Technological innovation and complexity theory

Abstract

Complexity theory has become influential in recent models in social science. In the context of innovations and new technologies, most applications have focused on technology adoption and technology diffusion, whereas the topic of the innovation process has received less attention. This paper discusses three families of complexity models of technological innovation: fitness landscape models, network models and percolation models. The models are capable of analysing complex interaction structures (between components of technologies, between agents engaged in collective invention) while avoiding ‘over-parameterisation’. The paper ends with discussing the methodological challenges and critiques regarding the application of complexity theory that remain.

Keywords:

• Innovation
• Complexity
• Fitness landscape
• NK-model
• Percolation
• Complex networks

Acknowledgements

This paper has been written during the EXYSTENCE Thematic Institute on ‘Innovation and Complexity’, Tech Gate, Vienna, 6 September–8 October 2004 funded by the European Commission under no. IST-2001-32802. I thank Giorgio Fagiolo, Cristiano Antonelli and three anonymous referees for their helpful comments. The usual caveats apply.

Notes

¹Some models of technological innovation in evolutionary economics suffer from over-parameterisation, which means that the possible behaviours of the models for different parameters were not always properly understood. This problem has become less urgent given the rapid increases in computing power; yet, even some recent contributions do not systematically explore the whole parameter space (on this, see Windrum, 2004).

²Other approaches to fitness landscapes assuming a continuous design space are discussed by Hartmann-Sonntag et al. (2004). See also Kwasnicki (1996).

³For application of the NK-model to technological innovation, see Frenken et al. (1999), Auerswald et al. (2000), Kauffman et al. (2000), McCarthy (2003, 2004), Frenken and Nuvolari (2004), Hovhannisian (2004), Hovhannisian and Valente (2004), Sommer and Loch (2004) and Frenken (2005). For applications of the NK-model in organisation theory, see Kauffman and Macready (1995), Levinthal (1997), Levinthal and Warglien (1999), Gavetti and Levinthal (2000), Marengo et al. (2000), Rivkin (2000, 2001), Dosi et al. (2003) and Rivkin and Siggelkow (2003).

⁴For formal properties of NK landscapes, see Kauffman (1993) and the review chapter by Altenberg (1997).

⁵Following the Central Limit Theorem (Kauffman, 1993: 53–54).

⁶This also holds for genetic algorithms that recombine of substrings of two parent strings to create a string (Holland, 1975; Birchenhall, 1995).

⁷Other search algorithms are greedy search (evaluating all N neighbouring strings and choosing the one with highest fitness), extremal search (mutating the element with lowest fitness) and recombinant search (recombining of substrings of two parent strings to create a new string).

⁸In fact, it is more accurate to speak of small worlds properties than of small world models.

⁹This mechanism introduces a self-reinforcing rich-get-richer dynamics similar to the Arthur-model of increasing returns (Arthur, 1989).

¹⁰Also known as neural networks (Plouraboue et al., 1998).

¹¹There have been important recent advances in the establishment of complexity science in the social sciences as a field of scientific investigation. A number of complexity-minded journals started recently including free electronic journals like JASSS (http://jasss.soc.surrey.ac.uk/JASSS.html) and e-JEMED (http://beagle.u-bordeaux4.fr/jemed/). Web portals are available on agent-based modelling (http://www.econ.iastate.edu/tesfatsi/atechevo.htm) and networks (http://www.econ.iastate.edu/tesfatsi/anetwork.htm). A simulation platform is available from SWARM (http://www.swarm.org/). More specifically tailored to the community of evolutionary economics is a more recent initiative called the Laboratory for Simulation Development (http://www.business.auc.dk/lsd/). See also Valente and Andersen (2002).

¹²The KISS principle has recently been challenged by the KIDS principle (Edmonds and Moss 2004), referring to ‘Keep it Descriptive Stupid’. The KIDS approach simulates phenomena in the most straightforward way possible, taking into account the widest possible range of evidence. Simplification is only applied when and if the model and evidence justify this. In contrast, the Kiss principle advocates that one should start with the simplest possible model and later extended in necessary.

Additional information

Notes on contributors

Koen Frenken

E-mail: [email protected]