One of the most interesting concepts I learned about while studying Resilience Engineering was the distinction between Work-as-Done (WAD) and Work-as-Imagined (WAI). This concept was mentioned in all sorts of texts, but with almost no references to where it would come from. This entry is a tour of the sources I checked to try and understand the concepts better.
Introduction
The best blog post I could find on it is Steven Shorrock's The Varieties of Human Work, which further expands the definition to include:
- Work-as-Imagined: the work that we imagine others do and the work that we imagine we or others might do, currently or in the future. It is influenced by the other types of work, but never quite the same.
- Work-as-Prescribed: the formalization of the other types of work, usually in terms of laws, regulation, rules, procedures, checklists, standards, and so on. It is often assumed to be the right way to do the work, but is usually an inadequate synthesis of what actual work requires, and is therefore incomplete.
- Work-as-Disclosed: how people report, explain, or exemplify the work they do. To communicate effectively, people may simplify their descriptions or explanations. Alternatively, in less safe environments, people may willingly hide facets of their work.
- Work-as-Done: the actual activity people do. It is situated in a specific context, includes workarounds and trade-offs made to reach goals, and is hard to observe, sometimes even to the people doing the work themselves.
The post goes into far more detail and I encourage you to read it, but I wanted to provide a quick summary for folks who won't. Basically, it's useful to consider these types of work as distinct, because people who study work have to consider multiple kinds. As Shorrock says:
The analysis of work cannot be limited to work as prescribed in procedures etc (le travail prescrit), nor to the observation of work actually done (le travail réalisé). Similarly, it cannot be limited to work as we imagine it, nor work as people talk about it. Only by considering all four of these varieties of human work can we hope to understand what’s going on.
I've referred to this concept in my talk, This is all going to hell anyway, and of the need to keep that gap as narrow as possible, even if it's impossible to actually close the gap. A couple of years ago, I decided to do a deep dive into the concept, which got sidetracked because I accidentally found Dekker's Past the Edge of Chaos, one of the early papers advocating for Resilience Engineering, which also obliquely pushed for efforts in narrowing that gap between procedures and reality.
I ended up reaching out to Steven Shorrock about where the concept came from. He mentioned Hollnagel, but that it always came back to L'analyse du travail by Jean-Marie Favergé and André Ombredane. Fortunately for me, the book is in French. Unfortunately, it is from 1955, out of print, and was impossible to find outside of some archives in France. Earlier this year I revisited the topic out of curiosity, and found out a French university had actually brought back the book into ebook form in 2023. I ordered a copy, and read it as early as I could.
L'analyse du Travail
Interestingly, Ombredane and Favergé never really discuss the concepts in the terms I am familiar with. Instead, it is deeply concerned with the proper measurement of manufacturing work, and tries to draw a distinction between proper approaches to work optimization that actually look at real work as it is performed, compared to "try and see" experiments to improving performance and good old Taylorism. The closest mentions I could find to "Work-as-Done" and "Work-as-Imagined" concepts came from the introduction, and a comment that referred to the introduction. I'm loosely translating here:
In the study of the work of a depanner in a sugar refinery he did in 1952, my student André Houyoux began by measuring the inadequacy if not the uselessness of the task descriptions and written requirements related to the task that could be given to him by the study office, but also the foremen. What are the challenges of the task and why does a mistake happen? Never does any explanation given get added to the determinism of work. Differences appear in the perspicacity of foremen and in the quality of their information, but we kept hitting a trend where job requirements were expressed in terms of psychological or physiological aptitudes free of any precision: "You need skill," "you need strength," "it's not hard for a tall strong guy." But there's more: when talking to the depanners themselves, Houyoux noticed that they neither agreed on the factors making the task challenging, nor on the causes of mistakes (such as breaking a sugar pan), and that they were embarrassed when having to explain the workarounds they used. At the end of the study [...] it became obvious that none of the workers, from worst to best, had any single idea of the true factors causing friction and of the potential consequences of the operational sequences they had adopted and automated.
And another bit of the intro:
Too often, the work analyst commits two fundamental faults. He describes the task requirements in terms of attitudes and human operations such as effort, attention, judgment, initiative, etc. and, positioning himself from a normative stance, states what the worker must do or is supposed to do instead of what he really does. [...] The word do is ambiguous, expressing both the result to obtain and the series of motions by which it can be obtained [...] A term such as unstick describes an effect more so than a sequence of actions, but the formula: enact a lateral traction with the three long fingers of the left hand laid at the superior internal corner of a sugar pan that must be unstuck describes a more operational sequence.
We can see bits here that refer to a distinction between an abstract high-level definition of goals, and a more opaque set of actions to reach them. They make it shorter by calling out the observer has similar issues:
Much like the foreman, we are generally content with considering what the observed should do to reach his results more than what he actually does.
There's also a bigger quote about how asking different people gives you different information:
We end this chapter by pointing out the risks of only investigating through directors and management. This method is based on the opinion that useful information will be obtained from those who manage and structure work, and, therefore, know it perfectly. Experience shows that often, the information obtained does not match reality, and particularly don't provide what we search for.
Professor Ombredane mentioned examples of the misleading nature of information provided by managers in the book's introduction. The worker he describes at the [sheet metal] presses does not use the signals we thought he would. There are others we must discover by interacting with him and not by talking to people who aren't doing the work.
In a general sense, and this is natural, the directors will tell you how they planned the factory, the technicians will tell you what a robot they'd build would do to replace the workers, the foreman will tell you of his own work, but you will need to find by yourself what the worker does, and that's where the analysis of work begins.
As you can see, the ideas expressed there do line up in spirit with the Work-as-Imagined and Work-as-Done framework, but they are clearly not the same. We don't explicitly talk of imagination, and we don't explicitly talk of this divide between the two in quite the same way.
Back to other sources
I decided to go look at other sources, one of which was Hindsight25 (Summer 2017 issue), dedicated to the concepts of Work-as-Done and Work-as-Imagined. There are tons of really good texts there, and scouring through this document's own sources, I found more references. The earliest I could find (aside from Ombredane and Favergé) was a book by Hollnagel titled FRAM: The Functional Resonance Analysis Method (Ashgate 2012). It has been on my to-read list for a while, but a quick browsing of a copy revealed no footnotes or citations I could use for my specific deep dive here.
At that point, I decided to contact Professor Erik Hollnagel directly and he replied:
Even though WAI and WAD got a renaissance with Resilience Engineering, Dekker mentioned it in his chapter in the first RE book, the origin is, as you have noticed in French psychology and ergonomics, since French makes a clear distinction between tache (task) and activite (activity), where the Anglo-Saxon tradition focused on the task, as in task analysis, not least the static version Hierarchical task analysis. The leading person in French psychology was Jacques Leplat, no longer among us unfortunately, who wrote a number of papers (also in English) often in the excellent French journal Le travail Humain, (still active), often together with the editor-in-chief Jean-Michel Hoc (still among us, as far as I know). A good start is Leplat, J., & Hoc, J.-M. (1983). Task and activity in the psychological analysis of situations. Cahiers de Psychologie Cognitive et Cahiers de Psychologie Marseille , 3 (1), 49-63.
[...]
There is also an interesting paper by Daniellou, which I attached
The good news is that plugging these values into a search engine gave quick results: Hoc and Leplat's Tâche et Activité (1983) and Daniellou's The French-Speaking Ergonomists' Approach to Work Activity (2005), both of which I'll cover next.
Tâche et Activité
In this one, the authors spend a good bit of work making a solid distinction between Task and Activity. At a high level the distinction is that a task is what needs to be done, and activity is it happening.
Some extra elements come in play:
- A goal: the success criteria. Knowing what it is can't be sufficient; the worker will need to reframe it based on their means to attain it.
- Multiple constraints: these include state constraints (e.g., the materials available and their characteristics, minimal safe distance between aircraft), operational constraints (a machine's capacity), and procedural constraints (possibly dictated by the workplace or legal environment). These are never perfectly defined.
- A prescribed task: the task as conceptualized by whoever demands its enactment.
The prescribed task is a focal point here, and the authors state that not all descriptions are equivalent for any given worker. Some information is going to be explicit, some is going to be implicit, often depending on their training or assumed background. In some cases (e.g., scientific research), no prescription is given explicitly. In general, what is made explicit lines up with what isn't expected to be known of the worker, and what is made implicit is what is thought to be useless to mention.
If you can give a complete description, then the worker only needs to follow the instructions and only call upon previously internalized concepts. If the description is incomplete, the worker will have to fill in the gaps and devise adjustments on top of doing the prescribed steps.
By comparison, the activity is everything put together to accomplish the task. Some parts are observable, and some aren't, such as learning and problem solving. In some cases, the goals of the individual doing the activity may be different from the goals of the task, and unexpected constraints may also be added. For example, air traffic controllers could be told to only maintain safe distance between airplanes, but they might in practice also optimize their location to prevent future airspace conflict, or in order to create conditions where conflicts will be easier to solve.
This creates the need for the concept of actual task. If the prescribed task is what is expected of the worker, the actual task is describing what is actually done. When studying work, you want to study the actual goals of the actual task, and not just the prescribed ones. Some of these goals might be individual and subjective.
Do note that the actual task is a model of the activity. The activity might be to literally go from point A to point B, whereas the actual task would be the itinerary describing it. The activity will always be broader than and incompletely described by its model. The authors mention how the gap will always exist, but keeping it as narrow as possible is going to be important.
As you can see, we're now a lot closer to our modern definitions of Work-as-Done and Work-as-Imagined, with some room for Work-as-Prescribed.
The French-Speaking Ergonomists' Approach to Work Activity
This paper by Daniellou contains most of the missing answers. It was written in 2005 to try and explain to English-speaking ergonomists what the French-speaking ergonomists take for granted in their own tradition, specifically the term activity as described earlier.
While the author states that it is still unclear how the concepts actually emerged, the paper provides some missing links between the concept by providing the history of these ideas:
- Before 1975, the work of Favergé (the first book here, L'analyse du travail, being part of that) was influential, mainly with the thought of differentiating what needs to be done (task) and what workers really do—this is the discrepancy between prescribed and real work.
- Between 1972 and 1975, the concept of activity pops up, by borrowing ideas from psychology in general, but also in Soviet psychology specifically.
- Around 1977, a clarification is made between analyzing the conduct and the requirements and conditions to which the conduct is subjected
This got expanded over the following years, and the idea that one must study what difficulties the workers encounter and the adjustments they implement to deal with the variability has become fundamental. Other refinements are mentioned, and I list them here in no specific order:
- The activity has an internal representation, and it exists and lasts longer than any observation period and has some objectives that will only be met after the task is completed. The operator attributes meaning to things that are connected to various histories, which are connected, overlapping, and not necessarily aligned with the task itself.
- Mental representations do not purely reflect reality, but also functional deformations unique to each individual—some of which might be incompatible, even if many of them are in the same exact situation.
- A subjective dimension exists. Workers can develop defense mechanisms and modify their perception to protect their own mental health. Workers also have concepts of normality, have social relationships, have to resolve ethical dilemmas, and French ergonomists realized that they can't properly understand a situation while pretending "completing tasks" would be the sole objective in play.
- The rules, norms, and values in play within activities are not only those established by the organization. Activity is always social activity, within a social fabric, and the various sources of rules are often in tension.
- Activity is not only what is done, but also what the workers would have liked to do but could not.
- Ergonomists and designers are part of this system and these limitations apply themselves to this sphere as well
- You can't study what hasn't happened yet, and so design decisions have to be based on predictions and simulations. These must refer to existing situations regardless.
- Executives are to be seen as workers as well and must be taken into consideration
It's a much longer paper with a lot more details, but this felt like a good overview.
Closing Words
While I haven't exactly found a fully explicit definition of Work-as-Done and Work-as-Imagined named exactly under these terms, it's obvious that the concepts are tightly connected, and that a more detailed history seems difficult to find even for actual French ergonomists.
The good news is that there's nothing that appears to be a major misrepresentation between the original concepts and their more modern use in disciplines surrounding Resilience Engineering. What strikes me, however, is that the concepts appear to be laid out a bit differently:
- The French ergonomists appear to have them as layered abstractions and concepts: the prescribed task being more abstract than the actual task, which is a better but incomplete model of the activity.
- The Work-as-Imagined representation results in Work-as-Prescribed specifications, which get adapted in terms of Work-as-Done, which is observed as Work-as-Reported, which feeds into Work-as-Imagined.
I think these concepts are compatible; the former is based on how various representations can feed into each other, whereas the latter is a sort of simplified relationship of how information may flow into various types of work. They're both models of similar concepts, and I do appreciate having a better understanding of the underlying mechanisms and theories.
Late note: David D. Woods sent me extra context on the roots of Work-as-Done and Work-as-Imagined, which I included in a post on a paper about Resilience Engineering questioning ergonomics.