Preface: A death without a defendant:
On the 18th of March, 2018, an Uber with no driver behind the wheel killed Elaine Herzberg, a woman who was riding a bicycle, in Tempe, Arizona. She was the first person to die in an AV-caused pedestrian accident. She was picked up by the sensors in the car. The system deemed her as an “other thing. The system treated her as an “other object”. The Emergency brake system (EBS) was disabled. An operator who was the man behind the wheel of the machine holding a phone.
None of the perpetrators faced any prison sentences. Uber paid the civil settlement. Uber settled the civil case. The safety operator was charged (then acquitted) with the crime of “wild fires.” The manufacturer was not held criminally responsible. Now, of course, the AI encountered nothing whatsoever.
Elaine Herzberg’s death brought up a moral question. It opened one of the most significant legal questions of this century: When an AI kills, who is the defendant?
The legal gap at the heart of Autonomous Mobility:
Autonomous vehicles or the automatic vehicles (AVs) operate across six levels as defined by the Society of Automotive Engineers (SAE) international (SAE J3016). SAE J3016 is the recognized standard that defines the taxonomy and levels of driving automation for on-road vehicles. The level ranges from 0, i.e., (no automation), where the vehicle is controlled by humans, to level 5 (fully automatic), where no intervention of humans is required. As the vehicle approaches levels 4 and 5, the driver, who is traditionally the liability holder in traffic accidents, begins to disappear from the liability entirely.
Over the years, tort law has evolved significantly, and was built on a foundational assumption that “a human being made a decision that caused harm.” Under the doctrine of negligence, the plaintiff must establish the duty, breach, causation, and damage all related to the conduct of a person or any entity that is legally cognizable. When an AI system, which is operating beyond real- time human control, makes a quick decision that results in the death of a human being, the traditional laws or the frameworks collapse here.
The question then breaks down into three competing theories of liability: “the owner”, “the manufacturer,” and most provocatively “the AI itself.”
Theory one: The Owner’s Liability:
As per the principles of vicarious liability and negligent entrustment given in the tort law, the most legally accessible theory holds the vehicle’s owner responsible for the death caused by his/her vehicle. As per these principles, the owners have historically borne responsibility for how their vehicles are used. In Hertz Corp. V. Friend, 599 U.S. 77 (2010), the Supreme Court reaffirmed the broad interpretation of party responsibility in the commercial vehicle context. As per the Graves Amendment (49 U.S.C 30106), it shields the commercial lessors from liability in cases where they are not negligent. However, it implies by contrast that negligence on the owner’s part remains a viable theory.
For the private autonomous vehicle users, the argument remains inherent: you purchased a vehicle, you deployed it and, and you benefited from its use. Take an example, such as if a loaded weapon were discharged in your house, and it killed a visitor. Here, your ownership and deployment of that object will form the main reason for the negligence claim. In his influential 2011 paper named “Open Robotics” (94Maryland Law Review), Legal scholar Ryan Calo argues that the deployment of autonomous systems should carry with it a form of enterprise liability as the owner gets profit from the system’s operation and should internalize its risks.
However, this theory has certain significant limitations, such as: “What negligence did the owner actually commit? If the vehicle was well maintained, updated, and operated as per the manufacturer’s guidelines, it means that the owner may have done everything correctly. The vehicle made the decision, not the owner, nor did the owner direct; he/she anticipated. Holding the owner liable for it based on the enterprise liability approach, regardless of any fault, is a solution, but it remains unsettled in most of the jurisdictions.
Theory Two: The Manufacturer’s Liability/ Product Liability’s Finest Hour?
The Product Liability acts as the strongest existing legal framework for automatic vehicle (AV) casualties. Apart from this, the doctrine established in the Greenman case serves as the best legal framework for holding the manufacturer liable. In the case of Greenman V. Yuba Power Products, 59 Cal. 2d 57 (1963), the California Supreme Court established strict liability for manufacturers of defective products that cause injury. As per this principle codified under the law, it does not require proving the manufacturer’s negligence. The only requirement here to be fulfilled is that the product needs to be defective, and the defect caused the harm.
As it applies to the AVs, the argument is quite compelling: the vehicle’s decision- making algorithm is part of the product. If it has been decided that a correct system or a reasonable system should not have been made, it is defective. The entity that designed, trained, tested, and deployed that algorithm in the vehicle or the manufacturer bears strict liability.
The case of Elaine Herzberg points directly towards this theory. The National Transportation Safety Board’s investigation found that Uber’s system had “no system for the operator to be alerted to” If the vehicle misclassified an object consecutively. Then this is a design defect in the product liability sense and is an inevitable failure mode for which no adequate safeguard was designed.
Taking a view of the same on a global level, the United Kingdom’s “Automated and Electric Vehicles Act 2018” has codified a version of this approach, which states that when an accident is caused by an automatic vehicle running in self–driving mode, then the vehicle’s insurer, standing in for the manufacturer in a first – party claim, will be held liable. The act explicitly shifts the burden to the manufacturer or to the system’s designers and operators, not on the human passenger.
Similarly, the German law, as per their “Road Traffic Act” and the “Autonomous Driving Act,” creates a liability where the vehicle keeper who keeps their vehicle at level 4 operations bears primary liability with some rights against the manufacturer for system failures.
In his journal “Journal of Law and Technology” (2020), Professor Gary Marchetti argues that product liability is not merely the most applicable framework; it is the most just: “Those who get profit from the algorithmic decisions made at machine speed and machine scale should bear the costs when those decisions kill.”
Theory Three: The AI as Legal Person- Provocative but Premature:
The most philosophically bold counter to the liability query is whether AI-provoked systems, in themselves, need to be recognized as legally liable organisms that can be held accountable. The European Parliament resolution from 2017 on Civil Law Rules on Robotics discussed the possibility of establishing a new status for electronic persons that may, in principle, have rights and liabilities.
At present, it is more a thought experiment in jurisprudence than any doctrine that can be observed in the meantime. The initial hurdle is that the notion of legal personhood has always needed either biological humanity or the fiction of collective human actions and decisions, namely, the corporation as a bundle of human interests and decisions. Even though an AI system is very advanced, it lacks assets to stake, its reputation can’t be harmed, and no deprivation of liberty can be applied. It cannot be called evil within the meaning of the law, let alone within the soul of the law, according to its definition of evil.
These are examples of a construct of legal personhood, one made where necessary, which is removed where it hinders, as witnessed by the Eisner v. Macomber line of cases and more generally, by the doctrine of “corporate personhood” that was developed under Citizens United v. FEC, 558 U.S. 310 (2010). The question here is not philosophical, however, but pragmatic: is it fulfilling the compensation and deterrence goals of tort law to establish a liability vehicle under the name of an AI?
But, according to the study by Professor Ryan Abbott, titled “The Reasonable Computer: Disrupting the Paradigm of Tort Liability” (George Washington Law Review, 2018), it doesn’t (if Professor Abbott is correct, at least) — at least not yet. It is, he says, better to consider AI decision-making as a kind of manufacturer conduct and make the humans who created and deployed it account for working out what they reasonably could expect the system to be able to create and deploy.
The Causation Problem: When No One Decided:
Beyond the question of who is liable rests a deeper evidentiary crisis: how do you prove what an AI system has “decided” and “why?”
Barely getting a handle on deep learning, and our workhorses of billions and billions of weighted parameters that we can’t comprehend, is Advanced Deep Learning. If the system fails, training dates, model architecture, and run time logs – which many manufacturers do not release as public knowledge – will have to be looked at to be able to determine why the system failed. While the case was unrelated, the Supreme Court of Canada (“SCC”) in Uber Technologies, Inc. v. Heller, 2020 SCC 16, reaffirmed the rule that a lack of corporate transparency does not excuse companies from liability when it comes to algorithms. Despite the context being different, in Uber, the Supreme Court of Canada (“SCC”) reiterated that there is no algorithmic “holiday” for companies when it comes to liability.
Today, the need for “explainability,” that is, the need to describe the behavior of AI systems in natural language and modes understandable to humans, is part of the regulatory frameworks already in place such as the EU Artificial Intelligence Act 2024, which establishes specific and compulsory provisions on transparency, logging and human oversight for autonomous vehicles, defined as high-risk AI systems (Articles 9-17). The misgivings create in practice a ‘paper trail’ that could be used as a basis for legitimate liability claims in the future.
Toward a Framework: Shared, Graduated Liability
The recent trend of legal scholarship and comparative law suggests the existence of a general liability system, with degrees corresponding to the degree of autonomy.
Human drivers have a large role to play in lower automation levels (SAE Level 1-2): There is little modification of the negligence doctrine. A manufacturer’s responsibility for both system failure and its owner’s responsibility for misuse or failure to maintain it exist in parallel, at levels 3-4. From level 5 onwards, the most logical answer is for the manufacturer and developer to be strictly liable, with compulsory insurance coordinating pools, dispensing with all fault assessment, provided compensation can be recovered quickly.
From the analysis, it is noteworthy that India, a country with the fastest-growing AV pilots in the world, has no specific AV liability Laws. The Motor Vehicles Act, 1988, and the 2019 amendment thereto, say very little about automated vehicles, and place more emphasis on the general negligence or product liability concepts enunciated in the Consumer Protection Act, 2019, and in tort law, thereby leaving a growing lacuna.
Conclusion: The Law Must Catch the Machine:
In any meaningful way in which one can conceivably hope for justice, Elaine Herzberg’s death never goes unpunished. This is not just an injustice; it’s a sign of the weaknesses in the legal framework that regulates autonomous systems.
The owner could be a reasonable defendant, but he is a narrow one. The doctrinal target is the manufacturer, and the most immediate and doctrinal framework is strict products liability. For the moment, the AI will not be allowed to defend itself — but the problems of ‘accountability’ that it raises are going to be key to the next chapter of thinking in law.
One thing’s for sure: When an autonomous vehicle becomes regular on the road, the law cannot afford to be open to interpretation. Each mile that the driverless car rolls is also a mile of uncharted legal issues. A society that refuses to draw that map, as a seriously misguided society, will be a society that will continue to engender defendants who no one can call to account and a society that will continue to create victims who are not compensated for.
The machine has no conscience. The law must supply one.
THIS ARTICLE IS WRITTEN BY AKANSHA KUMARI FROM CHANDIGARH UNIVERSITY
REFERENCE :
(i) NTSB Report: Collision Between Vehicle Controlled by Developmental Automated Driving System and Pedestrian, Tempe, Arizona, March 18, 2018
(ii) Hertz Corp. V. Friend, 599 U.S. 77 (2010)
(iii) Ryan Calo, *”Open Robotics”*, 94 Maryland Law Review (2011)
(iv) Greenman v. Yuba Power Products*, 59 Cal.2d 57 (1963)
(v) UK Automated and Electric Vehicles Act 2018
(vi) Ryan Abbott, *”The Reasonable Computer: Disrupting the Paradigm of Tort Liability”*, George Washington Law Review (2018)
(vi) Uber Technologies, Inc. v. Heller*, 2020 SCC 16
(viii) EU Artificial Intelligence Act 2024, Articles 9–17
(ix) SAE International Standard J3016 — Taxonomy and Definitions for Terms Related to Driving Automation Systems
(x) Motor Vehicles (Amendment) Act, 2019 (India)