Monday, December 12, 2011

Organizing human cognition: Lessons from CERN

There's a hierarchy in big time science schools, and physics holds the crown. (Math majors are in a different league.) Physicists are, face it, smarter than the rest of us -- and they know it.

Our only consolation is that they often work for a pittance.

So, from my perspective as a corporate ant, it's fascinating to read John Conway's description of how physicists organize their collaboration on history's biggest physics project (emphases mine)...
Making the (Higgs) Sausage | Cosmic Variance | Discover Magazine
For the past year, physicists at the LHC experiments CMS and ATLAS have been analyzing ever–increasing data samples from the huge machine. Rumors are now circulating about what the experiments might announce at next week’s presentations at CERN regarding the search for the Higgs boson.
... As you probably know, each of the two big experiments has over 3000 physicists participating, from all over the world. Many, but by no means the majority, are resident at CERN; most are at their home institutions in Europe, North America, and Asia and elsewhere.
The main thing that allows us to collaborate on a global scale like this is video conferencing. We used a system called EVO, developed at Caltech, which allows us to schedule meetings and connect to them from a laptop or desktop computer, or even dial in by phone ...the experiments have gravitated toward having meetings in the late afternoon, Europe time, which makes it early morning for people like me in California.
.., In CMS, our whole system of producing physics results has a sort of pyramidal structure. Each experiment has a number of physics analysis groups which meet a weekly or biweekly, typically, and have two “conveners” who set the agenda and run the meetings. These convener positions are typically held by senior people in the collaboration such as professors or senior lab scientists, for two years at a stretch, one convener changing out each year. They report to an overall physics coordinator and his or her deputies.
Within the physics analysis groups are subgroups devoted to sets of analyses which share common themes, common tools, or similar approaches. Each of these subgroups in turn is led by a pair of conveners who establish the ongoing analyses and guide them to eventual approval within physics analysis group.
We have what I think is a pretty impressive internal website devoted to tracking the progress of each physics analysis. From a single website you can drill down into a particular physics group find the analysis you want get links to all the documentation, and follow what’s happening. In parallel, there is a web system for recording the material presented at every meeting.
The goal of every analysis is to be approved by its physics group, so it can be shown in public at conferences and seminars. This requires having complete documentation including internal notes with full details of the analysis, and a “public analysis summary” which is available to the public, and which often serves as the basis for a peer–reviewed paper which soon follows.
Every analysis is assigned an analysis review committee of three to five people with experience in the topic, who act as a sort of hit squad, keeping the analyzers on their toes with questions and comments at every stage of the analysis, both on the actual analysis details and on the documentation. After all, if we are not our own worst critics, someone else will gladly fill the role!
In parallel with processing the data that we record, we run full simulations of well–known standard model collision processes which represent our background when we are doing searches for new particles. There is a big organizational challenge in doing these simulations, which run on a worldwide grid of computers devoted to CMS data analysis. We make use of the Open Science Grid for this in the US, the EuroGrid in Europe, and other clusters scattered all around the world, comprising tens of thousands of computing nodes.
I'd love to see comparisons to organizational structures used in aerospace projects. There's nothing like this large scale organization in the industry I work in.

This framework for harnessing cognition reminds me of the original "computers" - humans who did large scale arithmetic calculations prior to the development of log tables. It's easy to imaging who this would map onto a cognitive unit made up of, initially, humans and AIs.

PS. Historical footnote: CERN was where Tim Berners-Lee, working as an independent contractor, led the development of the first web site and browser.

No comments: