The online interactive magazine of the Association for the Advancement of Artificial Intelligence

by Kenneth D. Forbus, Northwestern University

I first met Patrick when I started working at the MIT AI Lab in Fall of1973.  I was a freshman, doing a project with David Marr that was a step on the path to the Primal Sketch.  It was like a dream come true to work there. At the time, Patrick had just taken over as director, despite being an assistant professor. This was an unusual burden, but he handled it well, ensuring that it ran more smoothly while maintaining an exciting intellectual environment.

Patrick taught 6.034, Introduction to Artificial Intelligence, for many years. He was awesome, a rare combination of extremely organized and inspiring. In 1973, there were no personal computers and most students had no regular access to computers at all. This meant that programming assignments were mostly done via hand-simulation.  A defining feature of 6.034 was multiday take-home exams. These were brutal and sometimes fiendishly difficult. For example, a question on symbolic mathematics asked about a hypothetical procedure MathEQ, which would return t if the two expressions provided as arguments were equivalent, or nil otherwise. One part of the question asked us to attempt to write MathEQ.  Another part asked us to explain why MathEQ is “something that you don’t want.” Since writing MathEQ would be equivalent to solving the Halting Problem, this was of course impossible.  To be fair, they did suggest that we not try to spend too much time trying to write MathEQ.  To those of us who had yet to take an algorithms course, working on this problem was quite a learning experience.

Patrick also had a wicked sense of humor. At the end of the take-home exam, the staff provided beer for the students. In 1973 there were five questions, with boxes on chairs labeled 1 to 5 to accumulate answers to each part.  (No computers, remember, assignments were turned in on paper, commonly hand-written.)  Patrick asked me if I had any scratch paper.  He and the TAs then stapled it together, to look like a stack of answers, and put it in another box on a chair labeled 6.  Thus the hack was set.  The next student who came in confidently added his answers to the first five piles.    On seeing the extra chair, he stared at it for a moment, and then in a quavering voice, repeatedly asked “6?”, getting increasingly alarmed.  The TAs calmed him down with some beer, and the joke continued until the pile in the chair marked 6 was clearly much smaller than the rest.  This sense of humor extended well outside of class.  Patrick looked young, and sometimes passed himself off as a graduate student to unsuspecting visitors, or as his own secretary. He would sometimes commiserate with students coming in to see him, pretending that he, too, was waiting for Prof. Winston to arrive, and agreeing with their complaints about him.

Patrick was an incredible mentor, as I found out when he became my undergraduate advisor.  He was always supportive of his advisees, especially when things got tough. While Gerald Sussman was my thesis advisor, I continued to learn from Patrick in graduate  school. The lab ran on DARPA umbrella funding in those days, and Patrick was a master at writing proposals. I contributed to two of those proposals as a grad student, and what I learned from Patrick in this process has been invaluable throughout my career.

Patrick’s thesis and his subsequent work were landmarks in research on analogy.  His idea of near-misses, where a trainer uses their understanding of a learner’s concept to select examples that will be easier to learn from, is very important both for analogical learning and for active learning more generally.  He was the first to explore several algorithms for analogical matching and to use simplified English to automatically produce representations for his system to use.  Tragically, his thesis code did not survive due to the technology limitations of the day.  Disk space was scarce, so an automatic program, the Grim File Reaper, deleted files that hadn’t been touched after some period of time.  The expectation was that such files could be restored from backup magnetic tapes.  Unfortunately, those tapes were expensive as well, and so they were reused instead of archived.  Thus many programs from that era are now lost forever.

Patrick’s dedication to the AI enterprise showed through all of his thinking.  At one point he raised a racoon which had been abandoned by its mother, which caused him to reflect on the relative intelligence of people and other creatures.  A racoon, he noted, seems unlikely to be smart enough to think of making an artificial racoon.  While humans are definitely smart enough to think of building AIs, whether or not we are smart enough to do it remains an empirical question at this stage of the field.  Patrick was one of those pioneers who kept his eye on the big questions, and conveyed his enthusiasms through all his works.  He is missed.