Scientific Styles

November 29, 2022

Allen Newell (AI pioneer and Turing Award winner​) once shared his thoughts on the style of scientific practice. Even today his observations are helpful, particularly as we revisit our own styles and choose new research topics.

Newell’s ideas can be found in the excerpt below. This was originally taken from a retrospective talk he gave at CMU on December 4, 1991. I came across the excerpt while browsing the ML Theory channel on Amii’s Slack, after Csaba gave a general talk on the subject. Hope you enjoy!

So, I said I will reflect not just on my career, but on sort of scientific styles, which might be interesting to the group at hand, since most of you are [​sic​] either haven’t yet adopted a scientific style, or probably don’t like the one that you’re in. So, let’s talk about scientific style. Styles of scientific lives. There are a bunch of floating maxims around when we do this. The maxim associated with this slide is “To each scientific life, its own style, but each style defines a separate life”. OK. So, I want to just talk about different styles because I only exhibit one of these and there are lots of other styles.

So, the first one is the one I actually associate with algorithm complexity theorists, which I call a “nomadic existence”. Turns out that it’s easy to prove theorems and it’s very hard to prove theorems that are not almost like the theorems everybody else has proved. So, every so often in complexity theory someone proves a really new interesting theorem which opens up a new area, and then everybody picks up their tents and they all run over into this new area. And they pick up all the interesting nuggets around, and pretty soon you’ll look around and there’s nothing around except nuggets that are like everybody else’s nuggets. And finally, somebody gets a new one. Everyone picks up their tents. OK. And that’s the way you live your life. OK. You live your life moving from one theoretical area to another as it opens up.

The second kind of scientific style is sort of a general substantive theme. Gordon Bell actually has a pretty interesting thing, although he’s moved a little bit. He really played out the first half of his career understanding what the nature of computer architectures was. That has shifted for the last 15 or 20 years into understanding multiprocessors. He’s not focused on any particular thing. There is no particular machine that he wants to build. He wants to understand the space of multiprocessors and how to build effective multiprocessors. That is, in one sense, a kind of a lifetime effort of his. Although when I talked to him recently he was sort of running out of gas and asking what to do next.

There is another style that’s quite different, which I’ve labeled a sequence of strategic objectives. A kind of paradigm example was a scientist I knew about twenty years ago by the name of Werner Reichardt, who at the time I knew him, was studying the control system in insects that governed their flight. His view of the world — he was very articulate — was that you pick a particular scientific idea. That idea takes of the order of five years to build up and get deep enough so you really can do something, and then you pick another one. So, life for Werner is a sequence of five years, nominally five years project. Each one picked by looking at the state of science at the time.

There are in fact — their number is legion — a number of people, and I see lots of them around here, for whom the goal is simply to work on interesting problems. OK. You simply, in fact, if you can work on interesting problems, that’s all you ask of each particular day. There’s another variation on this, which is not so pleasant, in which you look at whatever seems publishable. You see a little project in life that looks like it’s publishable, let’s go for that one. OK. And I know a number of people whose scientific lives are a random walk among the publishable materials. [laughter] They, of course, shall all be nameless.

The last style I want to discuss is a single ultimate scientific question in which a scientist adopts a real goal out there, a scientific goal. I have three examples. Herb Simon is now famous for the fact that a single scientific goal of understanding how it is that humans, who are governed by bounded rationality, how all of the phenomena surrounding human cognition and decision making, can be understood by that. That in fact has driven his entire scientific career.

I haven’t checked this one out with Raj, but my belief about Raj is that he really has speech recognition as his only fundamental scientific goal. OK. Now Raj does lots of different things. He works on multimedia, works on vision, he’s worked on robotics. Only when he turns back to speech do you see sort of fantastic and interesting things happen. And he keeps coming back to it. So, for Raj, I infer — and I repeat, I didn’t ask him — I infer that his life is not all those things. It is really speech as the thing that he wants to really see done with his scientific life.

And for myself, I’m again one of these types of characters. My style is to deal with a single problem, namely the nature of the human mind. That is the one problem that I have cared about throughout my scientific career. And it will last me all the way, all the way to the end.

Scientific Styles - November 29, 2022 - john d. martin