Robot Rules

Jacob Turner’s Robot Rules arrived in 2018 (the paperback edition carries a 2019 date), published by Palgrave Macmillan in the window between the European Parliament’s 2017 resolution on civil law rules for robotics and the first wave of national AI strategies that landed in Europe, East Asia and North America over the following two years. Turner was then a barrister at Fountain Court Chambers in London — a commercial litigator by day and a frequent commentator on the legal implications of autonomous systems — and the book reads as a working lawyer’s attempt to tidy up a field that had, by the late 2010s, become loud with ethical manifestos and short on concrete regulatory design. It joined a cluster of volumes trying to translate the AI discourse into something that legislatures and courts could actually act on, alongside work by Ryan Calo, Jack Balkin and Harry Surden, but Turner’s framing is the most explicitly systematic of the lot: he sets out to describe what rules a lawyer would need to write, in what order, and under what authority.

The book’s central argument is that artificial intelligence is sufficiently unlike earlier objects of legal regulation that it deserves its own body of law, built around three questions that cannot be dodged forever: who is responsible when an AI system causes harm, what rights (if any) an AI system itself should bear, and how the moral codes that AI designers already encode by default should be written, scrutinised and enforced. Turner is careful not to overclaim. He does not say that current law has nothing to offer, and he spends several chapters showing where contract, tort, product liability and criminal law can be stretched to cover autonomous systems. His claim is narrower and more durable — that stretching only works for a while, that each extension creates new ambiguities, and that waiting for jurisprudence to settle these questions case by case is a decision in itself, one that leaves a decade of harms and windfalls to be sorted out after the fact.

The structure of the book moves from first principles into institutional design. An opening chapter sets out what Turner means by AI, settling on a working definition that leans on autonomy and the ability to act in ways not fully pre-specified by a designer — narrow enough to exclude every spreadsheet macro, broad enough to include the systems that regulators actually worry about. A chapter on unique features explains why, in his view, the standard legal tools fall short: AI systems can take decisions whose reasoning is not reconstructible, can learn behaviours their creators did not programme, and can be deployed in combinations whose emergent behaviour nobody owns. The responsibility chapter then works through the civil and criminal implications in turn, drawing on real cases and plausible scenarios. The chapter on rights takes seriously, without endorsing, the question of whether an AI could ever be a legal person — and uses the comparative legal history of corporations, ships, idols, rivers and animals to show that legal personhood has always been a social construction rather than a metaphysical one. From there the book turns to the design of moral codes themselves, and then to the institutions that would supervise them: a regulator, a licensing regime, something that works on AI the way the FDA works on pharmaceuticals or the SEC works on financial disclosure. A short epilogue widens the frame to international coordination.

The examples Turner draws on are the ones a reader of the 2010s technology press will recognise, which is deliberate — he wants to anchor the legal argument in facts that have actually happened. Microsoft’s 2016 Tay chatbot, which learned racial epithets from Twitter users within hours of launch, illustrates how learning systems can produce liabilities their creators did not foresee or intend. The Random Darknet Shopper, the 2014 Swiss art project whose bot purchased ecstasy on a dark market and had the pills exhibited in a gallery, surfaces the question of how criminal intent travels across a decision chain that includes an autonomous agent. DeepMind’s AlphaGo, and its successor AlphaGo Zero, serve as the standard examples of systems whose moves are opaque even to their designers. Uber’s 2018 fatal collision in Tempe, Arizona and the sequence of Tesla Autopilot incidents frame the autonomous-vehicle liability discussion. The 2017 awarding of Saudi Arabian citizenship to Hanson Robotics’ Sophia robot is treated with the irony it deserves, but Turner uses it to open a longer discussion of what granting legal status to a machine would actually imply. The ProPublica investigation of the COMPAS recidivism scoring algorithm supplies the running example for algorithmic bias and due process. He returns repeatedly to the European Parliament’s 2017 resolution — the one that proposed “electronic personhood” for the most autonomous robots — not to defend it but to show how a legislature reaches for the nearest analogy when it does not have the right one.

Autonomous weapons appear in the book, though they are not its focus. Turner devotes a section to the Campaign to Stop Killer Robots and the discussions under the Convention on Certain Conventional Weapons in Geneva, and he quotes figures including Stuart Russell on the specific risks of delegating lethal decisions to algorithms. His treatment is careful: he notes that existing international humanitarian law — the principles of distinction, proportionality and precaution, together with the Martens Clause — already applies to any weapon a state fields, and that the harder regulatory question is not whether these rules bind but whether they can be enforced against systems whose decision processes cannot be audited after the fact. Readers coming to the book for a definitive position on a ban will not find one. Turner lays out the arguments on both sides and declines to adjudicate between them, which is in character for the book as a whole.

Alongside the technology, Turner spends considerable time on legal history, and this is where the book is at its most interesting. He traces the gradual acceptance of corporate personhood from the medieval universitas through the great nineteenth-century chartering acts to the modern multinational, and he does the comparative work of showing how different legal traditions have treated non-human legal subjects — Hindu temple idols in Indian jurisprudence, the Whanganui River in New Zealand, ships in admiralty law, slaves in the antebellum United States, animals in a growing cluster of contemporary jurisdictions. The point is not that any of these is a template for AI personhood; the point is that the law has repeatedly invented new categories of legal subject when it needed to, and that the question for AI is not whether such invention is conceivable but whether it is useful. Turner’s own answer leans sceptical in the short term — he thinks electronic personhood is likely to be solving the wrong problem, and that insurance pools and strict-liability regimes do the work people reach for personhood to do — but he treats the question with the seriousness it deserves rather than dismissing it.

The chapter on writing a moral code is where the book pivots from description to prescription. Turner surveys the ethical frameworks that AI designers have in fact been drawing on — the Asilomar Principles, the IEEE’s Ethically Aligned Design, successive corporate codes from Google and Microsoft, the OECD guidelines — and he observes something that has become more conspicuous since publication: that none of these documents resolves the trade-offs that AI systems make every time they run. Fairness, accuracy, privacy and interpretability conflict in specific ways that depend on the deployment; abstract principles cannot resolve them, and nor can trolley-problem thought experiments, which Turner treats as the illustrative devices they are rather than as guides to actual policy. His own proposal is procedural rather than substantive: codes should be written through a defined process, with specified stakeholders, with mechanisms for revision, and with the understanding that a code is a document that lives alongside the system it governs rather than a once-settled constitution.

Reception since the book appeared has clustered around a few points. Legal scholars have generally welcomed the clarity of Turner’s framing while pushing back on specific claims. Reviewers in law journals pointed out that his treatment of criminal responsibility underweights the problem of mens rea for distributed systems — the question of what intent means when no one person held the mental state that led to the harm — and that his sketch of a regulator leaves many of the hard institutional questions to be worked out. The debate about electronic personhood that Turner joined has moved on, generally in his direction: the European Union’s AI Act, which took shape after the book’s publication and reached final text in 2024, abandoned personhood as a route and settled for a risk-based regulatory framework with product-liability updates alongside it. The question of AI responsibility continues to divide courts, with the first wave of autonomous-vehicle liability cases working through tort doctrine much as Turner predicted, and the growing use of generative models in professional work raising new variations on the same themes. Robot Rules did not resolve any of these debates, but it stated them cleanly enough that later work could begin further on in the argument.

For a reader navigating a shelf on AI and its governance today, Robot Rules sits in the lawyer’s quadrant of that shelf. It pairs naturally with Ryan Calo’s work on robotics and the law, with Virginia Dignum’s Responsible Artificial Intelligence for the practitioner’s angle on ethical design, and with Mireille Hildebrandt’s Smart Technologies and the End(s) of Law for a deeper theoretical treatment of the questions Turner raises. For a reader whose interest is specifically in autonomous weapons and the law of armed conflict, Paul Scharre’s Army of None covers the ground more comprehensively, and the activist literature around the Campaign to Stop Killer Robots offers the case for a treaty-level ban; Turner’s chapter is a useful short orientation but not a substitute. Where his book earns its place is in showing a lawyer’s instinct for sequencing — what has to be decided first, what can wait, what can be left to case law — and in doing so without the breathlessness that marked many of the AI governance books published around it.

The parts of Robot Rules that look most durable five years on are the ones Turner wrote with the most care: the comparative-law chapters on personhood, the procedural chapter on moral codes, the diagnostic argument that stretched-existing-law is a stopgap rather than a solution. The parts that have aged less well are the ones that depend on the state of the technology at the time of writing, which is unavoidable in any book of this kind — the specific capabilities and limits of the systems Turner treats as cutting edge in 2018 look modest against the generative models that reshaped the conversation from 2022 onward. The regulatory landscape has also moved. The European Union’s AI Act is now law in a way it was not when Turner wrote, and the risk-tiered approach that Act adopted has become the default template that governments outside Europe are either copying or arguing against. A reader coming to Robot Rules today reads it less as a map of the current terrain and more as a well-made account of how the terrain was first surveyed — a snapshot of the moment when AI regulation was still a set of first-principles questions, before the first-generation statutes made them something else.