Customer Reviews

Moral Machines: Teaching Robots Right from Wrong

5 star:		(5)
4 star:		(1)
3 star:		(3)
2 star:		(0)
1 star:		(0)

 

 

Allen and Wallach's Moral Machines is the best text yet in the rapidly expanding field of robot ethics - and their work offers insight into the morals of not only robots, but ourselves as well.

Wallach and Allen examine the strengths and limitations of traditional approaches to ethics, such as deontology and utilitarianism, and the issues that arise in attempting a top-down programming of such rules into a robot. But the history of ethics is replete with controversy over the adequacy of any proposed set of rules - for instance, it might seem logical to switch the track of a runaway trolley that would kill five workers, even if it would thereby kill one person on the other track - switching maximizes utility. But should a doctor then harvest organs from a patient in for a checkup to save five people in the next room needing transplants?

So what should a robot do? An alternative is to attempt a 'bottom up' approach, and teach ethics to robots by trial and error, as we do children. The authors argue that this approach has both technical and rational limitations as well; principles are especially useful in resolving the difficult moral situations we call moral dilemmas. So they argue that a hybrid approach is probably best, and discuss in thought-provoking ways whether robots would need emotions, and how human-like we should desire these robotic agents to be.

Wallach and Allen convincingly argue that even if full moral agency for machines is a long way off, it is already necessary to start instilling into robots a type of functional morality, as robots are already engaged in high-risk situations and are already equipped with lethal weapons (e.g., the Predator drones now flying in Pakistan).

The text is anchored in near-term considerations and hence is light on some of the more far-reaching aspects of robot ethics - for instance, if full human-type ('Kantian') autonomy for robots is possible, should it be allowed? Or should robots be forever relegated to a 'slave morality', so they could never ultimately choose their own life's goals - lest they be harmful to humans? But the failure to engage in these more long-term debates simply underlines the near-term strengths of this text. For those wondering (or worried) about moral questions involving robots over the next decade, this is a must-read.

P.S. They also have a nice blog with updates: [...]

This book combines the ideas of leading commentators on ethics, methods of implementing AI, and the risks of AI, into a set of ideas on how machines ought to achieve ethical behavior.

The book mostly provides an accurate survey of what those commentators agree and disagree about. But there's enough disagreement that we need some insights into which views are correct (especially about theories of ethics) in order to produce useful advice to AI designers, and the authors don't have those kinds of insights.

The book focuses more on near term risks of software that is much less intelligent than humans, and is complacent about the risks of superhuman AI.

The implications of superhuman AIs for theories of ethics ought to illuminate flaws in them that aren't obvious when considering purely human-level intelligence. For example, they mention an argument that any AI would value humans for their diversity of ideas, which would help AIs to search the space of possible ideas. This seems to have serious problems, such as what stops an AI from fiddling with human minds to increase their diversity? Yet the authors are too focused on human-like minds to imagine an intelligence which would do that.

Their discussion of the advocates friendly AI seems a bit confused. The authors wonder if those advocates are trying to quell apprehension about AI risks, when I've observed pretty consistent efforts by those advocates to create apprehension among AI researchers.

Although this book is accessible to a popular audience, it has obvious academic merit. The authors thoroughly search-out all perspectives in this new field (i.e. it has a huge bibliography) and treat each perspective with skillful fairness. It both establishes itself as the authoritative reference, framing the issues for the new field of machine ethics, and establishes the credibility of the field as an academic pursuit. Good libraries ought to have this book.

This book was not intended as an introduction to ethics, but it is the book I would be inclined to assign as an ethics textbook. It covers an introduction to ethics, of course, but also covers material in related disciplines (psychology, economics, etc.), and gets technical about where our society assumes ethical faculties. It forces the reader to think about how ethics work, rather than just express opinions about contemporary moral issues, and is probably the very best book in existence for giving readers an appreciation for the ways the field of ethics will have to grow in the near future.

I totally enjoyed this reading, though the writing style could be more lively and engaging.
Most of this is very speculative, on the edge of philosophy and science fiction, but possible in the next 50-100 years. My only objection is that I desired more technical information on the robots, or at least their computer capabilities. I am not a tech person, but you should not write a book on robots without more concrete details, if at all possible. Also, I am very surprised and disappointed to find oonly one poor photo of a robot here. Where are all the good photos or illustrations? Well, try YOUTUBE because none are here.
`
Instead readers get much information on ethical theory (such as Kant and Aristotle) and how that might apply to robots. Interesting enough and well explained, but someone totally new to philosophy may have trouble understanding that. It was easy for me with my advanced background. Useful for a course on philosophy of technology and just good reading. A long bibliography is included and much research was done.

What strikes one most reading the book is how it informs us not merely about the challenges of creating a robot or program with moral or ethical reasoning. The subtext really goes much further. Namely, what is it that actually informs morals and ethics for humanity, both for us personally and as a race. The book provides an insightful overview of the challenges scientists face in defining and in embuing a machine with the capacity for moral reasoning. That challenge is amplifed by the obvious question, what is it that gives us our capacity for the very same thing? As the reader is taken across the landscape of the current and future thinking and development of moral reasoning in machines, we begin to understand the problem is so complex and difficult because it is likewise difficult for humanity to agree upon what it is the allows us to make moral decisions. It may be that like the famous insight about pornography, "I can't define it but I know it when I see it". The authors take us on a journey that helps us see why defining it is so challenging.

All that said the main thrust also take us deeply into the exciting developments in the developments of Artificial Intelligence, from the actual to the possibilities that science fiction has presented. This will be a great resource for any reader to begin an in depth look at this facinating area.

Moral Machines: Teaching Robots Right from Wrong
We have been in uncharted waters for at least 94 years since 1914.
Some few gifted observers have tried to explain the past, clarify the present, and glimpse the future.
Wendell Wallach and Colin Allen have succeeded in giving us a cautionary and yet hopeful view of a future world that we are likely to be sharing with increasingly intelligent computers and their active agents...robots.
Can we form a reasonably secure community together and, if so, how can we go about achieving it.
Here in this volume in both an entertaining and highly informative manner, Wendell Wallach and Colin Allen have given us the framework for understanding the challenge.
Howard G Iger, MD

A friend gave me Moral Machines. I didn't know an AMA (Autonomous Moral Agent) from a CPA. With each well-argued case, I came to recognize how essential it is to understand, and take an active stance, on the fast-coming day of robots on the battlefield, in my bloodstream, and mining my internet messages. Professors Wallach and Allen have written a treasure-trove volume, wide in research and modest in editorial style. It will surely show up on university book lists for new-science courses and, more important to me, it should be the culminating text in every course on the history of ethics. For, as we humans try to implant moral guiding mechanisms in machines, we are forced to learn the essence of our own human moral guidance.

Having spent part of my carrier as a systems engineer in developing robotics, I was interested in where AI has come in the last 10 years or so, and this book seemed just the ticket. I was hoping the authors, being philosophers, would provide useful insights into implementable ethics and morality, and being in AI would relate well to the science and engineering of robots. What I got instead was very high level, vague, descriptions of the philosophical approaches with tenuous tie-ins to robotic systems that were also described (at best) at the early block diagram level. Vague philosophical generalities pervade the whole book - I have very little idea of any actual progress that has been made in the field from this book, nor do I have any idea of which moral/ethical ideas are easy/hard to implement nor what (engineering) approaches are being taken to do so. So if you are a philosopher and/or a consultant, this may be just the book for you. If you are interested in the realistic progress of pragmatic AI, my advise is try another book.

Wallach and Allen explore ways in which machines might emulate human moral decision making. But human morality is badly flawed and what is really needed is machines which have a morality which surpasses the merely human. (Theories of Value, its Origin, and Value Change, Trans. Kansas Acad. of Science, vol. 109, pg 254, 2006 R. Jones and Asa H: A hierarchical architecture for software agents, Trans. Kansas Acad. of Science, vol. 109, pg 159, 2006 R. Jones) Evolution has imposed on humans and other animals an economic utility (fitness) approximated by U=(N-2)/L where N is the number of offspring a pair of mammals has and L is the animal's lifespan. It is difficult for a creature to decompose this utility U into a judgment about any particular action. In animals evolution has hardwired in a set of heuristics (drives, aversions, etc.) which perform this decomposition (i.e. pain, pleasure, sex drive, hunger, thirst, discomfort, innate fears, sickness, loneliness, curiosity, etc.) Note that all of these produce a much more immediate reward (feedback) than U, N, or L can. It is, however, possible to give artificial intelligences a BETTER value system than this. (see my 2 references cited above and also: An autonomous software agent for industrial process control, Trans. Kansas Acad. of Science, vol. 107, pg 32, 2004 R. Jones)