The Way Out Must Remain Open

Kant (1784) defines enlightenment as ‘the human being’s emergence from self-incurred immaturity’ (AA 8:35). This immaturity is self-incurred when its cause lies not in a lack of understanding but in a lack of courage to use one’s own understanding without another’s guidance. And it is convenient: ‘If I have a book that has understanding for me, […] I need not make any effort myself’ (AA 8:35). Kant had books, spiritual advisers and doctors in mind. Today, the sentence can be applied almost word for word to systems that compose texts, help form judgements and propose decisions.

Yet Kant’s text contains a second demand, one that carries still greater weight for technological orders. One age, he writes, cannot ‘join together and conspire to place the next in a condition in which it must be impossible for it to expand its […] knowledge and purge it of errors’ (AA 8:39). His subject is binding religious doctrines. ‘Statutes and formulas, these mechanical tools for the rational use, or rather misuse, of one’s natural gifts, are the fetters of perpetual immaturity’ (AA 8:36). Transposed to technological orders, the question becomes: what commitments may the present make without depriving the future of the possibility of correction?

This is where the essay begins. Even someone who approaches a technology with understanding and courage may be unable to refuse its use when the conditions of participation require it. What is lacking is not necessarily independent judgement, but the possibility of acting on it. The way out remains open only if a society can still revise its own commitments – if later, better-grounded judgement can still alter the outcome.

The argument proceeds through conceptual analysis and normative reasoning rather than empirical testing or a systematic review. It brings selected accounts of expectations, path dependence and technological control into relation with self-determination understood as continuing agency. The hypothetical AI examples clarify a possible mechanism; the archaeological and human–automation examples illustrate the distinction between adopting a technology and preserving practical alternatives. These examples do not establish the prevalence of collective anticipatory self-binding. The assessment concerns the conditions under which later judgement can remain effective, not a general verdict on AI adoption.

This paper examines a specific mechanism of collective anticipatory self-binding: adaptation to expected future conditions of participation. Its normative question is whether the self-determination of those affected is compromised when later, better-grounded judgements can no longer correct a technological order in practice. It argues that such self-binding becomes problematic when it deprives later judgement of practical effect without being justified by a substantial shared purpose and the absence of less constraining alternatives. The sociology of technological expectations shows how promises about the future mobilise attention and resources; research on path dependence explains how decisions reinforce themselves and restrict later choices (Borup et al., 2006; Sydow et al., 2009). Collingridge (1980) describes the temporal tension between growing knowledge and declining capacity for change. Revisability supplies the normative link between these accounts: it asks whether later learning can still alter commitments made in response to present expectations and reinforced over time.

‘There is no way around artificial intelligence.’ This formula condenses a claim that goes beyond the performance of individual tools. Companies are expected to prepare their employees for the new era, universities to teach the competencies expected in the future, and applicants to demonstrate that they can meet those expectations. Even someone who takes a sober view of the present benefits may feel compelled to learn, invest and reorganise. What matters is not only present performance, but the expectation of no longer being able to participate without this technology in the future.

Preparing for new tools can be reasonable. Their performance can be investigated empirically for particular tasks; such evidence does not, however, establish a general obligation to organise all work around them. Shared infrastructure can likewise enable achievements that would be unattainable in isolation: with network externalities, the value of use increases precisely with the number of compatible users (Katz & Shapiro, 1985). The question begins not with a denial of these benefits, but at the point where a limited ‘can’ becomes a general ‘must’.

Even so, it is worth pausing over the sentence. Who exactly has established that there is no way around this technology? Does the statement describe a future that will come about independently of our decisions? Or is the statement itself among the forces that bring that future into being?

What we commonly call hype may be the visible expression of collective anticipatory self-binding. Because we expect a particular future, we invest in its tools, learn to meet its requirements and build its infrastructure. To the extent that these decisions make later alternatives more difficult, we do not merely predict the future: we bind ourselves to it and make other futures less likely. Here, ‘hype’ denotes the public concentration of attention around promises about the future whose claimed scope extends beyond applications already demonstrated. Such promises need neither be false nor follow a uniform course (van Lente et al., 2013).

Merton (1948) coined the term ‘self-fulfilling prophecy’ for related processes: an initially false definition of a situation elicits behaviour that makes it true. The case examined here differs in two respects. The initial expectation need not be false. And the issue is not only whether an expectation is fulfilled, but also the commitments it leaves behind – even when the expectation itself has long since faded.

Collective anticipatory self-binding denotes a process in which multiple actors orient their decisions towards an expected future and thereby generate dependencies that make later alternatives more difficult to access. The aggregate effect need be neither centrally decided nor intended by all participants. ‘Self-binding’ refers to the collective production of these dependencies, not to the voluntary consent of everyone affected. Merely experimenting or acquiring additional skills is not enough; what matters is the restriction of later choices. As an analytical working concept, the term connects the sociology of expectations with research on self-reinforcing paths (Arthur, 1989; Borup et al., 2006; Sydow et al., 2009).

The concept must be distinguished from Elster’s (1979) account of self-binding. Odysseus has himself bound to the mast because he knows his future weakness; the binding is consciously chosen and intended to protect him from himself. In collective anticipatory self-binding, by contrast, the later restriction of opportunities for correction need be neither intended nor jointly decided. Even consciously coordinated introductions of a technology can produce such consequences and bind people who have not consented to them. Whether the binding protects or fetters often becomes apparent only when correction is needed.

The mechanism examined here is narrower than forward-looking adaptation or investment in general. Its distinguishing feature is that anticipated conditions of participation become a reason for commitments that restrict later alternatives. Expected task benefits and coordination gains may coexist with this motive. The analytical question is which commitments actors would still make on the basis of those benefits, and which additional commitments they make because they expect others to require the technology as a condition of participation. If demonstrated task benefits and coordination requirements adequately explain the commitments, this mechanism supplies no additional explanation of that case. Identifying it in a concrete case would therefore require distinguishing these reasons and tracing how commitments attributable to anticipated participation requirements generate dependencies. The hypothetical chain below establishes neither the relative weight of these reasons in actual AI adoption nor the prevalence of the mechanism; it shows how the mechanism could operate.

An illustrative model shows how this feedback may operate. Imagine an applicant who highlights experience with generative AI on her CV, not necessarily because it has improved her work to date, but because she assumes that employers are looking for this competence. A company, in turn, lists this competence as a requirement in a job advertisement because customers and investors expect a business equipped for the future to use AI. Investors monitor AI adoption because other companies are investing and because they do not want to overlook what may prove a decisive development. Universities respond to the anticipated demand by introducing relevant courses.

Every decision in this chain can be justified. No one need act foolishly or dishonestly. Yet an unsettling question remains: where does the initial certainty come from? The applicant points to companies, companies to the market, the market to investments, and investors to the general expectation. Everyone observes signals that arose because others were observing the same signals.

Economists describe a related process as an information cascade. The decisions of earlier actors can reasonably serve as evidence of information one does not possess oneself. In the narrower cascade model, these decisions eventually become so decisive that private information no longer changes the decision. Visible agreement can thus arise whose informational content is lower than it appears (Bikhchandani et al., 1992). I use the term ‘expectation cascade’ here to refer to a broader process: actors increasingly derive their expectations of the future from the expectations and expectation-based decisions of others; their own decisions, in turn, generate apparently confirming signals. This also includes explicitly articulated promises and visions of the future (Borup et al., 2006). The crucial feature is the feedback: dependent signals can appear to be independent confirmations.

An experiment with an artificial music market made this effect visible. When participants could see which unfamiliar songs others had already chosen, the outcomes became more unequal and harder to predict. Quality remained important at the extremes; between them, success depended heavily on which early decisions prompted further ones (Salganik et al., 2006).

The process is more complicated in technological developments. A machine has testable properties, and its adoption may rest on good reasons. Yet here too, visible approval is not automatically the sum of many independent assessments. It may partly be the echo of a shared expectation.

At first, we are told that a tool can do something. Soon, we are told that many people will use it. Finally, we are told that we must use it. These three statements sound like stages of the same insight. In fact, they belong to different levels.

The word ‘can’ describes a capability. A system can recognise certain patterns or produce a draft more quickly. The word ‘will’ expresses an expectation about the behaviour of others. The word ‘must’, by contrast, can assert an obligation or describe practical pressure to adapt. The inevitability claim blurs this distinction: it treats the demand for adaptation as an already established necessity. Yet neither widespread adoption nor the need to organise all work around a tool follows from its demonstrated capability.

The shift can easily go unnoticed: if many people will use a technology, surely one must prepare. Yet those who prepare themselves contribute to its adoption. Promises about the future are therefore not regarded as mere descriptions; they mobilise resources and can help shape developments whose benefits are not yet established (Borup et al., 2006; Brown & Michael, 2003).

In this process, the question of a specific purpose can lose its power to guide decisions. A tool is no longer introduced solely because it performs a task better, but because its use is supposed to signal readiness for the future. Competence and belonging become tied to participation. Asking for a demonstrable benefit then appears not as part of the assessment, but as a sign that someone has not yet understood the direction of development. Not every instance of hype makes this shift from expectation to obligation. The essay examines the critical case in which public expectation turns into practical pressure to adapt.

An appeal to think more independently falls short here. No one checks every piece of information personally; knowledge and the division of labour presuppose justified trust in others. Nor is conspicuous opposition to the majority a form of independence. It takes the majority’s judgement as its reference point and merely reverses its sign. Judgement requires something else: distinguishing benefits, evidence supplied by others and social pressure to adapt.

Independent thinking begins not with thinking differently, but with recognising where one’s reasons come from.

In a technological decision, it makes a difference whether I use a tool because it solves my particular problem better, because knowledgeable people have demonstrated its performance in a way I can assess, or because I expect others soon to require me to use it. All three reasons can justify a decision. The justification becomes misleading when expected pressure to adapt is silently treated as evidence of concrete capability.

A simple thought test can make this shift visible. Would I also use the tool if no one knew about it? Would I make the same decision if I did not know how many others had already adopted it? What specific problem does it solve better than the available alternatives? And what skill or alternative do I lose if I rely on it permanently?

This thought test separates use value from social significance, but does not capture every form of benefit. A shared standard may be useful precisely because others use it (Katz & Shapiro, 1985). The relevant assessment then becomes: what coordination gains do we achieve, what dependencies arise, and does correction remain possible? The number of participants is neither proof of quality nor a reason to deny the benefits of shared procedures.

An individual may recognise these distinctions and nevertheless have to follow the prevailing path. Applicants cannot eliminate labour-market requirements through reflection alone; companies cannot ignore widespread interfaces without consequences. At this point, the individual appeal to courage is no longer enough.

A person may judge that using a technology is unwarranted yet find themselves compelled to use it by the conditions of participation. The problem is not an inability to judge independently. For Kant, immaturity denotes the inability to use one’s understanding without another’s guidance; the actor at issue here retains that ability. The essay calls this condition ‘the structural disempowerment of autonomous judgement’: the capacity to judge remains intact, but the ability to act effectively on that judgement is restricted. This is not another form of Kantian immaturity. It is a socially produced limit on an existing capacity for independent judgement. Courage remains necessary, but is no longer sufficient.

The decisive issue is not whether every individual judgement is implemented, but whether those affected can bring their judgements into a process of collective reassessment capable of changing the arrangements that bind them. Dependencies arising from the division of labour, shared standards or switching costs do not therefore automatically disempower autonomous judgement. Collective decisions may override well-grounded individual judgements without abolishing self-determination. Judgement becomes socially effective when it can inform the assessment and revision of collective decisions, even if it does not prevail. The critical case arises when even procedures for collective correction no longer offer a realistic possibility of revising those arrangements.

The conditions of such orientation can be clarified through social learning. Individual experimentation takes time; others’ experience opens up knowledge without requiring us to repeat every error ourselves. An experiment with human participants shows that difficult tasks and larger groups can encourage reliance on the majority (Toyokawa et al., 2019). Chimpanzees, too, preferentially copied the solution of an older, higher-ranking group member who had previously been successful (Horner et al., 2010). How informative such signals are, and who can shape them, helps determine whether social orientation provides access to knowledge or primarily reinforces dependency.

Alongside observing others, people orient themselves towards imagined futures. They can use language to envisage a majority that does not yet exist and organise their present around it. The word ‘inevitable’ provides a common path and interprets hesitation as a lack of preparation. Uncertainty about the future becomes an injunction for the present.

This movement requires no central plan. Yet not all participants shape the emerging path equally. Those who control capital, platforms, standards or access to labour markets can more readily turn their own expectations of the future into requirements for others. Jasanoff (2015, p. 4) describes influential visions of the future as sociotechnical imaginaries: collectively held, institutionally stabilised and publicly staged visions of desirable futures. Institutional stabilisation presupposes access to institutions. Some formulate what will supposedly be inevitable; others must prepare for it. Self-binding is collectively produced, but the power to shape the path and the costs of adaptation are unequally distributed. Sincere convictions and cautious adaptation can reinforce the same path. Participants explain to one another why a development will become inevitable, thereby changing the conditions under which it could be avoided.

A narrative about the future takes material form when education, workflows or infrastructure are reorganised on its basis (Borup et al., 2006; Brown & Michael, 2003). Each complementary investment then improves the conditions for the chosen path. The binding grows not only through agreement, but through practical additions to decisions already made.

At the same time, alternatives can lose ground. When resources, experience and complementary provision are concentrated on one path, learning, coordination and complementarity effects favour further decisions in its support; the scope for action can narrow to the point of lock-in (Arthur, 1989; Sydow et al., 2009). The dismantling of other training opportunities, equipment or knowledge would further entrench the path, but is not an inevitable accompaniment to every introduction of a technology. The later superiority of a path can thus arise partly from the conditions created by its adoption.

In such developments, increasing returns make later switching progressively more costly. Switching requires more than finding a better tool: actors must also replace, convert or work around established expertise, contracts, data and infrastructure (Arthur, 1989; Sydow et al., 2009). Path dependence provides the broader theoretical framework for explaining how earlier decisions can activate self-reinforcing mechanisms that constrain later choices. Lock-in denotes a particularly restrictive condition within such a process, in which viable alternatives have become extremely difficult to pursue. Within this framework, this essay uses ‘expectation-driven path entrenchment’ to denote a specific form of path-dependent consolidation in which expectations about the future have substantially contributed to the mechanisms reinforcing the chosen path. Switching costs alone do not amount to path entrenchment; what matters is whether these mechanisms progressively narrow the scope for action.

The concepts introduced here distinguish the process, its possible influences and outcomes, and its normative assessment. Collective anticipatory self-binding is the central process; expectation-driven path entrenchment is one possible outcome. Expectation cascades and hype may amplify that process, while an inevitability claim may frame it discursively. Neither these amplifying factors nor this framing is necessary for self-binding, and the relationships imply no fixed sequence. Self-binding can arise when expectations remain cautious and explicitly uncertain; expectations may fade without producing lasting entrenchment; and investments may constrain later choices before a technology attracts widespread attention. Revisability is not another stage in this process, but the normative standard for assessing it and its consequences. The structural disempowerment of autonomous judgement is one possible consequence of losing revisability.

The inevitability claim can reinforce this configuration at different points. Initially, it justifies commitments by invoking a supposedly inescapable future; later, the dependencies that have arisen appear to confirm that claim. Even those who doubt the necessity of the path can thus help consolidate it. The claim that there is no alternative must be distinguished from the actual restriction of alternatives. The fact that a path has become difficult to leave does not prove that there was no alternative to choosing it.

Hype can fade faster than the dependencies it has helped create.

A technology’s later success demonstrates its performance under the conditions that have emerged; it does not prove that these were the only possible conditions. Nor is every instance of path dependence a mistake. Shared standards enable cooperation and save effort. In their normative assessment, however, the possibility of review must be distinguished from revisability: a mistaken decision can be recognised yet remain practically almost impossible to correct. What matters is whether better knowledge still enables a viable change of course.

Collingridge (1980, pp. 11, 20–21) described the temporal tension in this problem: in early stages of development, a technology is still comparatively easy to change, while its social consequences are little understood. By the time these consequences become clearer, the technology is often so deeply embedded in routines and infrastructure that change has become difficult and expensive. His response centres on preserving opportunities for correction. Revisability is the broader concept here: later knowledge must still be able to bring about relevant changes in practice. Reversibility denotes the more specific possibility of undoing a commitment or returning to an earlier practice. It is one means of securing revisability, not its only form. The mechanism examined here complements Collingridge’s account by explaining how declining capacity for change can already be produced through adaptation to expected future conditions of participation. Normatively, what is at stake is therefore not only the governability of technological orders, but also the possibility of giving a later, better-grounded judgement practical efficacy.

The distinctive contribution lies not in introducing revisability as a general requirement, but in explaining how adaptation to anticipated conditions of participation can undermine it and thereby restrict the practical efficacy of autonomous judgement. This connects the essay with responsible innovation: Stilgoe et al. (2013) bring together anticipation, reflexivity, inclusion and responsiveness; Fisher et al. (2024) review the breadth of normative, empirical and theoretical scholarship; and the OECD (2025) emphasises adaptive, anticipatory governance for AI. Genus and Stirling (2018) also show why engagement with Collingridge must extend beyond the dilemma of control to the wider political context, including anticipatory, participatory and institutional aspects of governance. The essay therefore claims no new general standard for responsible technological development. It examines how preparation for an expected future can erode the practical conditions of responsiveness: not merely whether consequences become better understood, but whether that learning can still change the order that has been created.

A simplified narrative of the transition to agriculture shows how readily a development can appear definitive: hunting and gathering, agriculture, sedentary life, cities. In this sequence, each stage seems to be a necessary precondition for the next. Once people cultivate plants, there appears to be no reasonable way back. It is precisely this assumed irreversibility that can be tested against the archaeological evidence.

Research on the Fremont complex in present-day Utah and neighbouring areas presents a more fluid picture. The name does not designate a single Indigenous nation, but an archaeological umbrella category covering different communities. Madsen and Simms (1998) describe a mosaic of farmers, hunter-gatherers and groups combining different subsistence practices. Some switched seasonally; others intensified maize cultivation and later returned to greater reliance on wild foods. Isotope analyses from the Great Salt Lake region likewise indicate changing proportions of maize and wild foods; the decline in cultivation correlates with changes in regional moisture conditions (Coltrain & Leavitt, 2002). In eastern Utah, Finley et al. (2020) associate its intensification with multidecadal climate variability.

These findings challenge the assumption that development necessarily proceeds in a permanently fixed direction. They demonstrate neither a prehistoric expectation cascade nor deliberately institutionalised revisability; migration, climate and social change are part of the explanation. They do show, however, that using a fundamental technology does not fix its enduring share in a way of life. A narrative of progress that reads every reduction in that share as regression obscures the possibility of reasonable adaptation.

Agriculture could form part of a changing repertoire. This does not establish the general superiority of less extensive commitments, but raises a question about the conditions of adaptation: what combination of skills enables a community to respond to changing circumstances? A relevant alternative practice does not remain available merely because people know that it once existed. Knowledge, tools and social structures must make its resumption possible.

Human–automation interaction illustrates how an alternative can remain known yet cease to be practicable. Automation may leave humans responsible for intervening when something goes wrong, while normal operation gives them little opportunity to practise the skills those interventions require; those skills may then deteriorate (Bainbridge, 1983). Tool use does not inevitably produce such loss; it can also preserve existing skills or foster new ones. But if people no longer practise carrying out the work themselves and organisations no longer maintain the capacity to do so, a decision to dispense with the tool may be impossible to implement in the short term. Kant’s book ‘that has understanding for me’ then becomes infrastructure. What began as relief from effort can become an entrenched dependency that courage alone cannot overcome. The alternative is not prohibited; it has become practically unavailable.

In this sense, the way out must remain open: not as an ever-available return to an unchanged past, but as a real possibility of responding to new knowledge by acting differently. An alternative remains practically available to the extent that the necessary skills and conditions can be preserved or restored at reasonable cost.

The claim that there is no way around artificial intelligence can now be questioned more precisely. For which tasks does it hold? On what evidence of performance does it rest? What shared procedures does cooperation actually require? And how much of the alleged compulsion arises only through the decisions justified by invoking it? The inevitability claim bypasses these questions. It displaces concrete justification with belonging to an expected future.

Collective action requires commitments. Infrastructure requires investment, the division of labour depends on reliable interfaces, and education entails choices about which skills to teach. Avoiding every binding commitment would also forfeit opportunities for action that arise only through reliable cooperation. The relevant distinction is between making a binding decision and creating dependencies that make later alternatives harder to pursue. A commitment becomes self-binding to the extent that it creates such dependencies. The response to claims of inevitability is therefore not to refrain from making decisions. It is to create an order in which decisions are justified and effective possibilities for their later correction are preserved.

The normative concern is not that a legitimate decision ceases to count as an exercise of self-determination merely because it creates lasting dependencies. It is that self-determination is not exhausted by a single moment of authorisation. Decisions are made under limited knowledge, remain in force as evidence and circumstances change, and may bind people who did not participate in their authorisation. Self-determination therefore concerns not only the authority to decide now, but also the effective agency of those who remain subject to the decision. If their later, better-grounded judgements cannot enter a collective process capable of changing the arrangements, the original exercise of agency restricts the continuing agency of those affected. Revisability protects that continuing agency; it does not give each person a veto or require every decision to remain reversible. Its loss therefore creates a burden of justification rather than an unconditional prohibition.

Five dimensions of a revisability assessment follow from this relationship. The scope of a commitment determines whose possibilities for action are affected, how deeply and for how long. Uncertainty about its premises determines the significance of possible later knowledge; the severity of possible adverse consequences determines the weight of a necessary correction. Substitutability asks whether the same purpose can be achieved through less binding procedures. Finally, the power to secure correction determines whether a reassessment can have practical consequences: who can demand a revision, who can implement it, and who bears its costs? These dimensions thus connect the emergence of technological binding with its justification, without claiming to offer a complete catalogue of criteria for political legitimacy.

Consider a hypothetical organisation that requires the use of an AI system across its workflows. A revisability assessment would ask whose work and ability to participate are affected, how deeply and for how long, how uncertain the assumptions underlying the expected benefits remain, and how serious plausible failures could be. It would then ask whether less constraining arrangements could serve the same basic purpose, who has the authority to demand a change of course, and who has the practical capacity to implement it. The distinction between review and effective revision becomes visible if a later assessment finds the system unsuitable for some tasks. An organisation that has preserved transferable knowledge, the skills needed for alternative practices and accessible procedures for reassessment may be able to limit, modify or discontinue its use. An organisation in which those capabilities and procedures have eroded may accept the same finding without being able to act on it. The example does not show that mandatory use is always unjustified; it shows why the formal possibility of reviewing a decision is not yet a safeguard for effective correction.

The level at which correction must remain possible depends on the reach of the commitment. The organisational example illustrates practical capacity for revision; it does not by itself show that those affected are adequately represented in deciding whether revision occurs. Where a technological order binds people beyond an organisation, internal authority to change workflows cannot substitute for collective reassessment open to the reasons and interests of those affected. Conversely, the absence of an individual exit does not by itself establish the absence of societal revisability. At each level, the relevant question is whether those subject to the arrangements can bring reasons into a process with both the authority and practical capacity to change them.

The scope for effective correction must be safeguarded more strongly as a commitment becomes more far-reaching and lasting, its premises more uncertain and its possible adverse consequences more severe. Forgoing such safeguards requires a substantial shared purpose—a purpose that can be publicly justified to those affected—for which less constraining procedures are insufficient, together with a decision reached through a legitimate process. These are necessary, not sufficient, conditions: the relationship between shared benefits, burdens and lost opportunities for correction must also be assessed. Irreversible commitments are not ruled out in principle, but their present benefits alone cannot justify the loss of future power to secure correction.

Preserving one possibility of correction may consume resources needed to sustain other possibilities for action. Revisability therefore does not require maximum openness considered in isolation. An alternative is relevant if it can fulfil the same basic purpose differently or respond to a plausible failure of the chosen path. Parallel procedures, transferable knowledge and preserved alternatives require resources; how much effort it is reasonable to devote to preserving them depends on the uncertainty, scope, duration and possible severity of harm associated with the commitment. The assessment must compare the shared purpose secured by a commitment, the plausible need and importance of later correction, the burdens of preserving relevant alternatives and the opportunities forgone by doing so. A less revisable arrangement may be justified, but that judgement must account for whose agency it enables and whose future power to secure correction it restricts. Neither preservation costs nor present benefits settle the question on their own. High switching costs that have already arisen must not themselves become a sufficient reason to exclude an alternative; otherwise, the commitment would justify its own uncorrectability.

Under uncertainty, real options theory analyses not only waiting before irreversible investments, but also possibilities for suspension, resumption and abandonment, as well as hysteresis between entry and exit (Dixit & Pindyck, 1994, ch. 7, pp. 213–244; summarised on pp. 9–11). For collective technological orders, this economic perspective alone is insufficient. What also matters here is whether skills, infrastructure and institutional alternatives remain available. An individual exit option is not enough if shared infrastructure, education or the wider institutional order continues to bind those affected.

In practice, staged implementation subject to review, transferable knowledge and procedures that allow switching without forfeiting all previous work are means of safeguarding revisability. Which combination is appropriate depends on the commitment and the proportional assessment described above; no single instrument is universally required. Accessible reassessment procedures must admit new evidence, changed circumstances and reasoned dissent without requiring a later judgement to be recognised as better grounded before it can be considered. Whether that judgement warrants correction is a question for the reassessment itself. Not every minority position needs to prompt an entirely new decision, but such positions must receive reasoned consideration and must not be excluded by access requirements that cannot realistically be met. Decisions about correction and the distribution of its costs must not rest solely with those benefiting from immediate coordination gains; they must also take account of those bound without having participated in the original decision and those who would bear the costs of correction. Revisability thus requires a realistic possibility that dissent can change the order itself, not merely permission to dissent or adjustments within the established path.

Kant called it a ‘crime against human nature’ (AA 8:39) for one age to prevent the next from correcting errors. Applied to technological orders, this demand calls not for abstention from technology, but for arrangements designed to preserve the possibility of correction. Enlightenment under these conditions requires more than the courage to use one’s own understanding; it also requires preserving the conditions under which that understanding can still make a difference. The present may act without claiming final authority over the conditions of future action. A self-determining society is neither one that avoids every commitment nor one that treats its chosen path as destiny. It is one that can make commitments while preserving relevant possibilities for revision. The way out must remain open – and later knowledge must still be able to alter the outcome.

References

Arthur, W. B. (1989). Competing technologies, increasing returns, and lock-in by historical events. The Economic Journal, 99(394), 116–131. https://doi.org/10.2307/2234208

Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775–779. https://doi.org/10.1016/0005-1098(83)90046-8

Bikhchandani, S., Hirshleifer, D., & Welch, I. (1992). A theory of fads, fashion, custom, and cultural change as informational cascades. Journal of Political Economy, 100(5), 992–1026. https://doi.org/10.1086/261849

Borup, M., Brown, N., Konrad, K., & van Lente, H. (2006). The sociology of expectations in science and technology. Technology Analysis & Strategic Management, 18(3–4), 285–298. https://doi.org/10.1080/09537320600777002

Brown, N., & Michael, M. (2003). A sociology of expectations: Retrospecting prospects and prospecting retrospects. Technology Analysis & Strategic Management, 15(1), 3–18. https://doi.org/10.1080/0953732032000046024

Collingridge, D. (1980). The social control of technology. Frances Pinter.

Coltrain, J. B., & Leavitt, S. W. (2002). Climate and diet in Fremont prehistory: Economic variability and abandonment of maize agriculture in the Great Salt Lake Basin. American Antiquity, 67(3), 453–485. https://doi.org/10.2307/1593822

Dixit, A. K., & Pindyck, R. S. (1994). Investment under uncertainty. Princeton University Press.

Elster, J. (1979). Ulysses and the sirens: Studies in rationality and irrationality. Cambridge University Press.

Finley, J. B., Robinson, E., DeRose, R. J., & Hora, E. (2020). Multidecadal climate variability and the florescence of Fremont societies in eastern Utah. American Antiquity, 85(1), 93–112. https://doi.org/10.1017/aaq.2019.79

Fisher, E., Smolka, M., Owen, R., Pansera, M., Guston, D. H., Grunwald, A., Nelson, J. P., Raman, S., Neudert, P., Flipse, S. M., & Ribeiro, B. (2024). Responsible innovation scholarship: Normative, empirical, theoretical, and engaged. Journal of Responsible Innovation, 11(1), Article 2309060. https://doi.org/10.1080/23299460.2024.2309060

Genus, A., & Stirling, A. (2018). Collingridge and the dilemma of control: Towards responsible and accountable innovation. Research Policy, 47(1), 61–69. https://doi.org/10.1016/j.respol.2017.09.012

Horner, V., Proctor, D., Bonnie, K. E., Whiten, A., & de Waal, F. B. M. (2010). Prestige affects cultural learning in chimpanzees. PLOS ONE, 5(5), Article e10625. https://doi.org/10.1371/journal.pone.0010625

Jasanoff, S. (2015). Future imperfect: Science, technology, and the imaginations of modernity. In S. Jasanoff & S.-H. Kim (Eds.), Dreamscapes of modernity: Sociotechnical imaginaries and the fabrication of power (pp. 1–33). University of Chicago Press. https://doi.org/10.7208/chicago/9780226276663.003.0001

Kant, I. (1784). Beantwortung der Frage: Was ist Aufklärung? Berlinische Monatsschrift, 4, 481–494. Cited from: Kants gesammelte Schriften (Akademie-Ausgabe, edited by the Königlich Preußische Akademie der Wissenschaften), vol. VIII (pp. 33–42). Georg Reimer, 1912. English translations of Kant’s quotations have been prepared for this essay from the German text; references identify the Akademie-Ausgabe volume and page.

Katz, M. L., & Shapiro, C. (1985). Network externalities, competition, and compatibility. The American Economic Review, 75(3), 424–440.

Madsen, D. B., & Simms, S. R. (1998). The Fremont complex: A behavioral perspective. Journal of World Prehistory, 12(3), 255–336. https://doi.org/10.1023/A:1022322619699

Merton, R. K. (1948). The self-fulfilling prophecy. The Antioch Review, 8(2), 193–210. https://www.jstor.org/stable/4609267

OECD. (2025). Steering AI’s future: Strategies for anticipatory governance (OECD Artificial Intelligence Papers, No. 32). OECD Publishing. https://doi.org/10.1787/5480ff0a-en

Salganik, M. J., Dodds, P. S., & Watts, D. J. (2006). Experimental study of inequality and unpredictability in an artificial cultural market. Science, 311(5762), 854–856. https://doi.org/10.1126/science.1121066

Stilgoe, J., Owen, R., & Macnaghten, P. (2013). Developing a framework for responsible innovation. Research Policy, 42(9), 1568–1580. https://doi.org/10.1016/j.respol.2013.05.008

Sydow, J., Schreyögg, G., & Koch, J. (2009). Organizational path dependence: Opening the black box. Academy of Management Review, 34(4), 689–709. https://doi.org/10.5465/amr.2009.44885978

Toyokawa, W., Whalen, A., & Laland, K. N. (2019). Social learning strategies regulate the wisdom and madness of interactive crowds. Nature Human Behaviour, 3(2), 183–193. https://doi.org/10.1038/s41562-018-0518-x

van Lente, H., Spitters, C., & Peine, A. (2013). Comparing technological hype cycles: Towards a theory. Technological Forecasting and Social Change, 80(8), 1615–1628. https://doi.org/10.1016/j.techfore.2012.12.004

Winner, L. (1977). Autonomous technology: Technics-out-of-control as a theme in political thought. MIT Press.

Editorial Disclosure

Daniel Artur Raus is Founder and Scientific Editor of Interstitium. He took no part in the editorial assessment or publication decision for this contribution. The manuscript was handled and approved by Associate Editors.

Contributions

Daniel Raus de Baviera: Conceptualisation, Funding acquisition, Resources, Investigation, Writing original draft. 

Competing Interest

The author declares that he has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Publisher’s Note

Interstitium remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.