Institutions & Science

Busy as usual with talks, models, papers & grants. Less papers recently, actually -- except reviewing other people's. But anyway, the thing that stands out that I've been meaning to blog about for a while was a KLI talk by Sabina Leonelli. She is even more interdisciplinary than I am -- a philosopher of science, a historian and a sociologist. She is studying how the use of shared computer databases is affecting science.

A lot of huge results now, like the human (and other animal) genomes are available as resources to anyone over the internet. This allows a lot of work to be done by AI -- machine discovery of possible new drugs and so forth, where you just need to do giant pattern matching, which only takes time & data. In order to facilitate both human and machine exploration of all this data, there are large projects devoted to labeling all the important data in a uniform way. The part of the web that is marked up this way is called the semantic web, and the collections of labels that are agreed on for doing that are called ontologies.

I worked on the semantic web myself briefly for about four months with Lynn Andrea Stein right after my PhD in 2001. She said I had to write a paper about it, so I tried to find something to hack on, but as far as I could tell at least at that time there was no part of the semantic web that actually worked. It was just a bunch of protocols & languages people talked about. I wrote an article anyway about how services should be integrated over the semantic web. I sent it to the editors of a special issue on the topic, and they asked me for both a shorter version of the paper for their journal and a longer version for a book they were producing. (For some reason I can't get Google Scholar to show both versions on the same page.) Anyway, taken together those papers have been cited 58 times, which makes it like the fifth most academically-popular thing I've done, even though there's no code in it. Or experiments. I think this mostly just shows how many people are working on the semantic web. But what people do like about the article is the image of web services as extensions to the intelligence of agents that work for people, rather than being either just passive data or themselves active agencies. Though there is one other useful contribution of the paper which is trying to get the logicians who were designing the semantic web languages to think about representing and integrating components that are unreliable real-time systems.

Well, back to the topic, apparently the semantic web is actually getting usable these days, and Sabina Leonelli is studying the impact of this on science. When there is a single agree-on ontology, does that become perceived as the truth? The nature of science and knowledge is that we are always improving on it. That's why the articles about how many articles in leading journals are "wrong" are so ridiculous -- if articles get corrected then they are in the scientific process. The ones that are not corrected or extended are dead.

The concern that ontologies bias science is to me an example of two concerns that I run into a lot. First, that once you introduce technology a problem changes qualitatively. This is a lot like the hopes & fears for AI stuff I've written about before. There is almost always a dominant thread of scientific opinion, but there is hardly ever only a single thread. In Sabina's talk I mentioned the official lists of psychological disorders, which at one point included homosexuality and now do not. What's in the list doesn't determine where science goes, but it does determine a lot of stuff about public spending, what insurance will pay for, and so on. So the introduction of ontologies just gives us an occasion to think about the nature of science, the influence of money and so on again. But so far I don't think it actually creates new problems.

The other concern is one that has come up over & over in scientific discussions at the KLI in the last two years. The question is, is it right that science should ever be biased. Should we believe the mainstream? Won't good ideas get lost?

There's something very attractive about the idea that being objective means being without bias, but anyone who works on computational models of planning or learning knows that bias is actually a necessary and useful mechanism. Do we really want to think that it's as easy to light a fire with water as with a match? If we were without bias we'd be without knowledge and our actions would be random.

In this sense, science & evolution both work a lot the same. They select the theories / genomes most likely to do well, but they also need to have enough variation around so that in science, ideas & knowledge can keep moving forward, and in evolution, so that changes in the environment can be dealt with and better solutions to living and reproducing found. Sticking with science, one way to think about it is that as a community we should allocate our time exploring a particular idea in proportion to its probability of being right. In practice, this usually means that there are a bunch of people who believe different things, and they & their labs study these different ideas and try to convince other scientists they are right.

The kind of bias that is bad is when people are blocked from researching ideas that they think are probably true, or forced to spend lots of time researching ideas they think are probably wrong. That throws off the system, and means useful & important ideas might get neglected. But that most current scientists study things at least mostly compatible with the mainstream view just shows that we respect and (more importantly) use the knowledge discovered by the previous generations of scientific culture.

Anyway, I loved Sabina's talk. Fortunately, she has a job in Exeter now, and her husband is a smart & fun mathematician, so we hope we will be able to get together and talk about science a lot when I get back to Bath.

I've given versions of my own talk on evolution, culture & cognition in Brighton (Sussex), Budapest (CEU), GrĂ¼nau (Vienna people), & Oxford in the last month. I'm getting a lot of help and enthusiasm from people at the talks -- hopefully eventually I'll get that translated into some more publications and grants! Robin Dunbar, who inspired my entire approach to action selection with some of his talks in Edinburgh in the 1990s, has moved his lab to Oxford & has very smart postdocs. Again, that's quite close to Bath so hopefully I can visit more once I go back.

Sorry for the fuzziness of the Oxford picture, it's from the laptop, taken out the window of the room I stayed overnight in at Magdalen College. I really need to get a phone with a good camera in it again. I will probably do so quite soon now. The second picture is from Patricia, who managed to convince both sons & partners to go to Ludlow for lunch.