...Moral Machines represents a valuable addition to, and extension of, the current literature on machine morality. As the development of autonomous artificial moral agents becomes closer to being realized, I suspect that this book will only gain in importance.
URL for review: http://metapsychology.mentalhelp.net/poc/view_doc.php?type=book&id=4873
Review - Moral Machines
Teaching Robots Right from Wrong
by Wendell Wallach and Colin Allen
Oxford University Press, 2008
Review by Ryan Tonkens
Apr 28th 2009 (Volume 13, Issue 18)
Moral Machines: Teaching Robots Right from Wrong is the first book-length discussion of issues arising in the nascent field of Machine Ethics, offered by two of its more veteran thinkers. The authors do an admirable job at using language accessible to an interdisciplinary audience, which also makes the book open to a more general public readership. It will be of interest to anyone concerned with the ethical, social, and engineering issues that accompany the quest to develop machines that can act autonomously out in the world.
As a response to the expanding (and seemingly limitless) scope of artificial intelligence and robotics research, a surge of recent work has focused on issues related to the development of artificial moral agents (AMAs) (or moral machines or ethical robots). These robots will (or in certain cases already do) have the capacity to perform ethically relevant actions out in the world, in varying ways and with varying degrees of autonomy. As the capacities of such robots increase, so too should our demand that such machines act ethically. The cutting-edge discipline of Machine Ethics--made up of engineers, artificial intelligence researchers, and philosophers--is important because it investigates whether or not the development of AMAs is possible (and desirable), and helps us to prepare just in case it is.
The main themes of Moral Machines are twofold: An examination of the motivations we have for creating AMAs and how we should go about developing machines that behave ethically. Each chapter of the book focuses on certain specific issues that need to be attended to if the project of Machine Ethics is to be successful. Some of the more noteworthy questions posed by Wallach and Allen include: 'Is machine morality necessary?', 'Can robots be moral?', 'Does humanity want machines making moral decisions?', 'What are the roles of engineers and philosophers in the design of AMAs?', 'What methods and moral frameworks are best suited for the design of AMAs?', and 'How can machine morality inform human morality?'. Through their attempt to answer these questions, the authors offer a detailed and thorough survey of the relevant research being done on machine morality, and offer preliminary (and often quite insightful) answers to these and other questions (although they humbly admit that much more work needs to be done in the future).
The authors also make some more substantial claims about how ethics could be implemented into machines. For example, after discussing the benefits and shortcomings of both top-down (rule-based) and bottom-up (evolution- or learning-based) approaches to the design of moral robots, the authors spend some time arguing for a hybrid approach (Ch. 8 and Ch. 11). One example suggested by the authors is an approach that appeals to a virtue ethical framework, since virtue ethics focuses on virtuous character traits which are acquired through training and habit formation (and hence may accommodate both top-down and bottom-up computational approaches). The authors argue that a hybrid approach holds much promise for overcoming the problems associated with pure top-down and bottom-up approaches to implementing ethics into machines. This proposal has some initial appeal and plausibility, and warrants the attention of further research.
Despite the value of the book as a whole, a few critical notes are worth mentioning. For one thing, although the authors touch upon issues surrounding the nature of moral agency, they do so only somewhat superficially, leaving many of the more complex and important issues unattended (and unresolved). For example, there is a rich debate over whether or not consciousness is a necessary condition for being a moral agent, and, if so, whether robots could be sufficiently conscious so as to possess moral agency (akin to humans, perhaps). Although the authors do mention the issue of machine consciousness (and moral agency in general), they do so only in passing (Ch. 4).
Furthermore, although the authors discuss the relationship between ethics and engineering, and the different (and often conflicting) roles of ethicists and engineers, the authors seem to champion the task of the engineer. In other words, although the book is devoted to the topic of machine morality, the authors focus primarily on the design, implementation, and engineering aspects of creating AMAs, with the consequence of leaving other (ethical) issues by the wayside. For example, in their discussion of which sort of ethic we should implement into machines, the authors focus on which frameworks work best in terms of their computability or implementability. There is no doubt that this issue is important. Yet certain ethical questions may demand attention, prior to the implementation stage. For example, whether the moral codes we are trying to implement into our machines allow for the development of those types of machines is never asked. Moreover, from an engineering perspective, the moral frameworks appealed to for designing AMAs are assessed based solely on whether they are conducive to implementability. Yet, ethicists may be reluctant to accept that all (or most) moral frameworks start on an even playing field, the problem simply being a matter of which frameworks are most conducive to implementation. Some discussion of the longstanding debates in Ethics between competing moral frameworks may be necessary here. Although the authors argue for a hybrid approach to designing AMAs, perhaps one that adopts a virtue-based moral framework, they do not ask whether we would want our machines to be virtuous, in the sense that virtue ethics is the best moral framework on offer (as compared to duty-based or consequentialist ethics, for example).
Despite these unattended issues, Moral Machines represents a valuable addition to, and extension of, the current literature on machine morality. As the development of autonomous artificial moral agents becomes closer to being realized, I suspect that this book will only gain in importance.
© 2009 Ryan Tonkens
Ryan Tonkens, Department of Philosophy, York University, Toronto, Canada