Machine learning, social learning and the governance of self-driving cars
Open accessThe DOI matches its Crossref record; 115 of the 115 references it cites were found in a registry or an index; none of the 99 sources its claims cite is checked for support yet.
Abstract
Self-driving cars, a quintessentially ‘smart’ technology, are not born smart. The algorithms that control their movements are learning as the technology emerges. Self-driving cars represent a high-stakes test of the powers of machine learning, as well as a test case for social learning in technology governance. Society is learning about the technology while the technology learns about society. Understanding and governing the politics of this technology means asking ‘Who is learning, what are they learning and how are they learning?’ Focusing on the successes and failures of social learning around the much-publicized crash of a Tesla Model S in 2016, I argue that trajectories and rhetorics of machine learning in transport pose a substantial governance challenge. ‘Self-driving’ or ‘autonomous’ cars are misnamed. As with other technologies, they are shaped by assumptions about social needs, solvable problems, and economic opportunities. Governing these technologies in the public interest means improving social learning by constructively engaging with the contingencies of machine learning.
Overview
CoChat Summary
TL;DR
The paper argues that self-driving cars are misnamed and that their governance is a high-stakes test of social learning, because machine learning systems remain opaque, contingent, and shaped by social assumptions. Using the May 2016 Tesla Model S fatal crash, it shows how industry and regulators often focus on human error and aggregate outcomes rather than the processes and purposes of innovation, and it proposes data sharing and democratized learning as governance responses.
Key findings
- “Self-driving” and “autonomous” are misnomers: cars are not self-contained, self-sufficient, or self-taught; Tesla’s “fleet learning” means each car learns from the accumulated experience of thousands of sister vehicles, and autonomy only happens when tasks and contexts are sufficiently constrained.
- In the May 2016 Tesla crash, the car was in Autopilot and hit a crossing truck at 74 mph in a 65 mph zone; NTSB data showed Joshua Brown’s 40-minute journey included 37.5 minutes on Autopilot, his hands were off the wheel for 37 minutes, he touched the wheel eight times, the longest gap was six minutes, and neither he nor Autopilot applied the brakes. The NTSB calculated he would have had at least ten seconds to react had he seen the truck, and NHTSA noted automatic emergency braking was not designed for crossing-path collisions.
- Governance responses often privatized learning: Tesla called Autopilot a “public beta,” Musk framed acceptability through aggregate math (over 1M auto deaths per year worldwide, approximately half a million saved if Autopilot were universally available), and NHTSA’s report largely directed recommendations at drivers. California DMV disengagement reports showed Tesla’s four test vehicles covered only 550 miles in 2016 and disengaged on average every three miles, while Tesla claimed customers had covered more than a billion miles in Autopilot mode by the end of 2016.
- The paper identifies constructive alternatives: Mobileye’s “semantic abstraction” versus Nvidia’s “end-to-end” learning, recognition that handoffs from automation can take as long as 40 seconds, and governance proposals for data sharing, event data recorders, and “ethical black boxes.”
Methods
- Design: qualitative STS/conceptual analysis and single-case study of public debate around self-driving car innovation, centered on the May 2016 Tesla Model S fatal crash and its aftermath.
- Data: official investigations (NTSB, NHTSA, Florida police), Tesla blog posts and statements, regulatory guidance (NHTSA, California DMV), media reports, and informal online exchanges among Tesla users.
- Analysis: interpretive tracing of competing problem definitions, promises, and learning practices; contrasts machine learning opacity with social learning and governance proposals.
Limitations
- The paper states no formal limitations section, but it acknowledges evidentiary opacity: the NTSB could determine that detection of the truck did not occur but not why, and Autopilot’s decision-making remained largely off-limits to investigators.
- Evident from the text: the analysis is a single-case study focused on Tesla and the US regulatory context, limiting generalizability to other self-driving car governance settings.
- Evident from the text: it relies on public/official documents, media, and online exchanges rather than systematic primary data, and it is normative/conceptual rather than a quantitative test of safety or governance outcomes.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 2 Not valid · 85 Not checked · 73 Own results and interpretations
In the paper
“The car's brain remained largely off-limits to investigators.”
What the sources say
- Not checked yetcites NTSB, 2017c
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“The first official report of the May 2016 crash, from the Florida police, put the blame squarely on the truck driver for failing to yield the right of way (Traffic Crash Records, 2016).”
What the sources say
- Not checked yetcites Traffic Crash Records, 2016
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“Their final word on the probable cause of the Tesla crash added a concern with Autopilot's 'operational design, which permitted [the driver's] prolonged disengagement from the driving task and his use of the automation in ways inconsistent with guidance and warnings from the manufacturer' (NTSB, 2017c).”
What the sources say
- Not checked yetcites NTSB, 2017c
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“'Self-driving' or 'autonomous' cars are misnamed.”
- Also stated in: § 'Autonomous vehicles' and 'self-driving cars' as misnomers · plain
“Just as self-driving cars cannot be self-driven, so autonomous vehicles can never be truly autonomous (Stayton, 2015).”
What the sources say
- Not checked yetcites Stayton, 2015
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 'Autonomous vehicles' and 'self-driving cars' as misnomers · plain
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Focusing on the successes and failures of social learning around the much-publicized crash of a Tesla Model S in 2016, I argue that trajectories and rhetorics of machine learning in transport pose a substantial governance challenge.”
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
51 more claims
Create a free account to see all 160 claims and to verify the 99 references still open.
- It may state two separate claims as one. 48
- Probably not a substantial claim. 76
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“According to the recent literature, factors contributing to the overall lack of algorithmic transparency include the cognitive impossibility for humans to interpret massive algorithmic models and datasets; a lack of appropriate tools to visualise and track large volumes of code and data; code and data that are so poorly structured that they are impossible to read; and ongoing updates and human”
From The ethics of algorithms: key problems and solutions (2022) · cites this paper as Stilgoe 2018
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Democratizing learning
“(GHSA, 2017; Policy Network 2016; Waymo 2017)”
§ References
“GHSA (Governors Highway Safety Association) (2016) Autonomous vehicles meet human drivers: Traffic safety issues for states.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Constructing human deficiency
“Musk claimed that these updates would 'minimize the possibility of people doing crazy things with it' (quoted in Charton, 2015).”
§ References
“Charton A (2015) Tesla owners need to stop doing crazy things with Autopilot, says Elon Musk. International Business Times . 4 November.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ The gamification of driving
“(Lipson and Kurman, 2015: 219).”
§ References
“Lipson H and Kurman, M (2016) Driverless: Intelligent Cars and the Road Ahead . Cambridge: MIT Press.”
Citations
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 1×
“As of 2016, these neural networks’ greatest feats have been in digital image recognition, where it is claimed their abilities have surpassed those of humans (Krizhevsky et al., 2012), and voice recognition (LeCun et al., 2015).”
BackgroundLeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
Human-level control through deep reinforcement learning
Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al. (2015) · Nature · doi:10.1038/nature14236 · Cited by 30,357 · cited 1×
“…and modes of machine learning (for example, reinforcement learning, which adopts a trial, error and reward approach to optimization, described in one leading paper [Mnih et al., 2015] as ‘deeply rooted in psychological and neuroscientific perspectives on animal behaviour’) is inescapably social.”
Advances in Neural Information Processing Systems 25
Zhang, Y., Sutton, C. A., Storkey, A., & Zoubin Ghahramani, A. (2012) · Cited by 12,653
Policy paradigms, social learning, and the state: The case of economic policymaking in Britain
Hall, P. A. (1993) · Comparative Politics · doi:10.2307/422246 · Cited by 7,369 · cited 1×
“These frameworks have an analytical basis in theories of politics as a form of learning (Hall, 1993), with policymaking seen not just as the playing out of stakeholder interests, but as ‘puzzling together’ (Hoppe, 2011), in the tradition of Dewey (1916).”
BackgroundThe ethics of algorithms: Mapping the debate
Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016) · Big Data & Society · doi:10.1177/2053951716679679 · Cited by 4,567 · cited 1×
“As it has become clear that citizens’ lives are increasingly shaped and cajoled by the influence of algorithms and artificial intelligence, social scientists, legal scholars and philosophers have developed critical approaches (Mittelstadt et al., 2016).”
The Black Box Society: The Secret Algorithms That Control Money and Information
Pasquale, F. (2015) · doi:10.4159/harvard.9780674736061 · Cited by 3,796
End to End Learning for Self-Driving Cars
Bojarski, M., Del Testa, D., Dworakowski, D., Firner, B., Flepp, B., Goyal, P., Jackel, L. D., Monfort, M., Muller, U., Zhang, J., et al. (2016) · arXiv · doi:10.48550/arxiv.1604.07316 · Cited by 3,198 · cited 2×
“Their emphasis is on ‘end-to-end’ machine learning, in which the system works out the rules that need to be prioritized (detecting the edges of roads, for example) for the solution of its task (Bojarski et al., 2016a).”
BackgroundHow the machine “thinks”: Understanding opacity in machine learning algorithms
Burrell, J. (2016) · Big Data & Society · doi:10.1177/2053951715622512 · Cited by 3,196 · cited 2×
“The tension between rule-fol-lowing and rule-making in machine learning has started to be explored by literature on the politics of algorithms (Burrell, 2016).”
Developing a framework for responsible innovation
Stilgoe, J., Owen, R., & Macnaghten, P. (2013) · Research Policy · doi:10.1016/j.respol.2013.05.008 · Cited by 3,184 · cited 1×
“Frameworks of responsible innovation also aim to engage upstream with the processes and purposes of innovation, recognizing that, in addition to questions of risk, public concerns also relate to how and why innovation happens (Stilgoe et al., 2013).”
BackgroundSelf-citation
Research works printed but not linked: 13
- Ackerman E (2016) After mastering Singapore’s streets NuTonomy’s robo-taxies are poised to take on new cities. IEEE Spectrum 29 December. Available at: http://spectrum.ieee.org/transportation/self-driving/after-mastering-singapores-streets-nutonomys-robotaxis-are-poised-to-take-on-new-cities (accessed 12 March 2017).
- Filipowicz A, Liu J, Kornhauser A (2017) Learning to recognize distance to stop signs using the virtual world of Grand Theft Auto 5. In: TRB annual meeting 2017, Washington, DC, 8–12 January.
- Gomes L (2016) Google self-driving car will be ready soon for some in decades for others. IEEE Spectrum 18 March. Available at: http://spectrum.ieee.org/cars-that-think/transportation/self-driving/google-selfdriving-car-will-be-ready-soon-for-some-in-decades-for-others (accessed 12 March 2017).
- Guizzo E, Ackerman E (2015) Gill Pratt discusses Toyota’s AI plans and the future of robots and cars. IEEE Spectrum, 11 September. Available at: https://spectrum.ieee.org/automaton/robotics/artificial-intelligence/gill-pratt-on-toyota-robot-plans (accessed 8 November 2017).
- Lambert F (2016) Transcript: Elon Musk’s press conference about Tesla Autopilot under v8.0 update [Part 2]. Electrek 11 September. Available at: https://electrek.co/2016/09/11/transcript-elon-musks-press-conference-tesla-autopilot-under-v8–0-update-part-2/ (accessed 12 March 2017).
- Nourbakhsh IR (2013) Robot Futures. Cambridge, MA: MIT Press.
- Parson EA, Clark WC (1995) Sustainable development as social learning: Theoretical perspectives and practical challenges for the design of a research program. In: Holling ES, Gunderson LH, Light S (eds) Barriers and Bridges to the Renewal of Ecosystems. New York: Columbia University Press, pp. 428–460.
- Perrow C (1984) Normal Accidents: Living with High Risk Systems. New York: Basic Books.
- Ranft F, Adler M, Diamond P, Guerrero E, Laza M (2016) Freeing the Road: Shaping the Future for Autonomous Vehicles. London: Policy Network.
- Wetmore J (2003) Driving the dream: The history and motivations behind 60 years of automated highway systems in America. Automotive History Review 7: 4–19.
- Winner L (1977) Autonomous Technology: Technology-out-of-Control as a Theme in Political Thought. Cambridge, MA: MIT Press.
- Wynne B (1989) Frameworks of rationality in risk management: Towards the testing of naive sociology. In: Brown J (ed.) Environmental Threats: Social Sciences Approaches to Public Risk Perceptions. New York: Belhaven Press, pp. 33–45.
- Wynne B (1992) Risk and social learning: Reification to engagement. In: Krimsky S, Golding D (eds) Social Theories of Risk. Westport, CT: Praeger, pp. 275–300.
Cited by
The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities
Floridi, L. (2023) · doi:10.1093/oso/9780198883098.001.0001 · Cited by 507
In AI We Trust: Ethics, Artificial Intelligence, and Reliability
Ryan, M. (2020) · Science and Engineering Ethics · doi:10.1007/s11948-020-00228-y · Cited by 600
“So far, this has been a fairly effective procedure, despite some controversial incidents such as the Tesla Model S fatality in 2016 (Stilgoe 2018).”
BackgroundEthical governance is essential to building trust in robotics and artificial intelligence systems
Winfield, A. F. T., & Jirotka, M. (2018) · Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences · doi:10.1098/rsta.2018.0085 · Cited by 493
“Given that real-world trials of driverless car autopilots have already resulted in several fatal accidents [46,47], there is clearly an urgent need for transparency in order to discover how and why those accidents occurred, remedy any technical or operational faults, and establish accountability.”
BackgroundFour Responsibility Gaps with Artificial Intelligence: Why they Matter and How to Address them
Santoni de Sio, F., & Mecacci, G. (2021) · Philosophy & Technology · doi:10.1007/s13347-021-00450-x · Cited by 490
“…even when this is not the intention of their authors, these proposals even run the risk of fuelling the temptation of “solutionism” (Morozov, 2013); (Stilgoe, 2017), the belief that complex socio-technical and political problems can be “solved” (or avoided) by the introduction of new techniques.”
BackgroundInfluentialAI governance: themes, knowledge gaps and future agendas
Birkstedt, T., Minkkinen, M., Tandon, A., & Mäntymäki, M. (2023) · Internet Research · doi:10.1108/intr-01-2022-0042 · Cited by 309
Metadata
- Authors
- Stilgoe, J.
- Affiliations
- University College London (GB)
- Corresponding author
- Stilgoe, Jack
- ORCID
- Stilgoe, Jack
- Venue
- Social Studies of Science
- Publisher
- SAGE Publications
- Published
- 2017-11-21
- Volume
- 48
- Issue
- 1
- Pages
- 25-56
- ISSN
- 0036-8539 (ISSN-L)
- 0306-3127
- 1460-3659
- OpenAlex
- W2607757716
- Semantic Scholar
- 95615c6bce2123f12e39c3d9eb293ebb759501aa
- Cited by
- 375
- References
- 113
- Citations by year
- 2026: 26 · 2025: 25 · 2024: 39 · 2023: 70 · 2022: 53 · 2021: 58 · 2020: 50 · 2019: 34 · 2018: 16 · 2017: 3
- FWCI
- 64.51
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Licence
- journals.sagepub.com/page/policies/text-and-data-mining-license (tdm)
- Locations
- Social Studies of Science (publishedVersion)
- SSRN Electronic Journal (acceptedVersion)
- PubMed (publishedVersion)
- UCL Discovery (University College London) (submittedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Law › Digitalization, Law, and Regulation
- Physical Sciences › Engineering › Automotive Engineering › Autonomous Vehicle Technology and Safety
- Keywords
- autonomous vehicles, machine learning, social learning, technology governance, algorithmic learning, transport policy, public interest
- MeSH terms
- Accidents, Traffic, Social Learning, Machine Learning, Humans, Technology, Motor Vehicles
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Machine learning, social learning and the governance of self-driving cars. https://app.cochat.ai/papers/10.1177/0306312717741687
- The paper
- Stilgoe, J. (2018). Machine learning, social learning and the governance of self-driving cars. Social Studies of Science. https://doi.org/10.1177/0306312717741687
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.