Human Supercomputer Workshop

From Google Summer of Code Mentor Wiki

This was a workshop to explore new approaches to brainstorming and problem solving using the resources of a group.

[edit] Planned Architecture

If each person is a processor node, then the group 'is' a multiprocessor system. Insights from high performance computer architecture can be applied. From this perspective conventional brainstorming is low bandwidth - only one person is talking at a time.

The original plan was to use the workshop to bootstrap and debug a Systolic Array of people. This has higher bandwidth and has other advantages too. The initial part of the workshop would have involved explaining the principles, how two contra-flow pipelines introduce variable delay feedback loops and avoids processing hot-spots - and why and how to adapt this to people. However, my plans to bootstrap such a structure were thwarted by a major stall in a prior processor. The single pipeline lunch scheduling meant that people were arriving at the workshop at random intervals - well after the workshop was due to start.

[edit] Actual Architecture

To accommodate this, we fell back to organising as multiple clusters of three. This architecture is much more robust to disruption and still increases bandwidth (= more exchange) over a one-way lecture or over sharing in a large group. When new people arrived we moved people from existing clusters to explain what was going on, without everyone having to stop.

One of the steps in problem solving is refining the question. In each cluster we experimented with the WFO pattern:

• Worker: - Refines a question.
• Facilitator: - Acts as a resource to clarify the question and identify unvoiced preconceptions about the correct answer.
• Observer: - Notices what interactions were most effective and gives feedback at the end.

We used three sessions - to brainstorm on architecture (x1) to brainstorm on improving GSoC (x2). Each subcluster took a different angle on the problem. We formed new subclusters after each 15 minute session so that information could travel through the nodes. Finally we closed with feedback in the whole group.

Because of the repeated rearrangements of clusters each person had a different experience of the group. For me the most interesting of the three sessions was on increasing the number of women in GSoC.

[edit] Feedback

We finished with a closing feedback circle. The general feedback was positive. The pieces I particularly remember were:

• People really enjoyed the opportunity to experiment in this way - it was fun.
• Three in a cluster is a very good number.
• People would have liked more time for the component sessions.

Although we didn't have time to build a systolic array I felt the workshop achieved its objectives.