Computing Revievs · May 1996


FRANKLIN, STAN (Univ. of Memphis. Mcmphis. TN)

Artificial minds.

MIT Press. Cambndge, MA. 1995. 449 pp., $30. ISBN 026206 1783.

As artificial intelligence (Al) has matured, it has provided a unified home for approaches to intelligence that manipulate symbolic xpresentations of knowledge. Research into other, nonsymbolic approaches to computational intelligence is much less accessible, scattered thmugh a dozen disciplines and hundreds of different conferences, joumals, and books. Franklin seeks to pull these scattered threads together to understand how one might implement intelligence on a machine. His approach is unabashedly infortnal and conversational. Though a mathematician and a researcher, he assumes the role of an amateur toward most of the technologies he discusses. and explicitly pt~sents the book as a tour of various research projects that he has found of interest.

The first three chapters set the tone. Chapter I, "Mechanisms of Mind," descnbes the domain that Franklin proposes to cover. He class)fies existing approaches as topdown versus bottomup and analytic versus synthetic. His focus includes AI and a bottomup synthetic group that he calls "mechanisms of mind." Chapter 2, "The Nature of Mind and the MindBody Problem," reviews mentalist, dualist, and matenalist theories of the relationship between mind and body. Franklin explicitly assumes the matenalist position without proof and seeks thmughout the book to exhibit a wide range of work that, on the matenalist assumption. is interesting and promising. The reader is also invited to consider whether the question "Can a computer have a mind?" might better be phrased. "How much of a mind can a computer have?" Chapter 3. "Animal Minds," illustrates this continuum approach by explonog how much mind can be ascnbed to various animals, based on observations of their behavior.

Franklin organizes thc body of his book around three debates conceming Al. Chapter 4 is a brief introduction to symbolic Al. Chapter S outlines "The First Al Debate": whether we can expect computers to think as humans do. Chapter 6, "Connectionism." shifts the focus from the classic Al model of mindascomputer to a model that relies on networks of computational elements that bear some formal resemblance to neurons in the brain. Chapter 7, "The Second Al Debate." reviews arguments between symbolic Al and connectionists. namely, whether a connectionist approach provides any capability beyond what can be achieved by direct symbol manipulation. At the end of these two debates, Franklin is guardedly optimistic about the possibility of machine minds, and favorably inclined toward connectionist models.

The next six chapters set the stage for the third Al debate: whether representation is necessary to implement artificial minds. Chapter 8, "Evolution. Natural and Artificial.' reviews computer models of natural selection. Chapter 9. "Artificial Life." shows how these algorithms contribute to the emergence of complexity from a community of individually simple organisms in Wilson's Animat. Chapter 10. "Multiplicity of Mind," applies this model to the mind. following suggestions that consciousness emerges from the interaction of many simple parts. Chapter 11 attempts to answer the question. "What Do I Do Now?" To decide an organism's next step, a mind needs access both to external information and to internal memories. Chapter l2. "What's Out There." expounds Freeman's model of concepts and argues that ontological categories do not exist absolutely out ;n the world, but are imposed on the world by the mind. Chapter 13 explains "Remembering and Creating." Chapter 14. "Representation and the Third Al Debate." engages the question of whether minds require explicit representations of concepts. Franklin believes that much situated activity can be done without representations, but that manipulation of abstract concepts probably requires them.

Chapter 15, "Into the Future." discusses a few leftover paradigms, including quantum models of consciousness and

Moravec's bush robot. Chapter 16. "An Emerging New Paradigm of Mind," gives Franklin's synthesis, which he teems 'the action selection paradigm of mind." "Mind" in this sense is the selection of the next action by an autonomous agent, so any agent that makes such a selection exhibits some degree of "mind." The mind creates information out of inputs. rather than processing preexisting concepts. Memories are recreated from partial clues rather than stored and retrieved as units. The mind is essentially modular, emerging from the interactions of many simple modules.

Franklin repeatedly emphasizes that he is an amateur leading a tour and not an expert. As a result. he has achieved a breadth of coverage for which an expert in a specific discipline could hardly claim credibility. and provides fellow travelers with a road map of how various approaches relate to one another and where work remains to be done. This volume will serve students and researchers alike as an entertaining and accessible pointer to the diversity of work available on artificial minds. The bibliography contains more than 250 references.

-H. Van Dyke Parunak Ann Arbor, MI