Generative Interactive Systems with Meta-Rules: Learning From Experience

Abstract
The paper describes recent developments in Video Constructs, the basic versions of which are generative abstract computer animations. In video constructs, the logic in the computer provides the underlying structure that leads to the form of the work. The most exciting element of the constructive video is, perhaps, the careful and very terse way in which a specification of what occurs in time is possible. The brevity of the specification is extremely important in the development of ideas. The inevitable exploration is so strongly supported by this aspect of the use of the computer that new ways of thinking about work emerge in their very construction. This has led to developments both in interaction and in changing behaviors as the result of experience.
The time-based video constructs have developed into interactive video constructs. It is not hard to understand how the structures in time can be so constructed as to react to events detected by sensor systems. A real time image analysis system is incorporated into the generative program. The behaviour of the piece, i.e. the generative path that it takes, is then reactive to what participants are doing. A video construct is searching through a set of rules and, as it does so, generating the sequence of images that form the output of the work. Each image represents the state of the search at that moment. In the earlier systems the sequence of states was entirely determined by the search strategy used by the software to explore the rules. In the interactive case, however, the search engine has available to it a stream of data that is a coded representation of the behavior of the viewer and this data modifies parameters in the search, thus leading to a sense of reaction by the system to the participant.
Because these interactive video constructs are described within the computer by a set of rules, it is possible to add meta-rules, that use the history of interactions between participants and the work to modify the generative behavior by changing the rules used, or changing which rules are used. The latest work that will be described in the paper does exactly that. By recording and analysing the interactions in real-time, the system applies meta-rules as it learns from experience about human reaction to it. The video construct changes its behavior in the light of its experience with human participants interacting with the work. Because, at its core, the work is a generative system, as it learns it changes the way that it develops rather than simply the stimulus-response rules that govern its behavior.
The learning interactive video construct is a living growing art system.