Featured in Five is a monthly section where we pose five questions to a Computing Reviews featured reviewer. Here are the responses from our April featured reviewer, Yishai Feldman (IBM Research - Haifa).
Q: What is the most important thing that's happened in computing in the past 10 years?
A: In my opinion, that would be the rocketing success of statistical techniques such as deep learning. The growth in computing power has sparked a renaissance in the relatively dormant field of artificial neural networks, yielding many new techniques and applications. Of particular interest to me is "transfer learning," where networks trained for one task are used for a different task. This seems to be essential for reusing learning networks and combining them for more comprehensive tasks.
The success of these techniques has been so compelling that they eclipsed the more traditional knowledge-based methods in artificial intelligence. Borrowing terminology from psychology, artificial intelligence research is now focused almost exclusively on System 1 ("fast") thinking, which is the automatic, instinctive, fast, parallel, and emotional part of the human brain, at the expense of System 2 ("slow") thinking, which is deliberative, conscious, slow, and logical. System 1 allows us to jump to conclusions, and is optimized for common cases. However, this causes it to have biases that result in wrong conclusions in unusual situations. System 2, on the other hand, is slower but more accurate, and can make up for the biases of System 1. (For more details, see Daniel Kahneman's Thinking, Fast and Slow -- a must read!)
Except for the most routine situations, human intelligence continuously switches between these two modes of thinking. I believe that machine intelligence must be able to do the same in order to progress to the next level, and I see this as one of the most important challenges in computer science for the next decade.
Q: By the end of your career, where do you think computer science will have taken us? What are you working on that might contribute to that?
A: When I started my career, computers were rare, and you had to travel to their shrines in order to see them through the glass windows; only a select few were allowed access to the computer room. These days, most people carry at least one computer (far more powerful than the titans of old), and many have more than one, in various shapes and forms. With the rise of IoT, we are surrounded by more and more computers. As a result, the ways we interact with our computers are changing rapidly, becoming more like the way we interact with each other. However, advanced modes of communication, like speech and gestures, are still limited by the level of understanding our computers have of the physical word and of us, our needs, motivations, moods, and morals. State-of-the-art technology is good at extracting simple phrases from spoken and written language, but far from being able to understand the subtleties of complex uses of language. This is true for more formal types of expression, such as legal documents, as well as for colloquial speech.
There is a lot of current research that aims to extend the capabilities of computers in all these areas: better interfaces, better understanding of the world and the people in it, and deeper understanding of language. I expect this to intensify in the future, and yield significant improvements. My own current research focuses on deep understanding of more formal but complex texts, such as engineering and regulatory documents. I find this a fascinating field, which would require advances in computational linguistics and semantics as well as commonsense reasoning in order to become scalable to the point it is widely usable without a lot of manual effort in domain-specific training.
Q: Who is your favorite historical figure? Why?
A: One of the greatest Talmudic scholars of all time, community leader, physician, scientist, and philosopher, Maimonides (Rabbi Moshe ben Maimon) left a legacy that is still a source of inspiration and innovation more than 800 years after his death.
Q: If you weren't working in the computer science field, what would you be doing instead?
A: Since I started programming in FORTRAN with punched cards on a CDC 6600 when I was 12 years old, I was hooked on computer science; it is therefore quite difficult for me to imagine doing anything else. Because I like to build things, I suppose I would have gone for an engineering degree if computer science weren't an option. But since all engineering disciplines now use computer-based tools, I guess my obsession with tools would have led me to very similar work to what I am doing now, working on intelligent tools for engineering.
Q: What is your favorite type of music?
A: I like to immerse myself in music as much as I can. I turn on the radio as soon as I get up, and I often leave it on timer when I go to sleep. The music is almost always classical (roughly Bach to Bartok), with a sprinkling of jazz, world music, gospel, and pop. I like acapella singing but not opera (with a few exceptions). In my youth, I preferred symphonic pieces, but I grew up to appreciate the more restrained beauty of chamber music. For me, there is nothing more soothing than a romantic string quintet (possibly including a clarinet)!