My notes and other stuff


Paper: Flexible Sketches and Inflexible Data Bases

This week I got nerd-sniped by one of Lorin Hochstein's blog posts into reading Kathryn Henderson's Flexible Sketches and Inflexible Data Bases: Visual Communication, Conscription Devices, and Boundary Objects in Design Engineering. This is a paper that was written as a piece of sociological work after the author embedded herself in a mid-size company that designs turbine engines, as a technical writer (participant-observer), for over a year.

The work is from the late 80s/early 90s and takes place in a company that has been shrinking for two decades as the nature of turbine design's market shrunk after the reductions of the aerospace industry, oil & gas, and competition from Japanese companies. The projects she was on had been part of a series of designs where new generation models needed to be more powerful at a lower manufacturing costs. The organization was undergoing a transition from manually done technical drawings that were then made official in CAD/CAM systems for their final copy, into using said CAD/CAM systems throughout the design for the first time.

That transition provides a great opportunity to observe technical drawings as boundary objects. Boundary objects are items that let members of different groups read different meanings from the same objects, for their own needs. They are seen as a tool for distributed cognition when they are flexible when used across groups, but focused within individual group use, since they let people with different backgrounds and perspective call upon their respective [tacit] knowledge. This concept registers in a broader one where technology is framed as a social process that is both shaped by society, but then shapes society back.

Put another way, when the engineers who know a lot about laying a turbine's support structure come up with a sketch of it, they can use it as a tool to talk to welders about their new design. The welders look at it, and they may be able to reduce the number of welds when it comes to assembly without any loss of quality. Taking that contribution back, the engineers will update the drawings and bake part of the expertise of welders into the design. That set of informal drawings are boundary objects that lets engineers and welders work together to shape the turbine's system, and this exercise will in turn create ongoing relationship across these departments.

The author points out that the flexibility that allows for multiple interpretations is a crucial quality that facilitates that sharing. A fully integrated approach that is absolutely unambiguous would not be able to serve the same role. The ability to join together the individual localized interpretations into a larger jointly inhabited frame is the core point of boundary objects. To do that, they have to be strongly structured in individual use, but weakly structured in common use.

The company studied here made a heavy use of these drawings as boundary objects:

[The] senior designer on the Mark IV, points out that [technical drawings] are used not only between design engineers but also to facilitate consultation between designers and those in the production cycle. He discusses his ideas early in the design process with people in "structures" [such as the unit that does welding].


[He] points out that a meeting with people in structures, where the package is actually built "works wrong if you can't communicate." He takes his drawings to "sit and talk," so others understand what he is trying to do. Likewise he draws sketches for workers so they can understand what they are supposed to do to create what he has in his mind that may not be clear in their minds from looking only at layout drawings. This interactive use of conscription devices as boundary objects knits together the input of people with expertise from shop experience such as the welders in the "structures" division with the combined knowledge of various designers. It taps individual expertise in piping, electrical circuitry, lubrication systems, and so on to socially construct a machine built through organized, distributed cognition.

While the initial drawings are done iteratively with feedback and a lot of throwaways, they are somewhat independent and context-bound. But once the first set is ready, the drawings then get switched to CAD, where they connect to all sorts of databases, and then get used for parts, inventory, sales records, and marketing. This then ties into more departments and longer-lived concerns, and can become the base for future designs as well.

In the adoption story told in this paper, the aim was to start using CAD throughout rather than just at the end, as more than an accounting mechanism. This, as it turns out, was a key risk:

The fluidity and flexibility that is part of the loose structure of boundary objects was paralyzed by the fact that the whole system was computerized. [...] The huge size and complexity of the interlocking systems intimidated people. Employees were afraid if they adjusted information in their part of the system and made a small error such as the misplacement of a decimal point—a common error—that monumental consequences could result, such as the ordering of excess inventory.

By connecting the drawings to so many more areas of concern, it reduced their flexibility and overextended them in such a way they could no longer function well as boundary objects. This was made worse because of assumptions built into the CAD system about how the worfkflow should work: idea, drawing, prototype, and then production. In practice, the design is very much iterative and far less linear. These assumptions are what made promises of efficiency but that in fact could disintegrate the communication mechanisms that made the work process self-repairing.

These issues, coupled with recent layoffs that made employees secretive about their knowledge (to protect their jobs), led to two parallel work systems being created. A formal system, using the CAD tooling that tracked everything, and an informal system, where people used their connections and hand sketches (the old tools, which were seen as safe) to get what they needed in terms of information and from their workflow.

This of course led to further problems. By using informal sketches, the inventory part of the system had been circumvented and parts were not ordered on time. As people on the shop floor made design alterations, those who held the final CAD designs could no longer check alterations and lost control. This created "hip-shooting" where changes also bypassed analysis steps for their impacts and costs. Further drift was created when some departments used the database as a direct check and no longer required to collaborate over boundary objects to get their answers.

The author state that if the boundary object is to become all-encompassing of all the interlocking information, it must either dictate the restructuring the whole setting (and change the workflow top-down), or must remain flexible enough.

The parallel system keeps going long, however, and marketing departments, engineers, and managers will all act as if the final drawings have accomplished the task rationally and efficiently, even if in practice the informal drawings keep existing and being used, with changes recorded after the fact.

The author concludes:

There are no one-way relationships between actants, mental models, representations, and the constructed technology. Rather, interactive practices contribute to the mutual and simultaneous construction of each by the others. The representations are the product of and resources for situated practice. The destruction of such visually oriented situated practices may occur because of a fundamental misunderstanding of their crucial role in the social organization of distributed cognition in team design work. When such fundamental misunderstandings are built into inflexible computer graphics programs designed with the misleading idea of a definable linear process from concept to design to production, then the social mechanisms that ordinarily repair frequently occurring problems are left out of the process with potentially disastrous results.

She slipped a line earlier in the paper that also sticks with me: the design reflects the organizational problems of the company that built it.