Addressing Manufacturers’ Liability in Accidents Caused by Self-Driving Cars

The rise of artificial intelligence may disrupt many aspects of life as we now know it, not least of which is tort law. It remains an open question how much the government should regulate artificial intelligence. The European Union’s Artificial Intelligence Act took effect in August 2024 [1]. The following month, the governor vetoed California’s contentious SB-1047, which would have required developers of powerful AI models to follow certain safety requirements [2][3]. Regarding tort law, it will not always be easy to determine whose actions caused an AI-related harm, who owed a duty to whom, and what constitutes negligence in developing or implementing AI.

Though there are many angles from which to approach the topic, this article will examine liability in accidents involving self-driving cars. One possible way to address such cases is to allow current liability framework (with modest adaptations) to compensate victims of such accidents. Another would be to create a fund to compensate victims similar to how workers’ compensation compensates workers injured on the job, as suggested by Professors Kenneth Abraham and Robert Rabin [4]. Though extensive research has gone into the latter option, there is as of yet insufficient certainty to dismiss the most organic solution of product liability. Current law holds manufacturers responsible for defective products that harm users, and the law should be given the chance to hold the producers of self-driving cars responsible before an alternative to tort is introduced.

To begin, we imagine a hypothetical: One fully autonomous vehicle with one passenger drives off the road into a wall. Let us further assume for simplicity that the vehicle’s capabilities reach the point that it requires no driver intervention, nor would a “driver,” more accurately a passenger, be negligent in not paying attention to the road. The passenger may wish to recover damages from the manufacturer, the developer whose autonomous driving software was installed in the car, and the car dealer. Who will be held liable?

Assuming that there was no negligence on the part of the passenger, such as failure to maintain the car properly, this case can be handled as a product defect. Nancy Birnbaum provides an overview of strict products liability, including some of the cases we will now consider [5]. First, we consider Thomas v. Winchester, wherein the defendant prepared and sold a medicine to B., who sold it to C., who sold it to the plaintiff [6]. The plaintiff gave the medicine to his sick wife, but the medicine turned out to be poison mislabeled by the defendant. The plaintiff’s argument is clear—one who either fraudulently or negligently prepares a poison mislabeled as medicine and sells it should be liable for the harm caused. The defendant has no factual defense, but legally he argues that he only entered into contract with B., to whom he directly sold the medicine, and hence only has a duty to him. Had B. taken the medicine and suffered harm, surely the defendant would be liable, but he has no contractual obligations to the plaintiff or his wife.

The New York Court of Appeals decided that since the defendant knew that his negligence could cause great harm, he had a non-contractual duty to any to whom his medicine might be sold [7]. He knew that B. was not the end consumer but rather that someone like the defendant’s wife would be. Hence, he owed her a duty, which he breached to her harm. In our hypothetical, we can establish that both the car manufacturer and the software developer cannot demur simply because they did not sign a contract with the plaintiff.

Similarly, the majority in MacPherson v. Buick Motor Co. held that the manufacturer of a car with a defective wheel (which it did not produce but purchased) can be held liable for injury the car causes as a result of the defect [8]. The majority opinion cites Thomas v. Winchester saying that its principle has never been challenged and that application of its rule has expanded. The majority reasons that though cars are not inherently dangerous, they are made dangerous by defective wheels, and consequently the car’s manufacturer has a duty to the user. The opinion concedes that applying the rule to non-dangerous products which are made dangerous by defect may not stem from Thomas v. Winchester but rather from a later case, Devlin v. Smith, which it fully upholds. (The dissent argues that Devlin v. Smith was about an inherently dangerous product, that the rule about non-contractual products defect liability applies only to products that are inherently dangerous, and consequently that the plaintiff cannot recover against the manufacturer [9].) Applying the majority's ruling to our hypothetical, we clearly see that the manufacturer has a duty to the passenger, for the car was rendered dangerous by the autonomous driving system’s defect.

Now let us consider another hypothetical. Suppose that the car for no apparent reason brakes and comes to a sudden stop in the middle of the road. Not only does this cause the passenger to hit his head on the dashboard, but the sound of the tires screeching startles a little boy on a trampoline in a nearby backyard, who falls and hits his head. Again, supposing no negligence on the passenger’s part, a defect in the car probably caused both injuries; without the defect, we can assume neither would have occurred. (Of course, data from the vehicle’s computer should be examined and presented to the jury to verify that the software caused the car to stop, and that there was no good reason to stop, such as a pedestrian about to jump in front of the moving vehicle.) But who can recover? In Palsgraf v. Long Is. R.R. Co, the court takes up a case where a railroad employee pushes a late-coming passenger onto the train, causing the passenger’s package to fall [10]. The employee was negligent and reasonably responsible for damage to the man’s property. But the plaintiff in this case was a bystander who was injured when the package, which happened to contain firecrackers, exploded and caused a scale to fall on her. Did the railroad have any duty to this woman? The majority rules no, for the wrong was done to the passenger whom the employee pushed onto the train [11]. The majority wonders: Who could foresee that the package contained firecrackers, much less that such a chain of events would lead to a scale falling on this unfortunate woman? The woman, the majority claims, is basing her claim on a wrong not to herself but to another. The dissent replies that people are nonetheless responsible for the effects of which their actions are the proximate cause and that the cause of the passenger dropping his package was simultaneously the proximate cause of this woman’s injury [12]. Yet even the dissent notes that for the sake of efficiency, a line must be drawn politically as to what constitutes a proximate cause. Returning to our hypothetical, we can use the majority opinion to arrive at the conclusion that whereas the manufacturer is liable for harm to the passenger, who bought the car for safe carriage from departure to arrival, it cannot be held liable for harm to the boy who falls off the trampoline. The manufacturer wronged the passenger, not the boy. We also note that even if we were to apply the dissent’s opinion, it is far from clear that the defect could be considered a proximate cause of the boy’s injury.

Now we jump to Greenman v. Yuba Power Products, which holds that strict product liability applies whenever a consumer uses the product as intended, and a defect causes him injury [13]. Initially, in its most basic form, non contractual liability could result from the sale of an inherently hazardous product. Over time, this principle expanded to products rendered hazardous by defective production, bringing defective cars within the scope of strict product liability. Thus, over the duration of more than a century, this rule developed from settling a case arising from poison sold as a cure to settling a case arising from a defective power tool. Given this historical progression, it would require great cause not to allow tort law also to respond to the development of self-driving cars.

Turning our attention to an article written by Professors Abraham and Rabin, we will consider another hypothetical: a collision between two fully autonomous vehicles. They note the difficulty victims face in proving a reasonable alternative design due to the highly complex and rapidly evolving nature of autonomous vehicle technology – a concern which must be addressed [14].

In their article, Professors Abraham and Rabin thoughtfully propose a tort reform similar to workers’ compensation that would in most cases limit victims’ recovery to a Manufacturer Enterprise Responsibility (MER) fund while eliminating the need to determine liability through highly technical engineering disputes [15]. But perhaps there is a simpler, more organic way the law could develop through a simpler, less technical way of determining liability. Following MacPherson, if the manufacturer produces the product in such a way that makes it a hazard to a proper user, the manufacturer can be found liable. This is the clear moral of the case, even though in that particular case a defective wheel could be positively identified as the cause of the hazard. Similarly, in a collision between two autonomous vehicles, assuming no negligence by either driver, neither vehicle should be a hazard, yet if a collision still occurs, clearly something made at least one vehicle hazardous. If these had been people driving the cars, a jury would assign a percentage of fault to each driver based on the evidence. This evidence might include actions the drivers took as drivers (e.g., steering, braking, etc.) as well as the state the drivers were in (e.g., intoxication, distraction, etc.). With autonomous vehicles, the latter category of evidence falls away while the former remains unchanged. Will some litigants retain expert witnesses to give highly technical testimony? Almost certainly. But will this testimony persuade the average jury more than extensive video evidence undoubtedly available from self-driving cars? This remains to be seen, but it is far from given. Despite not being able to articulate the reasonable alternative design as precisely as in MacPherson, in our hypothetical, a jury can certainly tell if an AI-driven car caused the collision. As to the existence of a reasonable alternative design, the defense can produce its opinion as to what abnormal circumstances caused the incident. There will be plenty of data from other autonomous vehicles that can show whether the computer program met industry standards—if under similar circumstances every car is having a wreck, then there is no product defect. However, if under nearly identical circumstances there are very few wrecks, then we have a defect. What may be more useful than technical engineering testimony is statistical evidence, which is communicated to a jury more clearly and effectively. Of course, there will be debate as to whether the situations from which specific data are taken are in fact similar enough to the incident. Though this debate may get technical, it may still prove significantly more understandable than a debate over a reasonable alternative design (consisting of technical computer algorithms) specified by the plaintiff. Before passing sweeping legislation at the federal level, perhaps it is worth giving the law time to evolve organically. If this fails, then perhaps legislation is necessary.

Left unaddressed to this point is whether the final manufacturer of the car, the software developer, or someone else involved in making the car should be held responsible. Professors Abraham and Rabin offer a perfect solution that simply uses indemnity agreements and price adjustments of component parts [16]. They also fully acknowledge that their proposed reform should not be implemented immediately—the current landscape of vehicles does not yet warrant moving beyond tort. But they also identify the in-between problem of accidents involving an autonomous vehicle and a human-driven vehicle. While it makes sense that the passengers of the autonomous vehicle would be compensated through MER, none of the three options for traditional drivers is satisfactory to the authors. They note these problems: Giving traditional drivers access to MER ignores their potential negligence, not to mention they have not paid into the system; forcing them to buy special insurance decreases autonomous-vehicle manufacturers’ incentives to make their vehicles safer; and leaving traditional drivers with recourse to tort leaves the complexities MER was meant to solve [17].

In his response to Professors Abraham and Rabin, Professor Ryan Calo argues that when widely adopted, autonomous vehicles will completely transform our relationship with cars [18]. In essence, though he recognizes the merits of the proposed reform, he thinks it is premature because when it comes time to implement the system, it may no longer work. This is another reason to advise caution in moving beyond tort for car accidents.

Though Professors Abraham and Rabin offer an impressive plan for addressing accidents involving autonomous vehicles, which are almost certain to cause not a little confusion, the product liability framework provides enough structure to address the problem that it merits a chance in the courts. As they mention, self-driving cars will drastically decrease the number of car accidents. It seems counterintuitive then to adopt sweeping reform to eliminate litigation over an already much smaller number of car accidents than we have now. Workers’ compensation addressed frequent accidents. By contrast, Professors Abraham and Rabin’s proposal would usher in radical reform to address infrequent accidents while severely limiting what victims can recover compared to the substantial verdicts they can receive under tort. Extensive video of the accident in question bolstered by statistical proof may prove a sufficient alternative to the technical engineering disputes that they warn could arise from car accidents. Further, in passing we mention Professor Calo’s caution that the world of autonomous vehicles may look completely different from today and may require a completely different system to one we could envision today. Tort law has adapted to countless past innovations. We should let it try to do so once more.

Bibliography

[1] “AI Act enters into force,” European Commission, August 1, 2024, https://commission.europa.eu/news/ai-act-enters-force-2024-08-01_en.

[2] “SB-1047 Safe and Secure Innovation for Frontier Artificial Intelligence Models Act,” California Legislative Information, September 3, 2024, https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB1047.

[3] Governor Gavin Newson, “I am returning Senate Bill 1047 without my signature,” Office of the Governor of California, September 29, 2024, https://www.gov.ca.gov/wp-content/uploads/2024/09/SB-1047-Veto-Message.pdf.

[4] Kenneth S. Abraham and Robert L. Rabin, “Automated Vehicles and Manufacturer Responsibility for Accidents: A New Legal Regime for a New Era,” 105 Va. L. Rev. 127 (2019).

[5] L. Nancy Birnbaum, “Strict Products Liability and Computer Software,” 8 Computer L.J. 135 (1988).

[6] Thomas v. Winchester, 6 NY 397 (1852).

[7] Ibid.

[8] MacPherson v. Buick Motor Co., 217 NY 382 (1916).

[9] Ibid.

[10] Palsgraf v. Long Is. R.R. Co., 248 NY 339 (1928).

[11] Ibid.

[12] Ibid.

[13] Greenman v. Yuba Power Products, 59 Cal.2d 57 (1963).

[14] 105 Va. L. Rev. 127 (2019).

[15] Ibid.

[16] Ibid.

[17] Ibid.

[18] Ryan Calo, “Commuting to Mars: A Response to Professors Abraham and Rabin,” 105 Va. L. Rev. Online 84 (2019).