MIT Artificial Intelligence Laboratory
Research Abstracts — 2001


The Artificial Intelligence Laboratory has been an active entity at MIT in one form or another since at least 1959. Our goal is to understand the nature of intelligence and to engineer systems that exhibit intelligence. We are an interdisciplinary laboratory of over 200 people that spans several academic departments and has active projects ongoing with members of every academic school at MIT.

Our intellectual goal is to understand how the human mind works. We believe that vision, robotics, and language are the keys to understanding intelligence, and as such our laboratory is much more heavily biased in these directions than many other Artificial Intelligence laboratories.

Our mode of operation is to attack theoretical issues and application areas at the same time. Even for theory however, we like to build experimental systems to test out ideas.

These pages are a collection of two page abstracts of many of the research projects being carried out at the Artificial Intelligence Laboratory and at the Center for Biological and Computational Learning. The latter, CBCL, is an interdisciplinary center within MIT's Brain and Cognitive Science Department—members of CBCL are all affiliated with the AI Laboratory.

There are thirteen sections:

  • Bio Machines — A new thrust for the AI Lab is building machines based on biological principles or biological materials—this ranges from the cell level to the whole body and encompasses engineering and clinical applications.
  • Computer Architecture — Despite help from Moore's law there are still yet to be discovered innovative ways of designing better computers.
  • Genomics — Genome data-bases are flourishing, and specialized machine learning techniques along with spatial reasoning techniques can be applied to them in order to gain a richer understanding of biological systems.
  • Humanoid Robotics — Robots with human form have long been the purview of science fiction, but they also provide a window into understanding how it is that human beings operate.
  • Information Access — Part of that support is better ways to access all the information that is buried within the mass of the web, and natural language front ends help us to do just that.
  • Intelligent Working Spaces — When we combine vision and speech, and other forms of input, we are able to build intelligent spaces which provide supportive computation to our intellectual work.
  • Machine Learning — Our intelligent machines must be able to learn from the world around them, and so we study better algorithms for learning from online data-bases and from sensory experiences.
  • Medical Vision — The inside of the human body, while variable and non-rigid also is relatively constrained and so we have made rapid progress over the last five years in providing vision based tools to pratical applications in surgery and diagnosis.
  • Mobile Robotics — As mobile robots become more physically robust there are more possible applications for them, and an increasing demand for intelligence on board so that they can carry out complex missions.
  • Reliable Software — Closely coupled with the architecture of our machines is the form and reliability of the software that is run on them; we all have our own personal understanding of how this remains an area where more research can be of benefit.
  • Speech — While we are not directly working on understanding the speech signal, many of our vision techniques are proving useful in manipulating and processing speech.
  • Vision — Our richest sense is vision; while much progress has been made in the last five years there are still many simple visual tasks that are not yet conquered.
  • Vision Applied to People — In specialized domains our vision systems can do much better than in the general case, and the regularity of the human form and behavior provides a strong constraint which has enabled our vision systems to better interpret scenes with people in them.

While these areas provide a mixture of theoretical and applied research, it is worth thinking a little on the proper role of a place like the MIT Artificial Intelligence Laboratory. I believe that as in the past forty years, it should continue to pull the research directions of the rest of the world. Our strength is not in getting products to market, even for early adopters. That is best left to spin-off companies, which we regularly produce, and to the national laboratories that can take our innovations and integrate them into large working systems. Sometimes outsiders get confused and think that everything we do should have immediate application. If we had stuck to that model since our inception, many of today's technologies would not exist, as so many of them had their first expressions here at the AI Lab in barely viable exploratory projects. Likewise, as researchers, we should not let ourselves be swayed too much by the need to build applications, but should remain bold and fearless and tackle the really difficult problems—they are where there is most ultimate pay off.

Our research is financed by many patient and generous sponsors. Critically and historically, DARPA and ONR provide the vast majority of our research support. Other government sponsors include NASA and NIH. More recently, NTT has become a major sponsor of the AI Lab's work. Over the last year we have joined forces with the Laboratory for Computer Science in MIT Project Oxygen. Many of the projects in these pages are parts of Oxygen—sponsored by DARPA, Acer, Delta, Hewlett-Packard, Philips, Nokia and NTT. Other sponsors, large and small, are essential to our endeavor, both for their financial contributions and in that they provide us with a diversity of requirements and points of interaction, both of which enrich our research. The support of all our sponsors is gratefully acknowledged.

Rodney A. Brooks
Director, AI Lab
September, 2001

[Back to Top]