Shuichiro Ogawa
日本語

Notes · updated 2026-09-22

What Social Coordination Costs Arise from Collaborating with Others on Creative Work

Adding people does not make work progress in proportion to headcount; the difference disappears into coordination loss and motivation loss (Steiner 1972), and into a relational loss from feeling that it is harder to get help (Mueller 2012).

Contents (6)
  1. Why Effort Doesn’t Scale with Headcount
  2. Six Sources of Coordination Cost
  3. Where Does Coordination Bite into Creative Work?
  4. Mechanisms That Make Coordination Cheaper, and What Is Lost in Return
  5. If the Partner Is AI, Does Coordination Cost Disappear?
  6. What Remains Unknown

Suppose two people are to write a single piece of text together. Twice the hands, surely, means it gets written in half the time.

But once writing begins, opinions split over the phrasing of a single paragraph, and work stalls over which version to adopt. Farkas wrote of the form in which an entire text is co-drafted by several people that “two (or more) people do one person’s work, and in fact often produce less than one person would have.” He also wrote that discord and frustration over word choice and sentence construction arise frequently.

Something similar happens when people pool ideas. Gather people for face-to-face brainstorming, and the quantity and quality of ideas produced are lower than when the same number of people think up ideas separately and their outputs are pooled.

Where does the added effort of extra people go? And when what is being made is a creative output such as a text or an idea, does the way it disappears take a distinctive shape?

Why Effort Doesn’t Scale with Headcount

Steiner formulated a group’s actual productivity as “potential productivity minus losses due to faulty process.” This loss is called process loss. Steiner split the loss into coordination loss, where the combination of effort goes wrong, and motivation loss, where individuals hold back effort.

The two losses can be separated experimentally. Ingham and colleagues created a pseudo-group condition in which a person was actually exerting effort alone but was made to believe they were doing so together with a group. With no one actually beside them, a failure to stay in sync could not occur. Effort still dropped. Coordination failure alone cannot fully explain the group’s loss.

Latané and colleagues, in an experiment using clapping and shouting, showed that the more people a participant believed were in the group, the lower the volume produced per person. This phenomenon is called social loafing.

Coordination has its own definition as well. Malone and Crowston defined coordination as “the management of dependencies between activities.” Coordination work arises wherever such dependencies exist — one person cannot start until another finishes, or two people use the same resource. In economics, Coase argued that where work gets bundled is decided by comparing the cost of transacting in the market against the cost of coordinating within an organization, and Williamson systematized this as transaction cost economics. That coordination itself carries a cost has long been a premise of the theory.

That cost grows with the number of people. Brooks pointed out that when n people communicate with each other, the number of communication channels becomes n(n-1)/2. Three people means three channels, six people means fifteen, ten people means forty-five. In the same book, Brooks also wrote that adding people to a late software project makes it later still.

Some losses do not fit within Steiner’s two-way split. Mueller examined 26 teams comprising 212 people and showed that relational loss — the sense, stronger in larger teams, that it is harder to get help — mediates the negative relationship between team size and individual performance. This is a loss that arises not from a failure of logistics nor from a drop in motivation, but from the perception of relationships. This note treats it as a third kind of loss.

Six Sources of Coordination Cost

Lumping everything under the single term “coordination cost” mixes things that arise for different reasons. The effort of staying in contact, the awkwardness of speaking up in front of others, and the negotiation over how credit is divided are all costs of collaboration, but the mechanism that produces them and the conditions under which they bite both differ.

The Effort of Aligning Logistics and Information

The first is the effort of keeping track of who has done what, and how far.

Kraut and Streeter surveyed 65 software development projects and reported that the larger the scale, interdependence, and uncertainty, the less formal procedures alone suffice, and the more informal communication is needed for coordination. Formal mechanisms such as procedure manuals alone do not keep logistics aligned.

This effort grows when crossing distance or organizational boundaries. Olson and Olson identified four factors that separate success from failure in remote collaboration: shared assumptions (common ground), the strength of the coupling of work, readiness for collaboration (collaboration readiness), and readiness for the technology of collaboration (collaboration technology readiness). Cummings and Kiesler examined 62 interdisciplinary collaborative research projects and reported that the more universities a project involved, the significantly less adequate its coordination, and the fewer its results.

Some research has counted this effort. Kittur and colleagues defined the cost of conflict and coordination on Wikipedia as “extra work not directly tied to new article content,” and showed that as Wikipedia grew, the share of edits going to articles themselves fell while the share going to indirect work — discussion, policy, procedure — rose. The community writing the encyclopedia, in other words, came to spend a growing share of its effort on work other than writing.

One thing studied as a support for coordination is a state in which others’ work is visible without effort. Dourish and Bellotti showed that awareness — grasping what others are doing, passively gained from a shared workspace — supports coordination. Gutwin and Greenberg organized this into a workspace-awareness framework for synchronous groupware.

Loafing

The second is motivation loss. Karau and Williams meta-analyzed 78 studies and reported that social loafing occurs robustly, and grows larger when an individual’s contribution cannot be identified and when the task carries little meaning.

In a co-authored output, whose judgment did what is often hard to see from outside. That structure tends to satisfy the first condition — contributions that cannot be identified.

The Awkwardness of Being Evaluated

The third is the loss brought by the presence of others’ eyes. Edmondson defined psychological safety as “a shared belief held by team members that the team is safe for interpersonal risk taking,” and showed, across 51 manufacturing teams, that it is linked to performance through learning behavior.

In brainstorming research, this awkwardness is called evaluation apprehension. Geerts and colleagues confirmed that evaluation apprehension arising from social anxiety lowers brainstorming productivity. In Diehl and Stroebe’s four experiments, however, the main explanation for the loss in brainstorming groups was not evaluation apprehension but waiting one’s turn to speak. Awkwardness explains part of the loss, but is not necessarily its main cause.

Conflict

The fourth is conflict. Jehn examined 105 groups and distinguished task conflict, over the content of the task, from relationship conflict, over the relationship itself. Relationship conflict damages satisfaction and trust. Task conflict can be beneficial on non-routine tasks, but not on routine ones.

This distinction conflicts with a later meta-analysis. De Dreu and Weingart reported that both task conflict and relationship conflict correlate strongly and negatively with performance and satisfaction, and that the negative association strengthens for more complex tasks. Creative tasks are non-routine, and usually complex as well. Following Jehn’s finding, task conflict should be able to help; following the meta-analysis, it should instead do more harm. Within the literature gathered for this note, which of the two applies to creative collaboration remains unsettled.

The Effort of Reconciling Differences in Interpretation

The fifth is the effort of people who see the same thing differently reconciling their views. Bucciarelli wrote an ethnography of an engineering design setting and conceptualized object worlds — the domains of thought and artifact that participants inhabit as they work on a particular part of a design. The engineer responsible for structure and the one responsible for electrical systems look at the same product, but live in separate object worlds. That difference is what turns design into a negotiation among participants.

Research on design teams has also documented this negotiation. Valkenburg and Dorst analyzed design teams using Schön’s framework of reflective practice (naming, framing, moving, reflecting), addressing the process by which a team forms a shared frame — a way of grasping the problem. Cross and Cross conducted a protocol analysis of a three-person design team, observing, alongside roles, information sharing, problem understanding, and concept formation, occasions on which conflict was resolved and occasions on which it was avoided.

In collaborative writing, this negotiation happens at the level of individual words. The discord and frustration Farkas described at the outset concerned which word to choose and how to construct a sentence. Posner and Baecker classify collaborative writing along four dimensions: roles, activities, document control, and writing strategies.

Negotiating Names and Shares

The sixth is the negotiation that follows once results are produced. Lissoni and colleagues argued that in science teams, the allocation of author and inventor names tilts toward the discretion of senior members, giving rise to problems such as gift authorship, where a person who did not actually contribute is added as an author. Whose name appears becomes an object of coordination in its own right.

In the creative industries, this negotiation takes the form of contracts. Caves called the property of having to assemble diverse skills into a single project a motley crew, and, in the context of coordinating the timing of production, discussed the problem of hold-up — the last-minute abuse of bargaining power. No peer-reviewed research addressing how credit or shares are negotiated in artistic or design co-authorship was found in this collection, however. Lissoni and colleagues’ subject is science, and Caves’ work is a book on contract theory.

Where Does Coordination Bite into Creative Work?

All six sources exist in non-creative work too. So what is different about creative work?

One is that coordination loss is built into the very act of generating ideas. Diehl and Stroebe, in four experiments, showed that the loss in brainstorming groups is explained mainly by production blocking — forgetting an idea, or holding back from voicing it, while waiting for someone else to finish speaking. In Mullen and colleagues’ meta-analysis, face-to-face brainstorming produced lower quantity and quality than nominal groups (the pooled output of individuals working separately), and the loss grew larger with group size, with an experimenter present, and when ideas were voiced aloud. The mechanism of speech itself — only one person can talk at a time — clogs up ideas in proportion to headcount.

The other is that the object of coordination is not logistics but the content of the output itself. In routine work, what is to be made is decided in advance, and what collaborators align is who does what and when. In creative work, what is to be made gets decided in the course of making it. Farkas’s word choice, Valkenburg and Dorst’s shared frame, and Bucciarelli’s object worlds were all negotiations over the content of the output. Reconciling word choice and problem framing can be read not as effort external to the work but as occurring as part of the work itself.

Nobody knows — the property of the creative industries, listed by Caves alongside motley crew, that what will succeed and how much demand there will be cannot be known in advance — is also likely to weigh in the direction of making this negotiation heavier. If it cannot be confirmed in advance which version will sell, the means of settling disagreement over which version to adopt by an external standard become scarce. Add the property of motley crew, and a single project gathers people who hold separate object worlds.

Collaboration does not, on the other hand, produce only loss. Sawyer discusses collaborative emergence, in which ideas arise from improvisational interaction (this is a book aimed at a general readership). Empirical research on group creativity is synthesized in the edited volume by Paulus and Nijstad. The research on loss has not shown that collaboration produces no gain; it has shown where the cost paid to produce that gain arises.

Mechanisms That Make Coordination Cheaper, and What Is Lost in Return

People who keep collaborating have cultivated mechanisms that make coordination cheaper.

On the side of art, Becker, in the chapter of Art Worlds dealing with conventions, argued that conventions coordinate the collaboration of numerous participants. If the notation of a score or the logistics of a stage are shared, there is no need to align everything from zero each time.

On the side of group memory, Wegner proposed transactive memory — a mechanism for sharing who knows what, and for encoding, storing, and retrieving knowledge as a group. Not everyone needs to know everything; it is enough to know whom to ask.

On the side of collaborative editing, Kittur and Kraut showed that adding more editors raises the quality of a Wikipedia article only when it is accompanied by appropriate coordination — explicit planning, or concentrating the work among a small number of editors — and damages quality when it is not. What made a larger number of people effective was implicit coordination, not explicit coordination.

Is coordination better, then, the cheaper it gets? Uzzi and Spiro analyzed the network of producers of Broadway musicals from 1945 to 1989 and showed that the relationship between small-world-ness — the property of combining dense clustering with short paths between clusters — and box-office and critical success is an inverted U. It can be predicted that the more a network stays clustered around the same faces, the further shared assumptions advance and the less effort reconciliation takes. Even so, success declines past a certain point. Past that point, cheaper coordination and better outcomes stop moving in the same direction.

If the Partner Is AI, Does Coordination Cost Disappear?

It is tempting to think that replacing a human collaborator with AI would remove social coordination cost. AI, after all, appears not to cause awkwardness, not to loaf, and not to demand credit. The research gathered here undercuts that expectation on three counts.

First, awkwardness did not disappear. Geerts and colleagues reported that even when a robot facilitated brainstorming, the effect of evaluation apprehension lowering productivity was not mitigated. The robot in this study was a facilitator, however, not a co-author.

Second, loafing did not disappear. Elshan and colleagues reported that social loafing occurs in human-AI teams too, that how it occurs is not uniform, and that it varies with how the AI’s presence is perceived and with how much self-efficacy a person holds about their own knowledge. Siemon and colleagues report that an AI teammate’s social presence has an indirect effect on the intention to collaborate creatively with the AI, mediated by willingness to rely on the AI and by commitment to the team. Even with AI as the partner, people change their behavior according to how they feel about that partner.

Third, behind the individual’s gain, a cost arose on the side of the collection of works. Doshi and Hauser, in an experiment with 293 writers, showed that short stories written with AI-sourced ideas were rated highly for novelty and usefulness, while the works came to resemble each other more. The rise in similarity, expressed as a proportion of the range of similarity in the human-only condition, was 10.7% in the condition given one AI idea and 8.9% in the condition given five. What is a gain for each individual is a loss of diversity for the collection of works as a whole.

Singh and colleagues reviewed 62 artistic human-AI co-creative systems and reported that the higher the user’s degree of control, the higher their satisfaction, trust, and sense of ownership (a finding from the arXiv version). Even with AI as the partner, the question of who decides remains, and how that question is answered shapes the maker’s satisfaction and sense of ownership.

From the above, it cannot be read that “with AI as the partner, social coordination cost becomes zero.” Some costs — the number of communication channels, for instance — probably can shrink once the partner is no longer human. But the research gathered here has not measured how much they shrink, and for what it did measure, awkwardness and loafing remained, with a separate cost — reduced diversity — added on top.

What Remains Unknown

  • No peer-reviewed research was found addressing how credit or shares are negotiated in artistic or design co-authorship. The research on credit allocation (Lissoni et al.) covers science teams, and the discussion of contracts in the creative industries (Caves) is a book that did not go through peer review.
  • The research that counted coordination cost as a volume of work (the share of indirect work in Kittur et al.) covers collaborative editing of an encyclopedia, and no study applying the same counting method to artistic or design production sites is included among this note’s 39 references. No study measuring coordination cost and the quality of the work separately was confirmed within this scope either.
  • Whether task conflict is beneficial or harmful in creative collaboration remains contested between Jehn’s study and De Dreu and Weingart’s meta-analysis. No verification restricted to creative tasks falls within this note’s scope.
  • For Luther and Bruckman’s study of leadership in online collaborative animation production, the bibliographic record was confirmed but the full text could not be reached, and its content is not covered in this note.
  • The human-AI research gathered here places AI as a facilitator, a source of ideas, or a teammate. No study tracking coordination when AI joins a production team over an extended period is included in this collection.
  • No Japanese-language research is included among the 39 references gathered here.

References

Author: Shuichiro Ogawa (Design Researcher / Consultant) About me →