, Rensselaer Polytechnic Institute
Pages: pp. 12–13
It's with a feeling of deep loss that I reflect on the recent death of Oliver Selfridge. His ideas on how humans learn, and how researchers could replicate this process with computers, profoundly affected the AI
field. In fact, if you're working in machine learning, neural networks, other soft-computing approaches, or agent-based computing, then whether you are aware or not, you owe a significant intellectual debt to Oliver. His seminal ideas in these and related areas helped define the nature of our work.
He was an organizer of the 1956 Dartmouth summer school, which is generally seen as the origin of American AI. During a career that spanned over half a century, he worked at GTE, BBN, and the MIT Media Laboratory, earning various awards and honors.
Oliver developed Pandemonium, one of the first AI programs to attack learning and reasoning. This system could recognize patterns by using what we would today call a self-organizing community of agents. Throughout the years, he continued to develop ideas about how such self-organizing distributed systems could work, and the ideas inspired a number of researchers to develop some of the main technologies we use in that area today. The AAAI, in its AI Topics page on machine learning ( www.aaai.org/AITopics/html/machine.html), chose two quotations from Oliver to begin the article.
In 2006, I invited Oliver to contribute an article to this magazine's "Future of AI" issue. Instead of resting on his laurels and writing about his early work in the field, or talking about the successes of machine learning that had arisen, Oliver chose a different course. His article "Learning and Education: A Continuing Frontier for AI" ( IEEE Intelligent Systems, May/June 2006, pp. 16–23) focused on the challenges that remain and how what we can do today seems simple when we compare it to the learning tasks that humans do almost effortlessly. His article posed several challenges to the community and sought to, once again, help shape our field's future.
To me, Oliver was not only one of the founders of our field but also a friend and mentor. Although I had met him before, we really got to know each other when I was at DARPA in the late 1990s. I invited Oliver to come down to DC to talk with me about the agent-based computing programs I was running and to give me feedback on some ideas. Oliver started by presenting a few "off the cuff" ideas about agent-based systems based on his long experience as an observer in the field. It became instantly clear that his ideas went well beyond what most people in the field, including me, were thinking, and we ended up talking late into the night. That and later discussions with Oliver significantly influenced all the agent programs I led at DARPA, including those that eventually helped lead to the Semantic Web and the use of agents in a number of real-world applications.
Oliver was one of the most interesting and energetic people I ever met, and one of the most fascinating to talk to. A dinner with Oliver would include anecdotes about early AI researchers, provocative ideas about what current researchers were missing, and insights into the field's future, as well as reflections into life, family, and the world in which we live.
Oliver cared about people and was, in the best sense of the term, a real gentleman. One night at dinner the discussion turned to our families, as discussions with Oliver invariably did (he never missed an opportunity to brag about the accomplishments of his children, including of course Mallory and Peter, both well-respected AI scientists in their own right). I mentioned in passing that my daughter, then in her early teens, was interested in mathematics, and that took us off into an anecdote about a famous mathematician whom Oliver had known as well as various other mathematical topics. About two weeks later, I received a package from Oliver—it included printouts of a couple of books about mathematics aimed at teens that he had written but not published. The books provided insights into complex mathematical phenomena at an appropriate level, and my daughter loved them. I was struck by the fact that not only had Oliver written such wonderful things but that he had also remembered the off-hand remark about my daughter's interests and acted on them. That's the kind of caring person he was.
Just a few weeks ago, I had written to some colleagues suggesting that we nominate Oliver for a major US government award, because he spent much of his career working with government projects, rather than promoting his ideas in academia. (In fact, I'm told that several of his major accomplishments were done in classified domains, so he's better known publicly for his early ideas than for his later work that helped to deliver on those early promises.) I pointed out how amazing it was that "Oliver's work impacted so many different parts of the AI field, and yet his ideas in recent writings are as fresh as those in his seminal early papers."
It's a very rare person who can, as Oliver did, provide compelling research visions throughout a long career. It's also rare to find someone as willing as he was to mentor younger researchers and spend time helping them advance their careers. And it's terribly rare to find someone as full of life and as interesting to talk to as Oliver was. Those of us practicing AI today are truly lucky that this was the field Oliver chose to work in, and he'll be sorely missed.