(See also: E. M. A. Ronald, M. Sipper, and M. S. Capcarrère, Testing for emergence in artificial life, in Advances in Artificial Life: Proceedings of the 5th European Conference on Artificial Life (ECAL'99), D. Floreano, J.-D. Nicoud, and F. Mondada, Eds. 1999, vol. 1674 of Lecture Notes in Artificial Intelligence, pp. 13-20, Springer-Verlag, Heidelberg.)

Hopefully, you will be enthralled enough to go and read the full paper...

Moshe's home page, Machine Nature: The Coming Age of Bio-Inspired Computing

Section 2.4: Formulating the emergence test

The difficulties we face in adopting a definition of the concept of emergence are reminiscent of the complications faced by early Artificial Intelligence (AI) researchers in defining intelligence. Nonetheless, where the equally elusive concept of intelligence is concerned, Alan Turing found a way to cut the Gordian knot, by means of an operant definition which is useful within the limited context of man-machine interaction [32]. Debate concerning the concept of intelligence is unlikely to subside in the foreseeable future, and the same, we believe, holds for emergence. We deem, however, that viewing the world through Turing-colored glasses might improve our vision as regards the concept of emergence---at least where modern-day Alife practice is concerned.

The Turing Test focuses on a human experimenter's incapacity at discerning human from machine when holding what we would now call an Internet chat session. Our emergence test centers on an observer's avowed incapacity (amazement) to reconcile his perception of an experiment in terms of a global world view with his awareness of the atomic nature of the elementary interactions.

Assume that the scientists attendant upon an Alife experiment are just two: a system designer and a system observer (both of whom can in fact be one and the same), and that the following three conditions hold:

• (i) Design. The system has been constructed by the designer, by describing local elementary interactions between components (e.g, artificial creatures and elements of the environment) in a language L₁.
• (ii) Observation. The observer is fully aware of the design, but describes global behaviors and properties of the running system, over a period of time, using a language L₂.
• (iii) Surprise. The language of design L₁ and the language of observation L₂ are distinct, and the causal link between the elementary interactions programmed in L₁ and the behaviors observed in L₂ is non-obvious to the observer---who therefore experiences surprise. In other words, there is a cognitive dissonance between the observer's mental image of the system's design stated in L₁ and his contemporaneous observation of the system's behavior stated in L₂.
  When assessing this clause of our test one should bear in mind that as human beings we are quite easily surprised (as any novice magician will attest). The question reposes rather on how evanescent the surprise effect is, i.e., how easy (or strenuous) it is for the observer to bridge the L₁--L₂ gap, thus reconciling his global view of the system with his awareness of the underlying elementary interactions. One can draw an analogy with the concept of intelligence and the Turing test: the chatty terminal might at first appear to be carrying on like an intelligent interlocutor, only to lose its ``intelligence certificate'' once the tester has pondered upon the true nature of the ongoing conversation.

The above three clauses relating design, observation, and surprise describe our conditions for diagnosing emergence, i.e., for accepting that a system is displaying emergent behavior. Some of the above points deserve further elaboration, or indeed invite debate. Before treating these issues in Section 4, we wish to demonstrate the application of our test to several cases.