In a world racing toward autonomous mobility, self-driving cars promise to reshape not only how we move, but how we think about responsibility, safety, and moral choice. These vehicles are not merely machines programmed to follow lanes and respond to traffic signs: they carry embedded ethical judgments in their software. As they navigate real streets shared with vulnerable pedestrians, cyclists, and human drivers, self-driving cars increasingly face scenarios where a collision may be unavoidable—and in those critical moments, the car’s programming must decide whose safety takes precedence.

This question forces us into a deeply philosophical terrain, where engineers, ethicists, regulators, and the public must collaborate on life-and-death trade-offs. It is here, in the intersection of moral theory and real-world engineering, that the so-called “trolley problem” comes into play. But unlike classical thought experiments, the dilemmas autonomous cars confront are not neatly packaged. They involve risk, uncertainty, and competing duties that must be pre-programmed in advance—not judged in the heat of the moment by a single person.

Moreover, the decisions made in these high-stakes moments do not exist in a vacuum. They reflect values about who we prioritize when danger looms: the occupants inside the vehicle, the innocent bystander on the sidewalk, or perhaps other drivers sharing the road. And these values are not universal: they vary across cultures, legal systems, and social norms. As a result, designing an accident-response algorithm is not simply a technical challenge—it is a moral undertaking, one that demands us to confront foundational questions about how much control we are willing to relinquish to machines, how we assign responsibility, and how we build trust in a technology that must sometimes decide who to protect.

The Moral Machine on Wheels: Programming Life-and-Death Decisions

One of the most striking ethical challenges facing autonomous vehicles is how they should act when a collision is unavoidable. In such scenarios—often compared to the classic “trolley problem”—a self-driving car's software may need to decide which course of action will minimize harm. Philosophers, engineers, and policymakers are deeply concerned about this: How should these cars weigh the lives of passengers versus pedestrians? Should they always minimize the number of fatalities, or prioritize protecting their occupants?

This problem is not purely hypothetical. Although fully autonomous cars may reduce human error by a large margin, they cannot guarantee zero accidents. In rare but critical moments, programmers must confront scenarios where harm is unavoidable. As noted by ethical scholars, these “accident algorithms” resemble the trolley problem—but with important differences. [1] Rather than a thought experiment where one person is sacrificed to save five, self-driving cars often navigate uncertainty, probabilistic risk, and rapidly changing dynamics. [1]

Adding to the complexity, these decisions implicate legal and moral responsibility. Designers, manufacturers, and regulators must decide where accountability lies. Who is responsible if a car chooses a path that knowingly increases risk to one party to save another? The way we code these vehicles reflects value judgments about life, risk, and responsibility. Scholars have pointed out that tackling these dilemmas by simply applying trolley-style logic ignores real-world nuance, such as the difficulty of modeling risk accurately and the legal liability for different actors. [1]

Meanwhile, public sentiment reveals deep tensions. In MIT’s Moral Machine experiment, participants from around the world ranked different outcomes when faced with autonomous driving scenarios. [2] The results showed that people generally support utilitarian decision-making—minimizing total harm—but also strongly favor cars that prioritize their own occupants. This contradiction raises a paradox: while people may endorse “sacrificing one to save many,” they may be less willing to ride in a car that would do exactly that.

Beyond Hypotheticals: Real-World Ethical Engineering

Critics argue that the focus on dramatic trolley-problem situations is misleading. Real-world autonomous driving ethics should grapple with far more than rare, high-stakes dilemmas. Scholars have urged a shift toward “real-world ethics,” one rooted in the complexity of actual engineering, social norms, and systems design.

In this broader view, the ethics of self-driving cars connect deeply with social systems: infrastructure, regulation, and the collective behavior of road users. Rather than simply programming binary choices (“swerve or don’t swerve”), engineers should integrate ethical frameworks into motion planning, risk management, and algorithmic decision-making. [1] Some researchers propose modeling risk not just as collision probability, but as a function of both probability and expected harm; then embedding that “risk cost function” into trajectories and planning. [1]

Moreover, critics warn that focusing only on stylized dilemma cases distracts from more pressing ethical and social issues. For example, who decides the ethical settings of these vehicles? Should the moral preferences of engineers, regulators, or the public dominate? [3] Others emphasize that responsibility for accidents isn’t solely technical: there are liability, legal, and social dimensions that must be addressed. Decision-making must account for not only what is morally “best,” but also what is fair, accountable, and legally coherent.

Some argue that autonomous vehicles should avoid making any complex moral tradeoffs: if a car must swerve, it should do so only when it doesn’t impose additional risk on bystanders. This principle reflects a conservative approach—limiting the moral agency of the car and constraining how far its decision-making can reach into high-risk ethical territory.

At the same time, new empirical research is moving away from artificial trolley-type scenarios. Academics at North Carolina State University have developed experiments using virtual reality to simulate more realistic traffic dilemmas—such as mundane yet morally relevant driving situations—grounded in what they call the Agent-Deed-Consequences (ADC) model. Instead of focusing exclusively on hypothetical life-and-death tradeoffs, these experiments aim to teach autonomous systems how humans make moral decisions in everyday driving. That shift, its proponents argue, could yield more realistic and ethically robust behavior from self-driving cars.

Adding another layer, public surveys show that many people make different moral judgments depending on whether they imagine themselves in a human-driven car or a self-driving one. In controlled experiments, participants tended to more strongly prefer self-driving cars that actively minimize harm, compared to how they judged human drivers in similar dilemmas. This suggests that people may hold autonomous systems to higher moral standards than human drivers—or at least different ones.

Further complicating matters, some researchers question whether an algorithmic “moral agent” is even the right framing. Rather than giving cars the ability to make active moral judgments, perhaps we should view them as tools constrained by design rules, safety protocols, and legal obligations. From this perspective, the focus shifts from programming moral reasoning into the car to setting up clear regulatory frameworks and responsibility structures for designers, manufacturers, and governing bodies.

These debates—between abstract moral dilemmas and practical engineering ethics, between utilitarian tradeoffs and risk-aware restraint, between public expectations and legal responsibility—are not academic luxuries. As autonomous vehicles become more common on our roads, the ethical choices embedded in their software will shape not only individual accidents, but social trust, liability systems, and even legal regimes. How we answer “who a self-driving car should protect” is not just a question of programming, but of values—and the answers will define the future of mobility.

Sources:

[1]: https://link.springer.com/article/10.1007/s10677-016-9745-2

[2]: https://en.wikipedia.org/wiki/Moral_Machine

[3]: https://academic.oup.com/book/44058/chapter-abstract/371949662

References:

https://research.chalmers.se/publication/532088/file/532088_Fulltext.pdf

https://news.ncsu.edu/2023/12/ditching-the-trolley-problem

https://pubmed.ncbi.nlm.nih.gov/31749736

https://www.mdpi.com/2504-3900/1/3/174