Communications of the ACM
A Complex Systems Perspective on Computer-Supported Collaborative Design Technology
Collaborative design is challenging in that strong interdependencies between design issues make it difficult to converge on a single design that satisfies these dependencies and is acceptable to all participants. Current collaborative design processes are typically characterized by multiple iterations and/or heavy reliance on multifunctional design reviews, both of which are expensive and time-consuming; poor incorporation of some important design concerns, typically later life-cycle issues such as environmental impact; and reduced creativity due to the tendency to incrementally modify known successful designs rather than explore radically different and potentially superior ones.
This article examines what complex systems research can do to help address these issues.
Defining Collaborative Design
A design (of physical artifacts such as cars and planes, as well as behavioral ones such as plans, schedules, production processes or software) can be represented as a set of issues (sometimes also known as parameters) each with a unique value. If we imagine that the possible values for every issue are each laid along their own orthogonal axis, then the resulting multidimensional space can be called the design space, wherein every point represents a distinct (though not necessarily good or even physically possible) design. The choices for each design issue are typically highly interdependent. Typical sources of interdependency include shared resource (weight and cost) limits, geometric fit, spatial separation requirements, I/O interface conventions, and timing constraints, to name a few.
Collaborative design is performed by multiple participantsrepresenting individuals, teams or even entire organizationseach potentially capable of proposing values for design issues and/or evaluating these choices from their own particular perspective. An example is manufacturability.
Some designs are better than others. In principle, we can assign a utility value to each design and thereby define a utility function that represents the utility for every point in the design space. The goal of the design process can thus be viewed as trying to find the design with the optimal (maximal) utility value.
The key challenge raised by the collaborative design of complex artifacts is that the design spaces are typically huge, and concurrent search by the many participants through the different design subspaces can be expensive and time-consuming because design issue interdependencies lead to conflicts when the design solutions for different subspaces are not consistent with each other. Such conflicts severely impact design utility and lead to the need for expensive and time-consuming design rework.
Insights from Complex Systems Research
A central focus of complex systems research is the dynamics of distributed networks, for example networks in which there is no centralized controller, allowing global behavior to emerge solely as a result of concurrent local actions. Such networks are typically modeled as multiple nodes, in which each node represents a state variable with a given value. Each node in a network tries to select the value that optimizes its own utility while maximizing its consistency with the influences from the other nodes. The global utility of the network state is simply the sum of node utilities plus the degree to which all the influences are satisfied. The dynamics of such networks emerge as follows: since all nodes update their local state concurrently based on their current context (at time T), the choices they make may no longer be the best ones in the new context of node states (at time T+1), leading to the need for further changes.
Is this a useful model for understanding the dynamics of collaborative design? We believe it is. It is straightforward to map the model of collaborative design presented onto a network. We can map design participants onto nodes, where each participant is trying to maximize the utility of the choices it is responsible for, while ensuring its decisions satisfy the relevant dependencies (represented as the links between nodes). As a first approximation, it is reasonable to model the utility of a design as the local utility achieved by each participant plus a measure of how well all the decisions fit together. Even though real-world collaborative design clearly has top-down elements, the sheer complexity of many design artifacts means no one person is capable of keeping the whole design in his or her head. Centralized control of the design decisions becomes impractical, so the design process is dominated by concurrent local activities.
How do such distributed networks behave? Let us consider a simple example: a network consisting of interlinked, binary-valued nodes. At each time step, each node selects its own value, the same as the majority of the nodes in which it is linked. We can imagine using this network to model a real-world situation wherein six subsystems are being designed, and we want them to use matching interfaces. The network has converged onto a local optimum (no node can increase the number of influences it satisfies by a local change), so it will not reach a global optimum where all the nodes have the same value (see Figure 1).
Generally speaking, networks may not always converge upon the global optimum, and in some cases (as we shall see with dynamic attractors), a network may not converge at all. Insights into whether and how global optima can be found in networks represent the heart of what complex systems research offers to the understanding of collaborative design.
The key factor determining network dynamics is the nature of the influences between nodes. We consider two important distinctions: whether the influences are linear or not; and whether they are symmetric or not. We will then discuss subdivided network topologies and the role of learning. Unless indicated otherwise, the following material presented on complex systems is drawn from [1].
Linear Vs. Nonlinear Networks
If the value of nodes is a linear function of the influences from the nodes linked to it, then the system is linear. Linear networks have a single attractor: a single configuration of node states that the network converges toward no matter what the starting point, corresponding to the global optimum. This means a "hill-climbing" approach can be used (where each node always moves directly toward increased local utility) because local utility increases always move the network toward the global optimum.
Nonlinear networks, by contrast, are characterized by having multiple attractors and multiple-optima utility functions, as shown in Figure 2.
A key property of nonlinear networks is that searching for the global optima cannot be performed successfully by pure hill-climbing algorithms because they can get stuck in local optima that are globally suboptimal. For example, consider what would happen if the system started searching at Point 1 (Figure 2). Hill-climbing would take it to the top of the local optimum, which is substantially lower than optima in other regions of the utility function. Hill-climbing would do even more poorly if started at Point 2.
One consequence of this reality is a tendency to stay near well-known designs. When a utility function has widely separated optima, once a satisfactory optimum is found, the temptation is to stick to it. This design conservatism is exacerbated by the difficulty when comparing the utilities for radically different designs. We can expect this effect to be especially prevalent in industries such as commercial airlines and power plants, which are capital-intensive and risk-averse, since in these contexts the cost of exploring new designs, (and the risk of getting it wrong), can be prohibitive.
An emerging range of techniques are appropriate for finding optima in ultrametric utility functions, all relying on the ability to search past valleys in the utility function. For example, simulated annealing endows the search procedure with a tolerance for moving in the direction of lower utility that varies as a function of a virtual "temperature." At first the temperature is high, so the system is as apt to move toward lower utilities. This allows a wide range over the utility function and finding new higher peaks. Since higher peaks are also typically wider ones, the system tends to spend most of its time in the region of high peaks. Over time the temperature decreases; the algorithm increasingly tends toward pure hill-climbing. While this technique is not provably optimal, it has gotten close to optimal results in most cases.
Annealing, however, runs into a dilemma when applied to systems with multiple actors. Let us assume that some actors are self-interested hill-climbers, concerned only with directly maximizing their local utilities, while others are annealers, willing to acceptat least temporarilylower local utilities in order to increase the utility in other nodes. Simulation reveals that while the presence of annealers always increases global utility, individually they fare worse than hill-climbers when both are present [3]. The result: globally beneficial behavior is not individually incented.
How do these insights apply to collaborative design? Linear networks have been used successfully to model routine design [2], involving highly familiar requirements and design options, for example, in automobile brake or transmission design [4]. Today's most challenging and important collaborative design problems (software, biotechnology, or electronic commerce, to name a few) are not just instances of routine design. They typically involve innovative design, radically new requirements, and unfamiliar design spaces. As a result, it is often unclear where to start to achieve a given set of requirements. There may be multiple and different solutions, and the best may be radically different than any tried before. For such cases, nonlinear networks seem to represent a more accurate model of the collaborative design process.
This has important consequences. Simply instructing each design participant to optimize its own design subspace as much as possible can lead to the design process getting stuck in local optima significantly worse than radically different alternatives. Design participants must be willing to explore alternatives that, at least initially, may appear much worse from their individual perspective than current alternatives. Designers often produce a good design for the subsystem, for which they are responsible, rather than conceding to make someone else's job easier. We need to find solutions for this dilemma concerning the lack of individual incentives for such globally helpful behavior.
Symmetric Vs. Asymmetric Networks
Symmetric networks are ones in which influences between nodes are mutual (for example, if node A influences node B by amount X then the reverse is also true). Symmetric networks do not have this property. Asymmetric networks (with an exception to be discussed later) add the complication of dynamic attractors, which means the network does not converge on a single configuration of node states but rather cycles indefinitely around a relatively small set of configurations. Let us consider the simplest possible asymmetric networkthe "odd loop" (see Figure 3).
This network has two links: one that influences the nodes to have the same value; the other influences them to have opposite values. Let us start with node A having the value 1. This influences node B to have the value 1, which in turn influences node A toward the value 1, which in turn causes node B to flip values again, and so on ad infinitum.
In most commercial airplanes, for example, the engine and wing subsystems are designed separately, taking advantage of standardized engine mounts, and allowing the use of a range of different engines. This is not the optimal way of relating engines and wings, but it is good enough and simplifies the design process considerably.
Current collaborative design practice is characterized by asymmetric influence loops likely to produce dynamic attractors and therefore nonconvergent dynamics. Feedback from later product life-cycle perspectives such as manufacturability and transportability, for example, tends to be weaker and slower than influences from design to these perspectives.
Subdivided Networks
Another important property of networks is whether or not they are subdivided (whether they consist of sparsely interconnected "clumps" of highly interconnected nodes). When a network is subdivided, node state changes can occur within a given clump with only minor effects on the other clumps. This has the effect of allowing the network to rapidly explore more states. This effect, known as modularization, is widely exploited in design communities. This involves intentionally creating subdivided networks by dividing the design into subsystems with predefined standardized interfaces, so subsystem changes can be made with few or any consequences to the other subsystems. The key to the success of this approach is defining the design decomposition such that the impact of the subsystem interdependencies on the global utility is relatively low, because the standardized interfaces rarely represent an optimal way of satisfying these dependencies. In most commercial airplanes, for example, the engine and wing subsystems are designed separately, taking advantage of standardized engine mounts, and allowing the use of a range of different engines. This is not the optimal way of relating engines and wings, but it is good enough and simplifies the design process considerably. If the engine-wing interdependencies were crucial, for example, if standard engine mounts had a drastically negative effect on the airplane's aerodynamics, then the design of these two subsystems would have to be coupled more closely in order to produce a satisfactory design.
Imprinting
One common technique used to increase network convergence is imprinting, in which the network influences are modified when a successful solution is found in order to facilitate finding similar solutions next time. A common imprinting technique is reinforcement learning. In this approach, the links representing the influences satisfied in a successful configuration are strengthened, and those representing violated influences weakened. The effect of this is to create fewer but higher optima in the utility function, thereby increasing the likelihood of hitting such optima in the future.
Imprinting is a crucial part of collaborative design. The configuration of influences between design participants represents a kind of social knowledge that is generally maintained in an implicit and distributed way within design organizations in the form of individual designer's heuristics about who (for example, which individual or design group) should talk to whom when, and about what. When this knowledge is lost, say, due to high personnel turnover in an engineering organization, the ability of that organization to accomplish complex design projects is compromised. It should be noted, however, that imprinting reinforces the tendency for organizations in nonlinear design regimes to stick to tried-and-true designs by virtue of making the previously found optima more prominent in the design utility function.
How Can We Help?
Once the design of a complex artifact has been distributed to many players, encouraging proper influence relationships and local search strategies is the primary tool available to design managers, and should therefore be supported by computer-supported collaborative design technology. This can occur in several ways. It can help monitor the influence relationships between design participants, allowing someone to track the volume of design-related exchanges or (a more direct measure of actual influence) the frequency with which design changes proposed by one participant are accepted as-is by other participants. This can be helpful in many ways. For example, highly asymmetric influences could represent an early warning sign of nonconvergent dynamics. Detecting a low degree of influence by an important design concern, especially as environmental impact that has traditionally been less valued, can help avoid utility problems down the road. A record of the influence relationships in a successful design project can be used to help design similar future projects. Influence statistics can also be used to help avoid repetitions of a failed project. If a late high-impact problem occurred in a subsystem with a low influence in the design process, this would suggest the influence relationships should be modified in the future. This has the effect of making a critical class of normally implicit and distributed knowledge more explicit, and therefore more amenable to being preserved over time (despite changes in personnel) and transferred between projects and even organizations.
Computer-supported collaborative design technology can also help assess the degree to which the design participants are engaged in routine versus innovative design strategies. For example, such systems could be used to estimate the number and variance of design alternatives being considered by a given design participant. This is important because, as we have seen, a premature commitment to a routine design strategy that optimizes a given design alternative can cause the design process to miss other alternatives with higher global optima. Tracking the degree of innovative exploration can be used to fine-tune the use of innovation-enhancing interventions, including incentives, competing design teams, and introducing new design participants.
References
1. Bar-Yam, Y. Dynamics of Complex Systems. Addison-Wesley, Reading, MA., 1997.
2. Brown, D.C. Making design routine. In Proceedings of IFIP TC/WG on Intelligent CAD, 1989.
3. Klein, M., et al. Negotiating Complex Contracts. Massachusetts Institute of Technology, Cambridge MA, 2001.
4. Smith, R.P. and Eppinger, S.D. Identifying controlling features of engineering design iteration. Management Sc. 43, 3 (1997), 276293.
Figures
Figure 2. An ultrametric utility function.
Figure 3. The simplest possible asymmetric networkan "odd loop."
©2002 ACM 0002-0782/02/1100 $5.00
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2002 ACM, Inc.