trained in the wide range of new tools, techniques and mathematics that make the I? agenda possible. We have received enthusiastic support from faculty colleagues for the idea of a new cross-departmental graduate training program in Integrative Intelligence. Speaking practically, the time and energy that individual faculty members have available for collaboration in their personal research may be limited and variable, but all recognize that graduate education must be highly collaborative. Students come to MIT because it is MIT – they come here for the opportunity to learn from and be challenged by a broad range of researchers and teachers who are the best in the world at what they do. An I? graduate program would complement rather than compete with existing departmental programs, and would help to recruit the very top students to MIT in all I?-relevant fields. Students would apply at the same time as they apply to the normal admission process of an MIT department, or in their first year here. They would take special core classes, be eligible for receiving funding from dedicated sources, and receive a special certification with their degree. A parallel postdoctoral or research fellow training program would also be extremely valuable.

Strategically, we expect fundraising should initially focus on short-term, small–scale projects. After an initial phase of 2-3 years -- and after establishing the promise of this interdisciplinary enterprise -- the time will come for a more ambitious set of longer-term projects. We have been encouraged by the strong support of the Dean of Science, as well as enthusiasm from the Dean of Engineering, the Provost and President, and heads of multiple MIT Departments and research labs. We have also received interest from government agencies such as NSF in funding integrative I? projects.

Phase 1 ($5M for 2-3 years):
(*) Fellowship support for graduate students and postdoctoral scholars. Fellowship support will be reserved for students or postdocs to engage in collaborative research with multiple laboratories, as a means to facilitate cross-departmental and cross-disciplinary interaction.

()* Seed grants for researchers starting on path-breaking research programs that more conservative federal granting agencies are not yet prepared to support. Small teams of researchers from multiple departments working on cross-disciplinary projects would be a primary target for these pilot funds.

Phase 2 ($50-60 million):
(*) 20 endowed graduate fellowships ($1 million each), supporting 8-10 PhD students per year over two to three years of their training (which typically will last five years in total).

(*) 6 endowed postdoctoral or junior research fellowships ($2-3 million each), supporting researchers just past the PhD to conduct independent work at MIT for several years in groundbreaking, cross-disciplinary areas covered by the initiative. These researchers would be charged with collaborating between existing labs, or setting up their own mini-labs with connections to multiple existing groups. -- 4 new faculty chairs ($5 million each) for researchers taking cross-disciplinary approaches to the problems of intelligence, either within the mind-brain-computation core or more broadly, at the levels of collective intelligence or molecular intelligence, or cutting across these levels. These chairs would free our hiring from the disciplinary confines implicit in existing faculty searches, instead targeting faculty in exciting areas of interest across multiple departments and schools. Emphasis would be placed on hiring faculty who could teach classes with broad I? relevance and contribute to the core curriculum of a new I? graduate training program.