VIII. Into a Future That Talks Back
On design as a one-shot intervention into a world the intervention itself remakes.
Note: This is the eighth essay in a series. The Brief Always Arrives Wrong argued that design’s most consequential work is the constituting of problems. The Web of Issues drew on Horst Rittel to argue that designing is the traversal of a web of issues, closed by judgment rather than by criterion. The Quarrel That AI Reopened and Two Stopping Rules set Rittel against Herbert Simon over the framing of problems and the ending of work. The Resolution of the Argument examined the representations in which the arguing happens; A Sketch Is a Question followed that argument into interactive products; No Ultimate Test argued that a generated artefact’s closures are unknowable. Each of those essays was about the making of a design. This one is about what a design becomes once it is loose in the world. I thank my colleague Pramod Khambete Ph.D. who drew my attention to the eighth of Rittel and Webber’s ten characteristics of wicked problems which motivated this essay.
Every design is aimed at a future. This sounds obvious to the point of emptiness, and it is neither. A design is not a description of how things are; it is a proposal for how things should be, built and released into a world that does not yet contain it. The designer works, always, in the future tense. And the future she works in has three properties that together make design unlike almost any other kind of problem-solving, properties the whole of this series has circled without naming directly.
The first is that there is, in effect, only one shot. Rittel put it plainly: solutions to wicked problems are one-shot operations [1]. You cannot run the benefits system as an experiment, observe the results, and roll back to try a cleaner version, because by the time the results are in, the system has already processed real claims, shaped real expectations, and rearranged the institution around itself. Every implementation counts. There is no trial without error, because there is no trial that is not also, already, the thing itself.
The second is that the design, once built, becomes part of the very future it was trying to anticipate. It does not arrive into the imagined world and confirm it. It alters the world, and so the future the designer imagined and the future that actually unfolds begin, from the first day of use, to diverge. Donald Schön described the designer’s materials as talking back to her in the studio; the sketch reveals consequences she did not intend. What this essay adds is that the world talks back too, on a far longer timescale and with far higher stakes, and that the talking back is not a flaw in the process but the basic condition of designing anything at all.
The third is that the divergence is not random. It follows, in part, from the path the designers took to reach the solution; from which issues they opened, which they closed without noticing, and in what order. The world that results carries the marks of that path, materially and actively, long after the path itself is forgotten. This is where the series’ argument about closure stops being a point about verification and becomes a point about consequence.
The Effects Arrive Late
The hardest thing about designing for a future is that the future does not report back promptly, and the most important effects are the slowest to appear.
A designed artefact has effects at several removes [2]. There are the first-order effects, which are what it was designed to do, and which are usually visible at once: the form collects the data, the feed shows the posts, the button records the approval. There are second-order effects, which follow from people adapting to the first: they change what they do because the artefact is now part of their world. And there are third-order effects, which follow from institutions, markets, and whole cultures adapting to the second. Each order takes longer to appear than the one before, and each is harder to trace back to the design decision that seeded it.
The history of interface design is, read one way, a catalogue of third-order effects that passed every available test at launch. The infinite scroll solved a real and well-formed problem: pagination interrupted reading, and removing the interruption measurably improved engagement. It worked, in the tame sense the previous essay described, completely. Its second-order effect, that people found it markedly harder to stop, took a while to become legible. Its third-order effects, on attention, on sleep, on the mental health of a generation of adolescents, took years, and were not even framed as effects of a design decision until the decision was a decade old and embedded in nearly every product on the market. The like button, the autoplaying video, the streak counter, the pull-to-refresh gesture modelled on a slot machine: each was a competent solution to an immediate problem, each passed its launch metrics, and each turned out to be designing something far larger and slower than the problem it was hired to solve. None of this was visible at the stopping point. By the time it was visible, the one shot had long been fired.
This is the deep reason design problems have no immediate test, and it is worse than the previous essay let on. It is not only that we cannot verify the wicked sense of working at the moment we must stop. It is that the world needs years to compute the consequences, and there is no version of testing, however patient, that can be run before the implementation, because the consequences are produced by the implementation. The designer is required to commit, irreversibly, to a future whose most important features will not be knowable until long after committing is the only thing left to have done.
Path Dependency, and Why the Corpus Deepens It
Now add the path. No deliberation is neutral with respect to its own history. The order in which a team opens its issues shapes which further issues it is able to see; an early framing closes off questions that a different early framing would have raised, and those questions do not come back, because nobody knows they were foreclosed. This is path dependency, and it is present in all design, human or otherwise [3]. The solution carries the shape of the road taken to it.
Generative tools do not remove this dependency. They deepen it, and they hide it. A model is trained on the accumulated output of past design, which is to say on the congealed residue of millions of paths that real teams once walked, each with its own foreclosures, its own unexamined closures, its own historically situated sense of what was obvious and what was unthinkable. When the model returns a solution, it returns the statistical sediment of all those paths at once, and it returns it as though it were the natural shape of the answer rather than the deposit of a particular history. The previous essay called the result unknowably well-formed, and argued that we cannot recover why the artefact is as it is. The point here is sharper and more material. The path the model encodes is not merely unknown. It is active. The foreclosures of a thousand past teams, the questions they never thought to ask, the populations their corpus underrepresented, the assumptions that were ordinary in the training data and are catastrophic in this new context, are all carried forward into the future the new design will help create, as live causal traces, working away beneath a surface that betrays no sign of them.
So the design that goes out into the world to remake it carries a path nobody walked, toward foreclosures nobody chose, with effects that will not surface for years. Each of the three properties of designing for a future, the single shot, the reflexive remaking, the late arrival of consequence, is intensified by generation, and all three are intensified invisibly.
The design that goes out into the world to remake it carries a path nobody walked, toward foreclosures nobody chose, with effects that will not surface for years.
What Designing for a Future Asks
If this is what design is, then a great deal of design practice and almost all of design education are training people for the wrong half of the job. We teach the making of the artefact, and the immediate evaluation of it, with real rigour. We barely teach the one capacity this essay has been describing as central: the disciplined anticipation of an artefact’s second and third order effects, the deliberate attempt, before the one shot is fired, to imagine how a particular future world will react to the thing being placed into it.
This is not forecasting, and it cannot be made into a method, for the same reason nothing else in this series could. The future a design will help create is wicked; it has no definitive formulation and no test in advance. But the impossibility of certainty is not permission to skip the attempt. A profession that took its temporal condition seriously would treat second and third order anticipation as a named and examinable skill, alongside research and critique. It would ask of every significant closure not only is this right now, but what does this make easier to do, and harder, once it is in the world; who adapts to it, and how; what does it become a part of. It would treat the cases of catastrophic delayed effect not as accidents but as the predictable result of stopping at the first-order test. And it would hold the designer answerable not only for the closure, but for its propagation, to whatever degree a serious imagination could have reached.
That word, answerable, is where this essay arrives, and it is the one the series has been moving toward from the start. The first essay argued that design cannot be displaced by machines because its real work is the constituting of problems. The intervening essays followed that work through argument, through stopping, through representation, through the unverifiable closure. This essay names the stakes the work was always carrying. A design is a one-shot, irreversible intervention into a future that the intervention itself remakes, after a delay long enough to sever every effect from its cause, along a path that the new tools render at once deeper and invisible. Nothing in that can be tested in advance, automated away, or recovered after the fact. What can be done is what could always be done, and what the machines cannot do at all: to imagine, as honestly and as far as a human mind can reach, the world one’s decision is about to help make, and to accept responsibility for having made it. The future talks back. The only question is whether anyone was answerable for what it was given to say. The final essay asks what makes a person willing to be.
Notes
[1] The one-shot character of wicked-problem solutions is the eighth of Rittel and Webber’s ten characteristics in “Dilemmas in a General Theory of Planning” (Policy Sciences, 1973): every implemented solution has consequences that cannot be undone, so there is no opportunity to learn by trial and error, and every trial counts. The related claim that every wicked problem is essentially unique, and can be regarded as a symptom of another problem, is theirs too. The reading offered here, that the implemented solution becomes part of the future it was attempting to anticipate, extends their framework toward what would later be discussed under the heading of performativity: the design does not describe a future state of the world but participates in bringing one about.
[2] The orders-of-effect language adapts a long tradition in systems thinking and technology assessment concerning first, second, and higher-order consequences, and the characteristic lag with which the higher orders appear. The interface examples are deliberately familiar; the point is not that any individual feature was malign, but that each was a competent, well-tested solution to a well-formed immediate problem whose larger consequences were structurally invisible at the moment of design, and became legible only after the intervention had already reshaped the behaviour it was responding to. Donald Schön’s account of the situation talking back is in The Reflective Practitioner (1983); the move made here is to relocate the talkback from the studio, where its timescale is minutes, to the world, where its timescale is years and its results cannot be sketched over.
[3] Path dependency is borrowed from economics and the history of technology, where it names the way early and often contingent choices constrain the space of later ones, sometimes long after the reasons for the original choice have lapsed. Its application to the internal deliberation of a design team, the way an early framing forecloses issues that are never subsequently seen, is the bridge back to this series’ second essay and its account of the web of issues; a foreclosed issue is one the web never opened, and on which, therefore, no judgment was ever owned. The claim that a trained model encodes the congealed path dependencies of its training corpus, and reactivates them as live causal traces in new contexts, is the temporal counterpart to the argument in No Ultimate Test: the earlier essay held that the rationale cannot be known, this one that the path is nonetheless still at work.
Essays In The Series
Index of essays on design practice in the age of generative AI