By Yolanda Gil
Joining the incoming PhD class at Carnegie Mellon in the late 1980s, I was lucky to have incredible opportunities for faculty advisors and mentors in AI. Jaime Carbonell was among the more junior faculty, continuing the research that he started in his PhD combining natural language, planning, and machine learning. His thesis work addressed how people with different perspectives approach a discussion topic, through reasoning about commonalities and differences, planning how to counter previous points made by the other party, and generating dialogue utterances that took the conversation in an intended direction. I remember asking him why he did not pursue a more focused topic that may have had more impact. He argued that taking an integrated view on intelligence enables us to do better research in AI. This perspective has greatly influenced my own work, and I believe is of increasing relevance for AI research today.
Jaime became my thesis advisor, and encouraged me to pursue a thesis on learning new planning knowledge by reflecting on prior knowledge and engaging in proactive experimentation. As a student, I would see Jaime travel continuously all over the world. I would often ask him what he was working on that week, and he was always happy to share his ideas and motivate new problems he was interested in. I will recount some of his work and influence in AI during the years that I was a graduate student.
He was greatly focused on machine learning at that time, making great research contributions to similarity-based learning, learning by analogy, explanation-based learning, and learning how to plan. He co-founded the field of machine learning through the first workshop in a series that continues today as the International Conference on Machine Learning, three co-edited books that introduced the breadth of the research, and as Editor-in-Chief of the prestigious Machine Learning journal. Machine learning was considered by design to be a field that included psychologists proposing models of human learning, AI engineers developing interactive knowledge acquisition tools, connectionist researchers creating novel neural networks algorithms, roboticists exploring learning from sensors and external feedback, and cognitive scientists studying scientific discovery. That inclusive perspective of the field of machine learning never changed in his view.
Concurrently, Jaime continued to pursue his interests in natural language and founded the Center for Machine Translation. He managed to attract great talent and it eventually became the Language Technologies Institute, with a pioneering PhD program in that area. One of the early projects he incubated led to the very successful Lycos web search engine, with great innovations in proximity search and scalable indexing. He had advised many countries about investing in machine translation, and always showed great pride on the tremendous improvements in the quality of both automated and human-assisted translation that were achieved over the last few decades.
Also during those years, Jaime was one of the few academics in AI deeply involved in startups and industry consulting. He would advise Fortune 100 companies on investing in AI technologies for knowledge systems and for machine learning. One of his first ventures was financial data mining, creating a startup on using AI to manage investment portfolios. Many others followed: on machine translation, on language tutoring, on cybersecurity, … All addressing a real need for AI and offering compelling technologies to customers.
He became an advisor to the NIH Human Genome project which was starting in the mid 1980s. Very early on, he saw the importance of AI and machine learning for advancing biology. Even though he had no background in the area, he started multiple collaborations with biomedical researchers. He was instrumental in the creation of the Computational Biology Department as part of the School of Computer Science. Over the years, he made important contributions in the application of AI to protein interactions, protein folding, and pathway analysis.
Undoubtably, Jaime was a polymath and a quick learner who never saw an interesting problem he did not like. He was a natural born leader, something he demonstrated many times over despite his understated presence. He was a wonderful advisor to a dozen or so PhD students at any given time. I remember that during his travels we would have group meetings with lively discussions, and while reporting back upon his return we would realize that all along he had the clearest understanding of any of us about the problem at hand.
Jaime was elected a Fellow of AAAI in 1991, and was a councilor from 1991 to 1994. He was a regular contributor to AAAI, with two papers at AAAI 2020: one in detecting “help speech” in social media to highlight support of Rohingyas refugees, and one in multi-view multi-task learning with scarce data. I always looked forward to learning about his most recent research interests, always fresh and inspiring.
The last time I saw Jaime was October 2019 in Pittsburgh, on a rainy Saturday where as usual he brought his own brand of sunshine. As always, he talked very proudly of his growing children, enthusiastically of his current and past students, excitedly about recent ideas in AI and other sciences, and energetically about the new ventures he was pursuing. He was the kind of advisor, mentor, and colleague that everyone should be lucky enough to have. He will be greatly missed by all of us whose lives he touched.