Monday, September 17, 2012

Research Areas and Social Identity

Last year, when I was giving the colloquium in Jyväskylä, my host introduced me as "leading the quantum gravity group at Nordita." I didn't object since it's correct to the extent that I'm leading myself, more or less successfully. However, the clustering of physicists into groups with multiple persons is a quite interesting emergent feature of scientific communities. Quantum gravity for example is usually taken to mean quantum gravity excluding string theory, a nomenclature I complained about earlier.

In the literature on the sociology of science it is broadly acknowledged that scientists, as other professionals, naturally segregate into groups to accomplish what's called a "cognitive division of labor": an assignment of specialized tasks which allows the individual to perform at a much higher level than they could achieve if they had to know all about everything. Such a division of labor is often noticeable already on the family level (I do the tax return, you deal with the health insurance). Specialization into niches for the best use of resources can also be seen in ecosystems. It's a natural trend because it's a local optimization process: Everybody dig a little deeper where you are and get a little more.

The problem is of course that a naturally occurring trend might lead to a local optimum that's not a global optimum. In the case of scientific communities the problem is that knowledge which lies at the intersection of different areas of specialization is not or not widely known, but there is a potential barrier preventing the community from making better use of this knowledge. This is unfortunate, because information relevant to progress goes unused. (See for example P. Wilson, “Unused relevant information in research and development,”. Journal of the American Society for Information Science, 45(2), 192203 (1995).)

So this is the rationale why it's necessary to encourage scientists to look out of their box, at least on occasion. And that takes some effort because they're in a local optimum and thus generally unwilling to change anything.

This brings me back then to the grouping of researchers. It does not seem to me very helpful to reach a better global optimum. In fact, it seems to me it instead that it makes the situation worse.

Social identity theory deals with the question what effect it has to assign people to groups; a good review is for example Stryker and Burke “The Past, Present, and Future of an Identity Theory”, Social Psychology Quarterly, Vol. 63, No. 4 (Dec., 2000), pp. 284-297. This review summarizes studies that have shown that the mere act of categorizing people as group members changes their behavior: When assigned a group, one that might not even be meaningful, they favor people in the group over people outside the group and are trying to fit in. The explanation that the researchers put forward is that "after being categorized of a group membership, individuals seek to achieve positive self-esteem by positively differentiating their ingroup from a comparison outgroup."

This leads me to think, it cannot be helpful to knowledge discovery to assign researchers at an institute to a handful of groups. It is also very punched-paper in the age of social tagging.

A suggestion that I had thus put forward some years ago at PI was to get rid of the research groups altogether and instead allow researchers to chose keywords that serve as tags. These tags would contain the existing research areas, but also cover other interests, that might be black holes, networks, holography, the arrow of time, dark matter,  phase transitions, and so on. Then, one could replace the groups on the website with a tag cloud. If you click on a keyword, you'd get a list of all people who've chosen this tag.

Imagine how useful this would be if you were considering to apply. You could basically tell with one look what people at the place are interested in. And if you started working there, it would be one click to find out who has similar interests. No more browsing through dozens of individual websites, half of which don't exist or were last updated in 1998.

I was thinking about this recently because Stefan said that with better indexing of abstracts, which is on the way, it might even be possible in the not-so-far future to create such a tag-cloud from researcher's publication list. Which, with an author ID that lists institutions, could be mostly automatically assembled too.

This idea comes with a compatibility problem though, because most places hire applicants by group. So if one doesn't have groups, then the assignment of faculty to committees and applicants to committees needs to be rethought. This requires a change in procedure, but it's manageable. And this change in procedure would have the benefit of making it much easier to identify emerging areas of research that would otherwise awkwardly fit neither here nor there. Which is the case right now with emergent gravity and analogue gravity, just to name an example.

I clearly think getting rid of institutional group structures would be beneficial to research. Alas, there's a potential barrier that's preventing us from making such a change, a classic example of a collective action problem. However, I am throwing this at you because I am sure this restructuring will come to us sooner or later. You read it here first :o)


  1. In the early days of PI, before you arrived, there were no research groups, and this was a deliberate choice. Everyone used to meet around the big table in bar area of the old building for lunch and we would discuss everything from AdS/CFT to Bohmian mechanics together as one group. However, gradually, over time, de facto groups emerged for various reasons, including grouping around seminars, the formation of discussion groups, grouping around mentoring, and the fact that people were used to being in groups at their previous institutions.

    By far the biggest factor was the difficulty in evaluating postdoc and faculty candidates across different fields. This really was a much larger problem than you have indicated, since reading a string theory paper was indistinguishable from reading gobbledegook for me at that point in my career and, similarly, non foundational researchers had a lot of trouble distinguishing good work in quantum foundations from utter nonsense. This led to the current system of field-based selection committees who present their findings for discussion to everyone else.

    I am not sure that the group structure at PI is any more "official" today than it was at the outset. It is an emergent phenomenon that happened as the institute got larger. Therefore, I don't think any sort of official disavowal of research groups on the website or anywhere else would work, and I am skeptical of researcher's enthusiasm for curating their own tags, especially when many of them can't even be bothered to maintain a publication list. Groups naturally form around research topics, no matter how much you try to prevent it.

    Probably the best way of preventing research groups from forming is to keep the institute very small (it's too late for PI to do this at this point). If there isn't a quantum information or foundations seminar/group meeting somewhere every day then I am much more likely to attend a quantum gravity or condensed matter seminar or group meeting. Once the level of scientific activity becomes so high that activities in one research area fill all of my time then I am not going to get exposed to other areas in practice, whatever my intentions.

  2. Management is a chain of command. Product is irrelevant. Amorphous engines of creation offer no quantitaive metrics for process control and managerial reward therefrom. Lucent Technologies tamed Hell’s Bells Laboratories by ending it. The abandoned building was a brick and mortar slum bereft of incentivizing jaunts into universal architecture space. The people mattered, not the container or its cork.

    Gravitation demands boson photon vacuum symmetries.
    Fermions are chiral in 4-D, left shoes and right shoes.

    Fermionic matter discerns a trace chiral vacuum left foot, hence parity "violations." Quantum gravitations and SUSY are not empirical because they are demonstrably wrong. Gravitation can only be repaired external to defective postulated truth. Whom would you assemble, Bee, to vex every Rex?

  3. Hi Matt,

    I think you're misunderstanding me.

    I didn't say and didn't mean to say that one should get rid of research groups. Read what I wrote: I wrote it's a natural trend and one that is beneficial - up to a point. All I'm saying is to get over this point one shouldn't reinforce scientist's ingroup behavior by officially assigning them groups upon arrival. You know as well as I do that there have always been candidates at PI who were assigned two different groups - and sometimes still didn't really fit.

    Neither am I saying everybody is supposed to read everybody's paper or applications, heaven forbid. Again, you're misunderstanding me. What I'm suggesting is far less radical than you think because for most people it will be almost the same as the usual group-structure. It can however more easily accommodate those people who sit at the intersections, and it can serve to prevent that people start identifying with some isolated research area if it's not necessary.

    Yes, I am skeptic too that researchers will take even the 3 minutes that it takes to select a few tags from a list. Though that's less work than setting up a website, and with some encouragement I think one should be able to get most of them to do it. However, this is why I was saying it's not far from hereon that the tags can simply be extracted from the abstracts of their papers.



  4. This comment has been removed by the author.

  5. Hi Bee,

    This piece has me wondering perhaps if fixed research groups are actually such good thing, as what has them to be found to be antiquated if no longer representative as being meaningful research directions. That is to ask, how does one avoid something from being continued as only for the sake of its continuance? I then tend to agree that perhaps tags being the more practical way to have both researchers and research to be identified, since tags are more easily dropped without having it seeming like something has been eliminated and more readily added to have something new able to enter the mix. I thus think that the only groups that should remain are the central ones, such theoretical physics and experimental physics, with even here wondering where one ends to have the other begin. That is in the end I think it more important to science that the openness of the dialogue should take precedent over categorizing what it need be about.

    “It is proposed that a form of free dialogue may well be one of the most effective ways of investigating the crisis which faces society, and indeed the whole of human nature and consciousness today. Moreover, it may turn out that such a form of free exchange of ideas and information is of fundamental relevance for transforming culture and freeing it of destructive misinformation, so that creativity can be liberated.......What is essential here is the presence of the spirit of dialogue, which is in short, the ability to hold many points of view in suspension, along with a primary interest in the creation of common meaning..”

    -David Bohm & F. David Peat, “Science Order, and Creativity”



  6. I feel like defending Matt Leifer.
    You wrote: "I clearly think getting rid of institutional group structures would be beneficial to research." You didn't write "- up to a point." which makes your whole argumentation sound very differently then previously. In particular I interpreted your text similar to Matt Leifer. But since you have clarified this issue now, there is at least for me no need to discuss the possible misinterpretations of your text any further.

    I can to some extend imagine that assigning people to a group may affect their behaviour (as in the paper you have linked to). So this might pose an obstruction to research. However as Matt Leifer already pointed out groups often emerge by themselves. Last but not least people are different in their way of learning and understanding things and people may tend to coalesce more easily with those who they think they can communicate with easier. So actually assigning people to groups without taking this "automatic" grouping like via communicational features into account may eventually even enhance "degrouping". For example at the mathematical institute Oberwolfach people are each lunch placed randomly to a table, i.e. they are "forced" to eat together with people who are belonging to a different field and this assignment to "table groups" actually furthers in some sense the "degrouping" of the research field groups.

    So I think identifying people via their research field may not pose big problems in comunication behaviour if one has mechanisms to adress this like in the above Oberwolfach example.
    I do however think that group structures in research may be annoying if there are organisatorial obstacles coming with them, like for example that one group has some money left, which another group could need, but that one can't transfer the money to that other group for organisatorial reasons.

    Regarding the tagging: I think Tag clouds, as you proposed them, may be a more or less useful supplement, however they may not be sufficient. Some time ago I had made a suggestion to an official at some academic institution here in Berlin that it would be good if members at this institution would identify their special expertises, because I was looking for an expert at that institution via searches and I couldn't find one. However the quest I had was so special that a tag cloud with such a speciality depth would get just too messy. You would need more something what is called "facetted search" (which could be e.g. visualized by a tag cloud you could zoom in) and/or better search algorithms. And of course as a precondition you would need that researchers identify their fields of expertise. Likewise by the way they may also identify their fields of interests etc. But as Matt Leifer pointed out it may not so easy to make people do that identification by themselves. So actually it could be that one would need to assign people to research areas in order to implement such search options across scientific areas. :O


COMMENTS ON THIS BLOG ARE PERMANENTLY CLOSED. You can join the discussion on Patreon.

Note: Only a member of this blog may post a comment.