Tè
Darwiniano
Tè
Darwiniano
In occasione del tredicesimo incontro del Tè Darwiniano (presenti: Laura Benigni, Sergio Benvenuto, Anna Borghi, Cristina Burani, Raffaele Calabretta, Silvia Caravita, Federico Cecconi, Daniele Denaro, Andrea Di Ferdinando, Riccardo Galbiati, Alessandro Laudanna, Davide Marocco, Francesco Natale, Stefano Nolfi), Domenico Parisi ha introdotto il tè...ma:
Artificial Life, and the Artificial Life community (the journal, the US and European conferences, etc.), have been in existence for more than a decade. Because of this, and because Artificial Life VII will take place in year 2000, it is seen as appropriate to identify and discuss a number of open general issues in view of the August 2000 conference in Portland. Here are some of these general issues.
1. What is Artificial Life?
We can all agree on some basic definition, e.g. the definition contained in the Artificial Life VII Call for Papers:
"Artificial Life is an interdisciplinary research enterprise investigating the fundamental properties of living systems through the simulation and synthesis of life-like processes in artificial media".
However, there are open questions and different emphases. Examples:
(a) What is a simulation (both digital and physical, e.g. a robot)?
Imagine we give the following definition of a simulation:
"A simulation is a new way of expressing a scientific theory (or model or hypothesis). Traditionally scientific theories are expressed in some symbolic medium such as ordinary language (systematized, formalized, etc.) and/or mathematical symbols. A simulation is a scientific theory expressed as a computer program. Unlike traditional scientific theories, in that they are computer programs simulations are "active" theories, that is, (a) they produce (or reproduce) the empirical facts the theory is intended to explain, and (b) they function as virtual experimental laboratories in which the researcher can control and manipulate variables and parameter values, and observe the consequences of these operations".
Simulations can be a useful addition to a scientist's toolbox in that, for example,
- they force the researcher to formulate theories and hypotheses with a higher degree of precision, detail, completeness, unambiguity, than is often the case with verbally formulated theories, because otherwise the simulation won't run in the computer or won't produce the desired results;
- they allow the researcher to verify if his/her theory or hypothesis actually implies (generates) the empirical phenomena or data it is intended to explain.
If we agree on the above definition of a simulation, the ultimate test of a simulation as a theory which must help us understand reality is that the results produced by the simulation match empirically observed phenomena and data.
Now, in many Artificial Life simulations this test is often not done. The empirical phenomena and data that constitute the external criterion for judging the quality/interest of a simulation, are either ignored or only very vaguely alluded to. For example, in a typical Artificial Life paper there are generally few references to the empirical literature.
Notice that the need to establish a link with the empirical literature is important even in those cases in which an AL simulation addresses very general issues and very general properties of phenomena.
This insufficient attention to the empirical literature and to the research issues that are of current interest to empirical researchers can have undesired consequences.
First, it can deprive AL researchers of landmarks and directions to follow that can guide their research. Hence, AL research may have a tendency to proceed in a sort of amoeba-like fashion or with too diffuse goals.
Second, AL research can elicit limited interest on the part of the empirical disciplines (real life biologists, psychologists, social scientists), whereas a strong interaction between AL and the empirical disciplines would be highly desirable for AL to get a larger audience.
This issue of the role of empirical data in AL simulations is related to AL's claim that AL investigates both real and possible phenomena. While the idea of studying possible and not only real phenomena is a very interesting one in that it can enlarge the range of evidence available to test theories and hypotheses, it still must always be shown that the study of possible phenomena actually improves our understanding of real phenomena.
(b) What are the phenomena of living systems?
The phenomena of the living world are studied by the biological disciplines, including some portions of the behavioral disciplines, e.g. ethology and brain/behavior relations. Human social, cultural, technological, and historical phenomena are studied not by the biological disciplines but by the social and historical sciences. Are these phenomena part of the phenomena studied by Artificial Life? If the answer is yes, what are the reasons for, and the consequences of, extending the Artificial Life approach to human social phenomena? For example, do we have an Artificial Life approach to human history?
2. Artificial Life applications
Artificial Life has many applications and more are likely to emerge in the near future (cf. D. Terzopoulos, Artificial Life for computer graphics, Communications of the ACM, 1999, vol. 42, n. 8; also Internet and robotics applications of AL). One can even view AL more as an applied discipline aiming at creating useful artifacts than as a basic science purporting to understand reality. But how good is to mix these two different goals? If an Artificial Life idea/realization leads to useful (e.g. economically useful) applications, is the idea/realization automatically illuminating with respect to the question how is reality? Or should an idea that makes us better understand reality automatically lead to useful applications?
AL is said to be concerned with both real and possible phenomena. But possible phenomena can be interesting to study because they extend the range of evidence to test our theories of reality (see above), or because they lead to useful applications. How much is AL driven by applications?
3. The sprouting of new, AL-related but independent, research communities
In the last decade a number of new research communities have emerged which are independent and separate from the AL community but have similar goals. Examples:
- The "social simulation" or "agent-based social simulation" community (cf. electronic journal JASS)
- The "adaptive behavior" community (SAB)
- The "evolutionary robotics" community
- The "evolution of language" community
The emergence of these new research communities shows that investigating real phenomena using simulations and theoretical approaches that are very similar to AL's approach (complex systems, individual-based models, etc.) is a productive and interesting new way of doing science. However, Artificial Life may be progressively emptied if an increasing number of important classes of phenomena are addressed by other research communities.
What should be done is to explicitly identify similarities and differences between the AL's approach to various phenomena and the approach taken by these new research communities and, if there are differences (even if only of emphasis), to discuss pro and con of the different approaches.
For example, the social simulation community appears to be less interested in the biological underpinnings of human social phenomena whereas an Artificial Life approach to these phenomena tends to be more biological. Artificial Life may prefer to simulate the behavior of individuals as controlled by a neural network (model of nervous system) whereas this behavior can be controlled by symbolically formulated rules in a social simulation approach. Or Artificial Life may be more interested in studying social phenomena in simple organisms or in simple human societies whereas the social simulation approach prefers to study modern human societies. Artificial Life may be more interested in relating cultural to biological evolution (similarities and differences, interactions, etc.) whereas social simulation tends to ignore biological evolution.
Another example: evolutionary robotics. In a sense evolutionary robotics (or, more generally, biomorphic robotics or biorobotics) is part of Artificial Life. But what are the advantages and disadvantages of using physical robots (either real physical robots or simulated ones) in physical environments (either real or simulated ones) rather than using simulated organisms that do not have a physical counterpart?
Third example: evolution of language. The evolution of language community is a restricted but active new research community. Are the simulations done within this community Artificial Life simulations? Or would Artificial Life address the evolution (and nature) of language differently?
These are questions that are not intended to identify and consolidate disciplinary boundaries for social/political reasons. Asking and trying to answer such questions serves to think more precisely about different ways of studying complicated phenomena and to explicitly discuss (and test) pro and con of different ways of approaching these phenomena.
4. Artificial Life and individual behavior
In Artificial Life there is a lot of interest for sub-organismic phenomena (at the molecular and cellular level) and at the super-organismic level (collections of organisms, population phenomena, social phenomena) but surprisingly little interest in the nervous system/behavior of single individuals. (Cf. the list of topics in the Call for Papers of Artificial Life VII). This can be due to historical/external reasons. There already exist two research communities distinct from the AL community that concern themselves with behavior and neurally-inspired models of behavior: SAB and the connectionist community. These research communities are as old as or even older than AL. They have already occupied the behavior/nervous behavior niche and AL is not trying to invade that niche.
The important question is: Would an AL approach to the study of the nervous system/behavior of individuals bring something new and interesting with respect to the SAB and standard connectionist approaches?
The answer might be yes. SAB appears to be more concerned with nonhuman behavior than with the behavior of Homo sapiens. Standard connectionism has a somewhat restricted view of behavior as controlled by the nervous system whereas AL might view behavior as caused not only by the nervous system but also by the rest of the body, by the physical and, in the case of humans, by the cultural/technological environment, and by the individual's inherited genetic information and the biological evolution that results in the genetic information inherited by the individual.
Most importantly, it would be nice to see more AL simulations of the behavior of Homo sapiens, that is, of those behavioral and cognitive capacities that tend to be typical of Homo sapiens and of how these behavioral and cognitive capacities have arisen, and when. (Of course, the evolution of language community is addressing an important part of this problem.)
5. Future of Artificial Life
Should Artificial Life become part of the various traditional disciplines (biological, behavioral, social), importing in the traditional disciplines the novelty of the artificial approach (simulations, artificial physical systems) and of the complex systems/individual-based approach, or should Artificial Life try to remain a distinct discipline (research community)? The first choice has the advantage of AL possibly having a greater influence on the traditional disciplines but perhaps the disadvantage of reducing AL's potential role in reshaping the traditional disciplinary landscape.
References