Leslie Pack Kaelbling
Massachusetts Institute of Technology
Leslie Pack Kaelbling is professor of computer science and engineering and a research director of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology. She has a bachelor’s degree in philosophy and a Ph.D. in computer science from Stanford University. Kaelbling’s research focuses on designing situated agents, mobile robotics, reinforcement learning, and decision-theoretic planning. In 2000, she founded the Journal of Machine Learning Research (JMLR) and serves as editor-in-chief.
For a long time, we’ve been electronically sharing papers. We were very early adopters of this system. The thing that is dramatically different is how we figure out what other work is going on. CiteSeer is an online system that indexes computer science literature and finds the citations for online papers. You just type in a name and click. It’s amazing. That’s changed everybody’s life.
When it was first possible to post papers on a Web site, journal publishers started to be worried and shake their fists. In the 1990s, some universities adopted policies to prohibit posting copyrighted papers on the Web, but authors typically ignored them. Scholars want people to read their stuff. We just want people to know about our work.
I was on the editorial board of a journal called Machine Learning, the main journal in the field. The price kept going up and we’d say, “This is ridiculous” – especially because libraries couldn’t afford it. Plus, the journal had an official policy about not putting stuff on the Web. We explained it was counter productive, to no avail. Finally, I got tired and said, “Forget it, let’s publish our own journal.” Two-thirds of the Machine Learning board resigned and started the new journal in 2000. (See jmlr.org – available free.)
It’s not that big a deal. We all spend a lot of time on service to the field. When you come up for tenure you are evaluated on teaching, research and service. It’s not much more work than being the editor of someone else’s journal. In the beginning it was more, but it’s fun.
By 2004 we had the second highest impact factored journal in all of computer science. ISI computes and gives an impact factor based on the number of citations per year papers get, on average. Some journals have a factor of one or less if hardly anyone cites them. With the 2004 figures, we have a 5.9 impact. That is great – it means that we are a key journal in the field. There was little resistance. I got four leaders in the field to write for the first issue. That signaled to everybody that this is where the action is. We do publish a hard copy annually for archival reasons to have a paper copy in library. It only costs a library $200.
It exists. People ask me if I’m sorry it exists. No. Some understand it to be less strong than JMLR. But it’s important to the field to have more than one venue for publication. A few months after we left, Kluwer relented and decreased the price and gave more public access – so that was a good effect.
It used to be a journal was a journal because of its role of gate keeping, certifying quality, and its role of dissemination. Now they don’t need to be coupled. If dissemination is your goal, just put your research on a Web page. If you want it registered with a date, then you can submit it to an archive. Dissemination is trivial. It’s no longer an issue. The problem with the Web now is that there is too much of it. That’s why people enjoy reading blogs – to save time from reading everything. What I would like to do is get 10 technically strong and witty friends and we could write a commentary or opinion about what’s out there. We could comment on papers that we see in the course of going to conferences, or reading other peoples’ web sites. We wouldn’t need to publish them. But it might be that other researchers would find our opinions or the set of papers we chose to comment on interesting.
We have Open Courseware at MIT. Five years ago, there was a commitment to put all of the MIT courses online – not the lectures, but all the material such as handouts, notes, syllabi. In computer science, that’s not that much of a revolution. But we also have an arrangement in Singapore where some of my lectures have been videotaped and posted publicly. I get emails from people all over the world. When I am about to lecture on a topic I don’t know a lot about, I go and look at other Web sties and see what lecturers at other universities do. I never end up with the exact same thing as anybody else, but it helps me understand different ways of looking at the material. I think that’s something good. Another example, it’s notoriously hard for faculty members to make problems for a problem set – to make them not too hard or too easy for students. We take problems all the time from the Web. It’s a great resource.
Better searching. Now searching is mainly based on names of people and technical terms. It’s harder to search on an idea. A computer is good at counting words and phrases. Often research gets replicated in two or three different fields because the vocabulary is different. There is an interest in doing a better job of reading text. In 10 or 20 years, a computer may be able to read papers and say: “Ah, this one is kind of like that one.”
You know that quotation from Newton: “If I have seen farther, it is by standing on the shoulders of giants.” Knowing what other people have done lets you build on it and not reinvent the wheel. It used to be that six people were working on the same thing at once and didn’t know it. That can still happen, but now as soon as one person writes it up in the literature you can know about it. It’s going to decrease duplication of effort, and free more people to work on things that are truly new and exciting. That’s the biggest impact.