I’d like to begin with the words of Gore Vidal, who said, “Every time a friend succeeds, a little something in me dies”. Well, quite a big something in me dies with the publication of A Legal Theory for Autonomous Artificial Agents, since it marks an enormous success for, not one, but two, old friends, Samir and Laurence. What makes it especially galling is the fact that they met through me (about 15 years ago when I was living in New York). Since then they’ve collaborated successfully on a number of projects, culminating in this book. Not once have they asked me to join in on any of these projects. So when Laurence naively asked me to speak here today, I started to plan my revenge. I also read the book, and decided that I’d have to postpone my revenge for another day, because, much as I hate to admit it, it really is a wonderful book, and it entirely vindicates their decision to exclude me (or their failure to even think of including me).
The first thing to be said about it is that it’s remarkably readable. This is remarkable, because this is after all both an academic book and a law book, yet it’s almost entirely free of both academese and legalese. It does, of course, contain some technical vocabulary, and even a bit of Latin, but these things are always explained clearly in plain English. What’s more the writing style is lucid and engaging, though I have to admit I found Chapter 2, which is on contracts, a bit heavy going. It reminded me of why I dropped out of law. But anyway that’s my problem. I’ll deal with it.
The other great thing about this book is its subject-matter, which achieves the trifecta of being extremely original, extremely interesting, and extremely important. This book opens up a genuinely new field of research, something that tends to be discouraged by academic publishers, who try to fit everything into pre-established categories. In this case, going by the blurb on the back cover, the publishers have clearly decided that this is yet another book on Artificial Intelligence (or AI). This misrepresents the contents of the book, and doesn’t do justice to its originality or its importance. This is a book about artificial agency, rather than artificial intelligence. As the authors say (on page 28) “we deprecate the terms intelligent agent or artificial intelligence as we wish to emphasize the embedded, social, real-world nature of artificial agents, rather than merely their disembodied intelligence”. This distinction is important. Although this book is in part a piece of speculative legal theory about how the law can and should respond to anticipated technological developments, it’s also about the here and now. It makes the case, very persuasively in my opinion, that (at least partially) autonomous artificial agents are already with us, and that the law needs to catch up with this fact. What is more it makes a wide variety of always interesting, and often compelling, suggestions about how that might happen, at the core of which is the recommendation that artificial agents should (at least in certain circumstances) be treated as legal agents.
For me, the most interesting chapters of this book are 3, 4, and 5. The subject matter is law. But profound philosophical issues lie at the heart of each of them, and the authors’ combination of legal knowledge and philosophical acumen is very impressive.
Chapter three addresses questions about knowledge and artificial agents. In particular it address two questions - “In what circumstances should we attribute knowledge to artificial agents?” and “In what circumstances should we attribute the knowledge of an artificial agent to its principal (that is the person on behalf of whom the artificial agent is operating).” The authors offer an analysis which draws on ancient philosophical debates about the nature of knowledge, as well contemporary debates about the practice of law. As the authors say (on page 75), they are engaged in “a delicate balancing act” in trying to devise an analysis of knowledge for artificial agents that meshes with intuitions, while not introducing undue complications to the law”. I won’t go into the details of the analysis they come up with (which is presented on page 76), but it seems to me that it succeeds in balancing these competing desiderata extremely well.
Not content with theoretical analysis, Samir and Laurence go on apply their analysis to a variety of practical legal issues. For me, the most interesting of these applications was their discussion of whether email filters can literally be said to read email, and whether companies like Google can violate their customer’s privacy by acquiring and using information when no humans have access to that information.
Chapter Four is about Tort Law. Here the authors draw on philosophical debates about the nature of causation to discuss the circumstances in which AAs should be held legally responsible for harms they cause, and the circumstances in which holding them responsible would “break the chain of causation” so as to alleviate responsibility from the designer, operator or owner of the AA.
Just to give you a feeling for the issues at stake here, my son, who is 8 years old and making a robot out of cardboard at school, asked me the other day what would happen if his robot (whom he calls Hymie) killed someone. Would he go to jail, would Hymie go to jail, would they both go to jail, or would neither of them go to jail? Well, I was disappointed not to find the answer to my son’s question in this book, because there is no chapter on criminal law. Nonetheless a variety of possible approaches to this question are implied by the book’s discussion of tort law, and I hope that the authors will be able to answer my son’s questions in a subsequent edition (and they’d better be quick about it because Hymie is about to become fully operational).
The fifth and final chapter is about whether, and in what circumstances, the law should treat artificial agents as people. It draws on longstanding philosophical debates about the distinction (or alleged distinction) between the concept of a person and that of a human being. It also draws on the fascinating history of the evolution of the legal concept of a person and the distinction between dependent and independent personhood. The authors make it clear that the personhood of AAs need not be an all or nothing matter. AAs may be treated (as corporations and a host of other entities have been treated) as legal persons for some purposes, but not for others. This chapter is important, not so much for the conclusions it reaches, as for the questions it raises. The authors do not offer (or attempt to offer) any precise criteria AAs would have to meet in order to be treated as legal persons, nonetheless they do make it clear that there can be no good a priori reason to rule out the possibility or desirability of ever treating AAs as people. Their book concludes with the following words
(Insert quote from final paragraph on page 191.)
And on that wise note, I’d like to congratulate Samir and Laurence and thank Laurence for inviting us all here to celebrate this wonderful book.