Futuramas of the present: the "driver problem" in the autonomous vehicle sociotechnical imaginary

The paper interrogates the contemporary sociotechnical imaginary of autonomous vehicles (AVs), focusing on how a widely cited claim—namely, that more than 90% of road crashes result from human (driver) error—constructs the problem that AVs purport to solve. Rather than speculating about distant futures, the authors examine how expectations about AVs shape the present by legitimating continued investment in automobility. They argue that the AV imaginary reproduces longstanding narratives of technological determinism and automobility’s promised futures, positioning AVs as a safety solution while rendering alternatives unnecessary or obsolete. The core research question asks whether the framing of "the driver" as the dominant cause of crashes is an empirically grounded fact or a methodological and discursive construct that serves to absolve automobility itself of inherent violence and risk. This focus situates AVs within a lineage of automobility imaginaries and highlights the political, ethical, and epistemological stakes of defining road violence as contingent and solvable primarily through technology.

The paper situates AVs within the framework of sociotechnical imaginaries (Jasanoff and Kim), noting how industry, governments, media, and academic actors co-produce and stabilize visions of desirable technological futures. It reviews historical and contemporary automobility imaginaries (e.g., Futurama exhibits; electric, connected, and autonomous vehicles), emphasizing their justificatory technological determinism and recurring promises. The authors draw on critical social science perspectives (e.g., Urry, Latour, Virilio, Baudrillard, Zizek) that conceptualize technology as intrinsically social and accidents as revelatory of underlying socio-technical orders. The road safety literature and accident causation models are canvassed: from early linear models (Heinrich’s domino theory) and the Haddon Matrix to systems approaches (Reason; Hollnagel’s CREAM) and derivatives such as DREAM and SNACS in Europe. The review shows how dominant Driver–Vehicle–Environment (DVE) schemas, widely used in US (Tri-Level Study; NHTSA NMVCCS) and European projects (EACS; ETAC; TRACE), consistently center human factors as primary causes, thereby normalizing automobility and under-theorizing vehicles, environment/infrastructure, and automobility as a socio-technical totality.

The study undertakes a critical, reflexive analysis of how the "93% human error" statistic was constructed, circulated, and stabilized across influential US and European crash causation studies. It reviews primary reports and methodological materials (e.g., codebooks, field coding manuals, analytic frameworks) from the Indiana Tri-Level Study, NHTSA’s NMVCCS, and European initiatives (EACS, ETAC, SafetyNet/SNACS, TRACE). The authors analyze the underlying theoretical assumptions (DVE schemas; definitions of "critical reason" and "probable cause") and how coding practices and variable construction distribute causality among driver, vehicle, and environment while excluding automobility as a causal category. The paper also reflects on the performative and rhetorical aspects of citation practices that transform qualified findings into "well known" facts. While proposing what an ethnomethodological "study of the study" would entail (following coders, observing on-scene data capture, tracing recoding/analysis, and dissemination), the authors do not conduct such fieldwork; rather, they provide a conceptual and document-based critique of accident causation methodologies and their epistemological commitments.

• The frequently cited claim that ~93% of crashes are due to human/driver error is a product of specific methodological definitions and coding practices in accident causation research, not a neutral empirical fact.
• US studies: The Indiana Tri-Level Study reported human factors as probable causes in 92.6% of investigated accidents (rounded to 93%), with environment at 33.8% and vehicle at 12.6%. NHTSA’s NMVCCS, using similar frameworks, assigned the "critical reason" to the driver in 94% (±2.2%) of crashes, ~2% to vehicle, and ~2% to environment.
• European studies: The European Truck Accident Causation Study attributed 85.2% of accidents to human error, with other factors playing minor roles; parallel projects (EACS, TRACE, SNACS) follow DVE schemas that privilege human factors.
• Through repeated citation and dissemination (e.g., Singh 2015 report cited hundreds of times), the statistic has become "well known" and "widely accepted," evidencing a performative transformation of a qualified, method-bound result into an authoritative fact.
• Accident causation models (from domino theory to Haddon, Reason, CREAM/DREAM/SNACS) define accidents and "critical reasons" in ways that predispose findings toward human culpability (e.g., defining the immediate precursor as a person’s failed action), thereby making driver error the default explanatory category.
• Automobility as a socio-technical complex is excluded as a causal entity; consequently, road violence is framed as contingent and remediable via technology (e.g., AVs), rather than intrinsic to automobility.
• An alternative framing would view the car–driver–environment assemblage as causal in 100% of crashes, shifting focus from removing the human to addressing the socio-technical system of automobility itself.
• The AV sociotechnical imaginary depends on the driver-error construct to justify AVs as a safety solution; in practice, this imaginary serves to reproduce and expand automobility rather than transform it.

The analysis demonstrates that the AV safety promise rests on a methodologically constructed premise: that human drivers cause the vast majority of crashes. By unpacking how crash causation studies define accidents, assign "critical reasons," and distribute causality, the paper shows that the dominant DVE schemas systematically center human error while bracketing automobility as a whole. This framing allows proponents to argue that substituting computers for human drivers will proportionally reduce road death and injury. However, if accidents are intrinsic to the car–driver–environment assemblage and to automobility’s socio-political order, then replacing the human component does not address the underlying production of road violence. The findings thus challenge the AV imaginary’s claim to radical transformation, suggesting instead that AVs perpetuate existing mobility regimes, investments, and externalities. Recognizing the performative creation of the "93%" fact invites broader ethical and political deliberation about mobility futures, beyond engineering-led determinism, and opens space for alternative imaginaries centered on responsibility, citizen engagement, and critique of automobility violence.

The paper concludes that the oft-cited statistic attributing ~93% of crashes to human error is a product of particular accident causation methodologies and citation practices. This construct underpins the AV sociotechnical imaginary, legitimating AVs as a safety fix while masking automobility’s intrinsic production of road violence. Rather than heralding a radically different future, the AV imaginary ensures the reproduction and expansion of automobility—more cars, roads, infrastructure, and associated externalities. The authors call for reflexive, empirically grounded research into how crash statistics are produced and stabilized, including ethnographic "studies of the study." They propose exploring alternative epistemic and methodological frameworks—such as phenomenological approaches focusing on embodied experience, situated practices, and near-misses—to rethink agency, risk, and road violence. Future research should pursue citizen-focused, participatory sociotechnical imaginaries that challenge technological determinism and consider non-automobility alternatives.

The authors did not conduct an on-the-ground ethnographic or ethnomethodological investigation of crash data production (e.g., observing on-scene coding, recoding, and analysis). Their critique relies on document analysis and conceptual examination of methodologies and citation practices. As such, while they outline the processes likely involved in constructing the "93%" statistic, a detailed empirical account remains to be undertaken, which may affect the specificity and generalizability of procedural inferences.