New Laws of Robotics

Frank Pasquale wrote New Laws of Robotics in the shadow of two prior debates he had helped shape. The first was the argument he made in The Black Box Society, published five years earlier, that opaque algorithms were quietly rewriting the rules of finance, search and reputation with almost no democratic oversight. The second was the wave of anxiety, cresting around 2018 and 2019, that autonomous systems were about to displace not just factory workers but doctors, teachers, lawyers, judges and soldiers. As a law professor working at the intersection of information technology and political economy — first at Maryland, then at Brooklyn — Pasquale was positioned to write about that anxiety without belonging to either of the loudest camps. He was not a Silicon Valley futurist, and he was not a Luddite. He was interested in what a serious body of law, professional ethics and industrial policy would look like if governments took the arrival of AI seriously enough to steer it rather than simply react to it. The book landed with Harvard University Press in 2020, into a policy environment already changed by the pandemic, in which the limits of automated triage systems, remote proctoring software and content-moderation algorithms had suddenly become mainstream news.

The central argument is announced in the title. Isaac Asimov’s three laws of robotics — do not harm humans, obey humans, protect yourself — were fiction, and Pasquale treats them as a starting point rather than a doctrine. He proposes four new laws, meant to speak to the actual regulatory questions raised by real robots and real machine-learning systems. Robotic systems and artificial intelligence should complement professionals rather than replace them. They should not counterfeit humanity. They should not intensify zero-sum arms races. And they must always carry the identity of their creator, controller and owner. Each of those is written as a design constraint and as a policy demand, and each is defended over the course of the book with a specific domain in mind. Behind the four laws lies a broader thesis: that the choice between full automation and human agency is not a technical inevitability but a political one, made by governments, procurement officers, professional bodies and unions, and that the reflex to defer to whichever technology is cheapest today is a form of surrender. Pasquale wants readers to see AI as a set of tools that can be shaped by law and by expertise, not as an autonomous force to which the working world must adapt.

The book is organised around the professions and institutions Pasquale believes matter most for the shape of the coming decade. An opening chapter lays out the four laws and their intellectual lineage, drawing on the sociology of professions, on labour economics and on the history of prior technological transitions. The next chapters take each major domain in turn. There is a long chapter on medicine and health care, which examines diagnostic AI, robotic surgery, telemedicine platforms and the growing tendency of insurers to substitute algorithmic scoring for clinical judgement. Another chapter focuses on education, where Pasquale surveys automated tutoring, essay-scoring software, MOOCs and the surveillance infrastructure of remote learning. A chapter on media and journalism examines algorithmic content ranking, automated news generation and the concentration of attention markets in a handful of platforms. A chapter on policing and armed force covers predictive policing, facial recognition, autonomous vehicles used by public authorities and, most directly for this field, autonomous weapons. Later chapters step back to look at the macroeconomic picture: the labour market consequences of accelerated automation, the political economy of platforms, the future of public services, and the question of what a democratic industrial policy for AI would actually contain. A concluding section returns to the four laws and to the international dimension — trade, standards, arms control — that Pasquale sees as inseparable from any national framework.

The concrete cases carry most of the argument. In medicine, Pasquale spends time on the marketing of IBM Watson for oncology, on the failures of that programme to live up to its early promotional claims, and on the slower, quieter integration of narrower diagnostic tools into radiology and pathology. He is careful to distinguish between systems that genuinely widen the horizon of what a doctor can see — pattern-recognition tools trained on very large image sets — and systems that are essentially cost-cutting devices dressed up as clinical decision support. His reading of health-insurance algorithms is closer to the second: he describes how prior-authorisation software and utilisation-review scoring have been used in the United States to deny care in ways that would be professionally indefensible if a human clinician were making the decision on the record. He argues that when a machine functions as a way to launder responsibility away from any identifiable human, the first and fourth laws are both violated at once. In education, he uses the example of essay-scoring engines that reward specific stylistic markers and can be gamed by students who realise what those markers are; he treats this not just as an accuracy problem but as an educational one, in which the presence of the machine reshapes what teachers teach and what students learn to produce. His discussion of remote proctoring — the use of webcams, eye-tracking and behavioural classifiers to invigilate exams — is more sceptical than most of the pre-pandemic literature, and it aged well as universities discovered during 2020 how badly those systems handled non-white faces, disability accommodations and modest home environments.

The chapter on media examines the shift from human editorial judgement to algorithmic ranking, and it draws on the by-then familiar catalogue of failures at Facebook, YouTube and Twitter, but Pasquale is more interested in the structural point than in any single scandal. He argues that the substitution of engagement metrics for professional editorial norms is a form of counterfeiting: a feed that looks like a newspaper front page is not a newspaper front page, because no one has taken responsibility for its contents in the way an editor once did. Automated news writing, of the sort pioneered by wire services for corporate earnings and sports scores, receives a more nuanced treatment; Pasquale reads it as an example of the first law working correctly, where machines take on the drudge tasks that free reporters to do things only humans can. The distinction between complementary automation and substitutive automation runs through every chapter and does most of the analytical work.

For readers coming to the book from the AI-in-warfare beat, the fifth chapter on force and coercion is the centre of gravity. Pasquale surveys predictive policing tools such as PredPol and the various risk-assessment scoring systems used in bail and sentencing, and he places them alongside the emergence of loitering munitions, autonomous drones and command-and-control software for the battlefield. He engages with the arguments of the Campaign to Stop Killer Robots and with the diplomatic conversation that had been running for years inside the United Nations Convention on Certain Conventional Weapons, and he takes seriously the counterargument, associated with Ronald Arkin and others, that a well-designed autonomous system might in principle apply the rules of international humanitarian law more consistently than a stressed and frightened soldier. Pasquale is not persuaded, and his argument is characteristic. He grants the theoretical possibility but treats it as the wrong question. The relevant question, for him, is what happens to the political and legal accountability structures that constrain the use of force when the actual killing is done by systems whose behaviour cannot be explained after the fact and whose builders, buyers and operators can each point at one another. His third law — do not intensify zero-sum arms races — is aimed squarely at this problem. He argues that the logic of autonomous weapons is inherently escalatory: a system that can react faster than a human forces its adversary to build a system that reacts faster still, and the political space for restraint shrinks with each cycle. He calls for arms-control instruments modelled loosely on the treaties that constrained chemical, biological and blinding-laser weapons, and he treats the absence of such instruments as a failure of political will rather than a technical impossibility.

The economic chapters step back from the professions to ask what happens to labour markets and public budgets when systems of the kind he has been describing spread across them. Pasquale is sceptical of the two dominant framings, one being the promise of a universal basic income that would compensate displaced workers, the other being the assurance that new jobs will emerge to replace the ones lost. He treats the first as a way of accepting the political disempowerment of workers in exchange for a stipend and the second as a piece of received wisdom whose empirical basis has weakened as automation has spread from routine tasks to cognitive ones. His preferred programme is a return to what he calls sectoral bargaining, professional licensing with real teeth, and public investment in the human infrastructure — nurses, teachers, social workers, care assistants — whose work is exactly the sort that cannot be automated without loss. He points at Germany and the Nordic countries as suggesting the outlines of such a settlement, though he is not naïve about the political difficulty of importing them into an American context.

The concluding sections return to the international dimension. Pasquale argues that a national AI policy is unworkable in isolation, because the incentives to defect are too strong. A country that regulates autonomous weapons will watch its adversaries build them; a country that regulates surveillance will watch its citizens’ data flow across borders to jurisdictions that do not. His proposed answer is coordination through existing multilateral bodies, augmented by new standards organisations, and by trade instruments that build regulatory floors into cross-border data flows. He is realistic about how difficult this is in a moment of great-power competition between the United States and China, and about the discomfort European policy-makers feel between the two, but he treats the alternative — an unregulated race to the bottom in each domain — as worse.

The reception of the book has been shaped by where readers sit. In the legal and policy community it has been treated as a serious contribution and cited alongside earlier work by Ryan Calo, Woodrow Hartzog and Julie Cohen on the regulation of machine agents. Reviewers in that world have generally accepted the framing of the four laws while pushing back on individual applications: some have argued that Pasquale is too pessimistic about diagnostic AI in medicine, others that he underrates the potential of algorithmic tools in access-to-justice work, and still others that his industrial-policy prescriptions understate the difficulty of reviving professional licensing regimes that have been eroded for decades. In the technology press the response has been more mixed. Writers sympathetic to a stronger regulatory state have welcomed the book; those closer to the productivity-and-growth argument for AI have accused Pasquale of romanticising the professions and of underweighting the gains automation can deliver to people who currently cannot afford expert human attention. The autonomous-weapons chapter has drawn interest from arms-control specialists but has not obviously moved the diplomatic conversation, which has continued in Geneva at its own slow pace. Since publication the surrounding landscape has changed rapidly, with the arrival of large language models, the Russian full-scale invasion of Ukraine and the resulting field experiment in drone warfare, and the passage of the European Union’s AI Act. None of those developments overturns Pasquale’s core argument, but each puts it under pressure, and readers now come to the book expecting to test its framework against events he could not have seen.

For someone reading widely on AI in warfare and adjacent fields, New Laws of Robotics fits a particular slot. It is not a survey of the technology, and readers who want that will do better with Paul Scharre’s Army of None on autonomous weapons specifically or with Kai-Fu Lee on the industrial race. It is not a first-principles ethical treatise of the sort Wendell Wallach and Colin Allen wrote earlier, nor is it a philosophy of automation in the mode of Nick Bostrom. Pasquale writes as a lawyer and a political economist, and the book pairs naturally with Shoshana Zuboff’s account of surveillance capitalism, with Virginia Eubanks on automated inequality and with Kate Crawford’s later work on the material and political costs of AI infrastructure. It is a book for readers who accept that the technology will keep improving and who want to think about the institutions that will decide what to do with it. Its treatment of autonomous weapons is a chapter rather than a book, and someone whose interest is exclusively military will need more, but the chapter earns its keep by placing armed force in the same analytical frame as the rest of the professions, which is unusual in the literature.

What is likely to age well is the framing. The four laws are elastic enough to survive new technical developments, and the demand that AI systems complement rather than replace human expertise has grown more rather than less urgent as generative models have made substitution look plausible in domains that previously felt safe. The insistence that systems must carry the identity of their controllers has become the mainstream policy position in Brussels and, in a weaker form, in Washington. What is already dated is some of the technological detail — the emphasis on rule-based expert systems in medicine, the treatment of essay scoring as the leading edge of educational AI, the assumption that autonomous drones would remain a niche procurement question rather than the dominant form of the ground war in Europe. Read now, the book works best as a framework for thinking about the trade-offs that governments and professional bodies keep having to make, one domain at a time, and as a reminder that the shape of the automated future is being decided by choices that are legal and political as much as they are technical.