Many of the unresolved challenges in the cognitive sciences are related to creativity, the process by which new knowledge and behaviors are acquired. We may ask: how does new information come about in natural systems? What features make a system creative? And why is it so difficult to equip computers with these capacities? Combining machine learning, biology and philosophy, I look for creative processes in nature, study their theoretical underpinnings and design computer simulations of minimally complex creative systems.
Creativity is sometimes characterised as producing something out of nothing. A similar (and similarly paradoxical) feature is found in many origin of life theories. The genesis of structure and function from a previously unstructured, high-entropy reaction-diffusion system has been modeled in the Autogenic Automaton, which provides a simulated account of how self-organisation, self-mending, and selection are intertwined. Simulation experiments are reported on in a paper that is currently under review at Artificial Life.
Creating computer games that are both challenging and fun to play is in many ways a form of art. Here, the ordered chaos provided by emergent systems is used as a tool to produce interesting and diverse game worlds. In this joint work with Joris Dormans, the systems theory behind game mechanics is scientifically studied using the Machinations framework (see jorisdormans.nl/machinations). We are aiming to exploit emergence in the game world by modelling its temporal, structural and hierarchical aspects.
The Nine Dots Puzzle originally gave rise to the phrase "thinking out of the box": drawing four connected straight lines, can all dots be connected? The answer depends on a leniency to think (or in this case: draw) out of the box. When programming a brute-force solution to this problem, this leniency is necessarily hardcoded into the program - or not. In neither case, the program can be said to be creative. To illustrate, a program was created (see leijnen.org/ninedots) as part of a presentation at Google Inc.
According to philosopher Paul Feyerabend, creativity needs to be understood as arising out of a society of agents, rather than the individual. The spread of ideas through creative innovation and not-so-creative imitation has been studied using a simulation of artificial agents called EVOC. In this joint work with Liane Gabora, a diversity of dynamical patterns was investigated, ranging from innovative leadership, to the optimal innovation-to-imitation rate, to the so-called artist loft effect realized by clustering innovators.
John Searle's Chinese room argument convey's a clear message: it is not necessary to understand language in order to use it. The symbolic shift often associated with language comprehension, rather than its indexical usage, is modelled here by a hierarchical ordering of three feedforward neural networks. The model is experimentally supported by a series of language training experiments that were originally conducted with Sherman, Kenton, Ericka and Austin, four young chimpanzees.
In a joint effort with Silvain van Weers and Andrew Koster, our university's robotics lab saw the birth of the Dutch AIBO team, ultimately bringing together 6 universities and ~60 students and staff members to successfully lead a team of autonomous four-legged robots to two Robocup World championships (Lisbon 2004, Osaka 2005) and two European championships (Paderborn 2004 & 2005). We eventually reached the quarterfinals in Osaka - losing against the Germans, but gaining a great deal of media attention on the way.