Artificial life: a constructive lower bound for artificial intelligence
I f you were at some of the first, small meetings of the IJCAI or AAA1 conferences, you probably remember the excitement that filled those events. There was hope that artificial intelligence might bring novel computational tools to bear on some of the fundamental questions that have dogged mankind: What is a mind? How do we see and run? How do we think differently from one another? Another set of questions has vexed philosophers and scientists for just as long: What is life? How has life
... How has life evolved on this planet? How would we recognize living systems on other planets? Computational investigations of such questions have been labeled "artificial life." ALife's goal is to abstract the "logical form" of life, independent of the particulars of the carbon-based biological life (BLife) forms that arose on this planet and with which biology is almost exclusively concerned. Some similarities between AI and ALife are fairly obvious. The first two ALife conferences (ALife-l and ALife-2) had all the enthusiasm of the early AI meetings and were attended by a similarly eclectic assortment of mathematicians, physicists, biologists, AI researchers, educators, computer security analysts, artists, comedians and philosophers. (See pp. 53-59 for a report on ALife-2.) Both AI and ALife address ancient, ambitious questions. Both depend on the computer as the tool to provide answers. ALife even has its own version of the "strong AI versus weak AI" dichotomy: If we successfully model a living system on a computer, have we created life? If so, is it then murder to terminate the program? Probably not, since the dictionary reserves the term "murder" for the malicious death of people; but there are many other troubling ethical implications of a strong ALife position that cannot be discounted so easily. Even the labels "artificial intelligence" and "artificial life" say a great deal about the fields' similarities. Both disciplines reconsider conventional notions by removing them from their "natural" settings; unfortunately, this means that neither AI nor ALife can be defined simply. And while these labels have successfully tapped raw enthusiasm (especially among students and the popular press), AI's history has proven the difficulty of building a solid body of scientific work under such an amorphous rubric. Like AI, ALife is still probably best defined by extension; for example, in terms of all the research projects reported at ALife-1 and -2. But even at this early date, several themes are beginning to emerge as common threads throughout much ALife work. 8 0885/9000/91/0200-0008 $1.00 0 1991 IEEE IEEE EXPERT Authorized licensed use limited to: Univ of Texas Pan American. Downloaded on February 1, 2010 at 15:09 from IEEE Xplore. Restrictions apply.