BRAIN AND COGNITION 34, 1–4 (1997)ARTICLE NO. BR970903

INTRODUCTION

Artificial Life: A Bridge toward a New Artificial Intelligence

Artificial Life (AL) has recently emerged as a confluence of very differentfields in order to study the kinds of phenomena that characterize living sys-tems from a perspective not attached to their local, temporal, or materialcontingencies. Examples of those phenomena are prebiotic systems and theirevolution, growth and development, self-reproduction, adaptation to chang-ing environments, evolution of ecosystems and natural selection, formationof sensory-motor systems, and autonomous mobile systems. Ultimately thereis an explicit purpose of producing organisms by artificial means, and theinherent difficulties of achieving such a goal with material components havenaturally brought in the computer as the privileged modeling tool of AL,and so the AL community is currently trying to develop a research programbased mainly (though not only) on computer simulations. As long as theappropriate interpretation procedures are adopted and awareness of their lim-itations is maintained, these simulations are very useful in the explorationof the limits of living phenomena.

There are important parallels between AL and the discipline known sincethe 1950s as Artificial Intelligence (AI). Both try to address extraordinarilycomplex phenomena for which classical scientific approaches are far fromproviding a unified and universal theoretical description, mainly becausetheir complexity is apparently irreducible to lower-level domains. Moreover,although life and intelligence are phenomena we intuitively know—or webelieve we do—we cannot forget that we don’t have unanimously accepteddefinitions for them within the biology and psychology communities. Giventhis situation, both AI and AL share a basic functionalist strategy: to under-take a computational study of the formal organization of life and intelligence,insofar as they can be realized in nonspecific material systems.

Despite these important similarities, the differences between AI and ALare strong and significant. AL probably owes its name (obviously inspiredby AI’s) to the desire to emphasize its differences from AI rather than to anattempt to suggest any kind of collusion with it. As a matter of fact, weseldom find in the literature an argument in favor of the technological, meth-odological, or epistemological proposals made by AL research program that

10278-2626/97 $25.00

Copyright 1997 by Academic PressAll rights of reproduction in any form reserved.

2 INTRODUCTION

don’t include a criticism—at least implicit—of the corresponding theoreticalpositions developed within AI.

Why is this ‘‘competitive dimension’’ between two sciences that shareimportant principles and presumably should not interfere in their comple-mentary scientific targets so important? The reason is that actually they dointerfere in a very important aspect: the explanation of cognitive phenomena,which lies in a problematic boundary region.

AI has traditionally used formal tools to attempt to study cognition as anabstract phenomenon, that is, as a disembodied process that can be graspedthrough formal operations, regardless of the nature of the system that dis-plays it. After almost four decades of research in this direction, the field hascome across several problems that have taken it to what many consider a‘‘dead end.’’ Why? The answer given from the AL perspective is that, asliving beings, cognitive agents are autonomous, situated, and embodied: au-tonomous, because cognitive agents are not designed to perform a task, butrather what they do (and what they are) is the product of its self-organizingprocesses; situated, because cognitive tasks are not abstract processes; andembodied, which means that cognitive processes should be considered froma threefold perspective: evolutionary (to encompass the fact that cognitionhas appeared through a long and collective process), integrated (AL triesto reproduce the whole system in which intelligent behavior occurs), andfunctional.

That is why the radically bottom-up and biologically grounded approachof AL hopes to contribute decisively to understanding cognition and to de-signing artificial cognitive agents, starting from its simplest instances butwith no a priori limitation. As Richard Belew says, ‘‘AL is a lower boundfor AI, (because) the dumbest smart thing you can do is stay alive.’’ Con-versely, it is obvious that intelligence is one of the most elaborate strategiesfor survival.

Thus, in addition to the more-or-less narrow scope devoted to phenomenabeing currently investigated by AL researchers, there is the far-reaching chal-lenge of settling the background for developing a consistent theory of allpossible forms of cognition, including its most complex instances in theframe of biological evolution, consequently giving birth to a new ArtificialIntelligence with entirely different conceptual foundations.

The idea of this Special Issue of Brain & Cognition was precisely inspiredby a workshop entitled ‘‘Artificial Life: A Bridge toward a New ArtificialIntelligence,’’ held in Donostia (San Sebastian, Spain) in December 1993.We invited a group of specialists (Mark Bedeau, Stevan Harnad, Phil Hus-bands, Gyorgy Kampis, Barry McMullin, Domenico Parisi, Tim Smithers,Evan Thompson, and Francisco Varela) from different disciplines (computersciences, psychology, biology, philosophy, etc.) who were interested in acommon set of questions. Our aim was to provide a forum in which to discussand contrast with them their respective positions on topics such as the prob-

INTRODUCTION 3

lems of functionalism in Cognitive Science and AL; modeling of neural net-works through genetic algorithms, autonomy, and robotics; the crisis of therepresentational models of cognition and its consequences; the use of ALsystems as problem solvers; emergence and evolution in artificial systems;and the relationship between living and (minimal) cognitive systems. Allthese topics were summarized in one question, which was the leitmotiv ofthe workshop: What is the impact that research in Artificial Life will haveon Artificial Intelligence? During two intensive days we had the opportunityof enjoying an extraordinarily fruitful meeting, in which each participantpresented a talk on a particular problem followed by a rich debate in whichthe flow of ideas generously exceeded our hopes.

In order to prepare this Special Issue, each author was required to presentan article that would focus on one of the above-mentioned points and wouldinclude his particular position held throughout the meeting, emphasizing theinfluence of the methodology and concepts appearing in AL in the develop-ment of new ideas about cognition that could eventually give birth to a newArtificial Intelligence.

The result is this volume, which includes eight contributions in whicheach author displays his ideas from his particular scope and background.Despite the deep differences in origin and divergences in proposals amongthese papers, there is still a common substrate evidenced in the workshopthat makes a global aspect.

In the first article, Bedau addresses the question of the potentiality of theso-called emergent computation for explaining both biological and cognitivephenomena. He explains, through some computational experiments, how theAL tools can account for such complex phenomena. McMullin’s contributionis a sharp attack on two classical ideas against functionalism in strong AI.Though exhibiting personal antireductionistic convictions, he concludes thatstill we don’t have sound and reliable arguments to support the widespreadbelief of computationalism as a discredited discourse. Thompson’s paperdevelops a criticism against the symbol–matter radical dualism that underliesclassical positions of functionalism in AI and AL. He concludes that thelesson given by the simplest symbol–matter system, the living cell, is thatthe symbol grounding problem will remain unsolved unless we take the ideaof autonomy seriously. This conclusion connects with the main subject ofVarela’s article, which shows how the meaning of autonomy in biologicaland cognitive domains is based on the operational closure of the organizationof the cell and the nervous system. Smithers shows how Maturana and Va-rela’s concept of autopiesis as a theory of life in a single cell can form thebasis for a proper definition of autonomy in robots and other agents. Morenoand colleagues discuss the necessity and difficulty of a biological groundingof a far-reaching idea of cognition. Taking cognition from its origins, thispaper argues for the necessity of a radical theory of embodiment, encom-passing the individualistic view of autonomy with an evolutionary perspec-

4 INTRODUCTION

tive. Starting from a more experimental investigation, Husbands and col-leagues show how the design of complex physical autonomous agents withcognitive capacities requires an evolutionary perspective. This approach per-mits one to work with very-low-level primitives and to explore many archi-tectures and behaviors before building physical agents. Finally, Parisi’s paperis a claim against the methodologies which pretend to limit the applicationof AL techniques and methodology to the lower-level processes underlyingcognition. On the contrary, he brings about arguments, experimental evi-dence, and prospectives in order to advocate an integrated approach up tothe most specialized and evolved cognitive functions.

The publication of the present issue has been the result of the collaborationof many people. First of all we want to thank to the authors for their enthusi-astic commitment to the project, which began with their valuable participa-tion at the workshop. Harry Whitaker’s advice was particularly decisive. Hehas been patient and helpful with the guest editors. We owe special thanksto the referees, who anonymously reviewed the articles of this issue. Theworkshop held at Donostia-San Sebastian owes part of its success to theconstant work of the members of the organizing committee, Blanca Cases,Arantza Etxeberria, Mari Jose Garcia and especially Julio Fernandez, andthose who sponsored the meeting (the Spanish Ministerio de Educacion,Basque Government, Kutxa, and Diputacion de Guipuzcoa).

Finally we make special mention of the continuous support provided byour home institution, the Euskal Herriko Unibertsitatea (University of theBasque Country), which has financed a number of research projects that havemade possible the preparation of this Special Issue.

Alvaro MorenoJesus Ibanez