A recurring exchange with different people at Berkeley:
Random person: Do you go into your lab often?
Me: Um, err… yeah?
Ignoring weekends, I go into the office daily, so why do I hesitate? I hesitate because I consider my work environment an office and not a lab. Answers.com defines a lab as “a laboratory.” There are more useful definitions for laboratory:
- A room or building equipped for scientific experimentation or research.
- An academic period devoted to work or study in such a place.
- A place where drugs and chemicals are manufactured.
- A place for practice, observation, or testing.
My work at Berkeley does not involve any experiments. In fact, the majority of my research there takes place on paper and then gets typed up as a report on my computer. If the first definition applies to my office, then it also applies to my home, the library, or even the bathroom. To make the second definition fit would involve further toilet humor. The third definition is also troubling because I do not take measurements. Furthermore, the idea of taking observations suggests some type of externally focused activity, which is not required to prove a theorem.
Should I care if there is no laboratory component to my work? While the motivation for many problems in information theory come from practical engineering applications, many of these have been well formulated and remain unsolved. Indeed, many information theorists make successful careers out of tackling open problems in the field and do not typically get involved with lab work directly if at all. Of couse, not all solutions to these problems find their way into practical applications, and a criticism against this approach is that applied math that is not applicable is neither math nor engineering. Feynman’s quick comeback: “What do you care what other people think?”
Accepting Feynman, the question then comes down to the type of research one wants to do. Even within one’s field, there is flexibility to choose problems that interest the researcher. At MSRI’s workshop on computational biology, Professor Reinhard Laubenbacher discussed how he changed from being a pure mathematician into one motivated by biology. Laubenbacher wanted to work on problems with practical significance, and this personal preference led to his research shift. Given that information theory is often housed within electrical engineering departments, I doubt many would complain if an information theorist had laboratory interests.
What type of lab would be useful for an information theorist so inclined? Again, this depends on the problems that interest the specific researcher, but as a case study, let’s consider the work of Claude Shannon, the field’s founder. Shannon earned his bona fides by working on Vannevar Bush’s differential analyzer, and during his career, he devised many inventions, including a mechanical juggler, mouse that solved mazes, and, among other game playing machines, a so-called mind-reading machine. The mind-reading machine inspired one of Shannon’s papers and played a role in the field of universal prediction.
Shannon’s mind-reading machine is a telling example since many implementations of it, including the one above, are available online (1 2). It just took these people some knowledge of the algorithm and Java. Indeed, computer programming provides a potential laboratory element to one’s office space that without adding extra equipment and purportedly a skill for which researchers of my generation are proficient. In fact, there was an occassion when I used MATLAB (short for “matrix laboratory”) to test if a result I wanted to prove was probably true. However, the type of programming I just mentioned was at a later stage when I had already formulated the problem I wanted to solve.
Can there be modern counterparts to Shannon’s mind-reading machine that lead to fruitful research directions? I’d like to believe this is true, but even if it’s not, it would be an opportunity to think outside the box (cubicle) and inside a lab.