This reviewer's first contact with robotic A.I. was Hans Moravec's "Mind Children" which I read in the late 1980's and which persuaded me that real A.I. was possible. Another influence was John McCrone's "Going Inside" dealing with human thought processes that seem to be reproducible at a machine level (although he didn't suggested it).

Storrs Hall in this excellent book shows how A.I. researchers lost the thread in the following decades with a fixation on coding everything and building systems that worked fine in closed environments with fixed rules (e.g. chess games) but hopelessly in unpredictable real life situations.

He concludes that robots need to learn and adapt to their environments (be autogenous) although they may have some hard wired basic abilities upon which they can develop a "self" against which to make environmental tests (i.e. increase the capability/ adaption of their "self"). Another interesting aspect of the book is his discussion from chapter 18 onwards of robotic/ A.I. ethics as applicable to this new "self" and he opts for the Boy Scout Law: "One should be trustworthy, loyal, helpful, friendly, kind, obedient, cheerful, thrifty, brave, clean and reverent", and he sees the task as, "... building a machine that understands what these qualities mean and what can we do to ensure that the machines that are built will have them."

Perhaps the author could have explored at greater length the concept of a robotic/ A.I. "self " to answer this question.

For example he says that A.I. would be rid of many human pressures like sexual jealousy, but if robotic A.I.'s adapt to different environments they will likely have differing abilities and "selves" that will differ in ability and capacity to protect the "self" (i.e. they may well be jealous and competitive if they are required to survive and adapt). Equally differing robotic "selves" may coöperate to gain a group advantage (e.g. a robotic Apollo, Aphrodite and Hephaestus or maybe the whole lot of them contributing their differing abilities).

His comparative advantage in human / robotic A.I. trade is not very convincing. We don't do a lot of trade with the great apes and we in turn may be even more distant from future autogenous A.I.'s.

Also he says that humans will have an "open source guarantee" with regard to robot/ A.I. code (i.e. they will have access to and will be able to delete undesirable variants) but this assumes 1) that they understand it and 2) that a robotic/ A.I. "self " will allow access (human or otherwise) to its code . It has invested a good deal in the evolution a viable "self" which could be put at risk with such a procedure.

Nevertheless it's a really good book, with Storrs Hall favouring good environments for autonomous learning machines and quoting the Christian Golden Rule, "Do unto others as you would have them do unto you" which seems like a good place to start with early autogenous evolving A.I.s. ( )

J. Storrs Hall wrote a very insightful summary and popularization of the AI endeavor in his book, which might as well have been titled 'Before AI' because of its concise but wonderful job in summarizing the history of the field. He is well versed not only with mainstream academic AI work, but also with the fringe AGI and Singularitarian ideas, particularly the emphases on machine ethics which academia has mostly neglected.

Aside from an overview, Hall includes, in my opinion, two very important concepts which are used throughout the book to support his thinking, and they are: autogeny and formalist float. Autogeny is vaguely similar to the ideas of recursion, self-replication, and self-organization. Formalist float is the general problem of codifying or formalizing the complexities of reality. These ideas are not new, and they can be found in many of the better books on AI, especially Hofstadter's GEB, but he does label them and treat them specifically as important concepts, which is nice to see. Though he doesn't explicitly juxtapose the two concepts, in my mind I see these two notions related to each other, at least within an engineering perspective; formalist float is a problem only when the code/formalism lacks continued autogeny, code is a frozen snapshot of a system, and if the represented system has a high degree of autogeny then formalist float will plague the formalism. The grandest scientific questions of our time and in the near future, AI included, are autogenic phenomena that, thus far, have been impervious to our intellectual probing precisely because they cause formalist float.

I had not heard of Hall before reading this book, but because of it I am now definitely an admirer of his thinking. ( )

I recommend this book if you are interested in an overview and survey of the current status of Artificial Intelligence (AI) research. The one book that I can compare it to somewhat is Moravec's Mind Children although I found Beyond AI more engaging and hopeful than that book.

What is always interesting about AI research is that it inevitably brings in a wide and eclectic variety of topics. Each AI researcher has their own particular emphasis in certain areas. The author of Beyond AI has a particular slant towards the philosophical which is not seen in many scientists. This was refreshing, and humanized the topic.

These types of books can also have a lot of free-style speculation about the "Brave New World" we can anticipate from AI. The author deftly retains his balance in this regard by utilizing various visions of science fiction writers who bring some breadth to these futuristic views, but without sacrificing the humanity.

This book isn't all philosophy and futurism, the core concepts and history of AI research is very well summarized without becoming mired in technical detail. ( )

Paraphrasing slightly from p 240: "There is a new excitement aboard the good ship Artificial Intelligence. All hands are on deck, the sails are swelling, and the flag snaps in the breeze. Clear sailing lies ahead, and anchors are aweigh." Well, maybe. Won't she have to look out for Luddite-launched torpedoes?