Integrating Computational Models of the Mind

Simon Award winner Marcin Milkowski speaking at the International Association for Computing And Philosophy 2016.

A vast majority of theoretical papers in cognitive science today describe computational models of cognitive processes. My focus in this talk will be on attempts to integrate separate computational models of the mind. Most models describe just how human and non-human subjects solve particular cognitive tasks. Since the 1970s, modelers are, however, acutely aware of the fact that partial models of particular tasks do not simply add up (Newell 1973; Newell 1990; but see Kosslyn 2006). There are several strategies that are supposed to help integrate them, or fill the gaps between these partial explanations. I will analyze three: cognitive architectures, interfaces, and experimental constraints.

The classical strategy of integration, pursued by proponents of production systems, is to describe complete cognitive architectures reused in multiple explanations (e.g., Anderson et al. 2004; Newell 1990; Sun 2004). In this case, theorists usually present simplified versions of previously offered explanations, by showing how they can be integrated in their architectures. For example, it’s possible to use Baddeley’s account to model working memory in ACT-R (Anderson, Reder, and Lebiere 1996). In the process, modelers implement features that were not in the original theory, and show implications for its further development. In other words, previous work is reused and adapted in integrated models of cognition.

The second, contemporary strategy is to build interfaces between different computational models and systems. For example, one may connect NEURON simulation software (Carnevale 2007) with LFPy, Python simulation package that computes biophysical properties of Local Field Potentials (Lindén et al. 2013). In such a case, the user simply can use LFPy to compute values necessary for his or her simulation of neural networks. A much more interesting case is building theoretically inspired interfaces between various computational subsystems, for example in hybrid cognitive architectures (Wermter and Sun 2000).

The last, much weaker, strategy is to constrain cognitive models by respecting certain known psychological limitations. For example, Herbert Simon and Allen Newell presupposed that the system may not have working memory bigger than 7±2 meaningful chunks (Newell and Simon 1972). It’s also presupposed that the model may not execute more than 100 operations per second, which is estimated to be the computational speed of the nervous system (Feldman and Ballard 1982).

I will consider the advantages and disadvantages of these strategies, in particular how they can contribute to building integrated mechanistic models of cognition.

References

Anderson, John R., Daniel Bothell, Michael D Byrne, Scott Douglass, Christian Lebiere, and Yulin Qin. 2004. “An Integrated Theory of the Mind.” Psychological Review 111 (4): 1036–60. doi:10.1037/0033-295X.111.4.1036.

Anderson, John R., Lynne M. Reder, and Christian Lebiere. 1996. “Working Memory: Activation Limitations on Retrieval.” Cognitive Psychology 30 (3): 221–56. doi:10.1006/cogp.1996.0007.

Carnevale, Ted. 2007. “Neuron Simulation Environment.” Scholarpedia 2 (6): 1378. doi:10.4249/scholarpedia.1378.

Feldman, J. A., and D. H. Ballard. 1982. “Connectionist Models and Their Properties.” Cognitive Science 6 (3): 205–54. doi:10.1016/S0364-0213(82)80001-3.

Kosslyn, Stephen M. 2006. “You Can Play 20 Questions with Nature and Win: Categorical versus Coordinate Spatial Relations as a Case Study.” Neuropsychologia 44 (9): 1519–23. doi:10.1016/j.neuropsychologia.2006.01.022.

Lindén, Henrik, Espen Hagen, Szymon Łęski, Eivind S Norheim, Klas H Pettersen, and Gaute T Einevoll. 2013. “LFPy: A Tool for Biophysical Simulation of Extracellular Potentials Generated by Detailed Model Neurons.” Frontiers in Neuroinformatics 7 (January): 41. doi:10.3389/fninf.2013.00041.

Newell, Allen. 1973. “You Can’t Play 20 Questions with Nature and Win: Projective Comments on the Papers of This Symposium.” In Visual Information Processing, edited by W. G. Chase, 283–308. New York: Academic Press.

———. 1990. Unified Theories of Cognition. Cambridge, Mass. and London: Harvard University Press.
Newell, Allen, and Herbert A. Simon. 1972. Human Problem Solving. Englewood Cliffs, NJ: Prentice-Hall.

Sun, Ron. 2004. “Desiderata for Cognitive Architectures.” Philosophical Psychology 17 (3): 341–73.

Wermter, Stefan, and Ron Sun. 2000. Hybrid Neural Systems. Edited by Stefan Wermter and Ron Sun. Berlin Heidelberg New York: Springer-Verlag.

The Herbert A. Simon Award for Outstanding Research in Computing and Philosophy recognizes scholars at an early stage of their academic career who are likely to reshape debates at the nexus of Computing and Philosophy by their original research.

It is with great pleasure the IACAP Board announces that Professor Marcin Milkowski has won the 2016 Simon Award for his significant contributions to the foundations of computational cognitive neuroscience.

Professor Milkowski serves as associate professor in the Institute of Philosophy and Sociology of the Polish Academy of Sciences. He is currently a managing editor of Przegląd Filozoficzno-Literacki (Philosophical-Literary Review). From 2005 to 2011, he served on the executive board of the Center for Philosophical Research, a new, independent scientific organisation that includes philosophers and scholars in humanities.

Professor Milkowski wrote his dissertation Konstrukcja umysłu. Intuicje zdrowego rozsądku a naturalizm w filozofii umysłu Daniela Dennetta (“Mind Design. Common-sense intuitions vs. naturalism in Daniel Dennett’s philosophy of mind”) under the supervision of Jacek Hołówka in Institute of Philosophy at Warsaw University. He received habilitation in Poland on the basis of his 2013 Explaining the Computational Mind (MIT Press, Cambridge, Mass.). His recent publications include:

  • 2015. Satisfaction conditions in anticipatory mechanisms. Biology & Philosophy, (February).
  • 2015. Function and causal relevance of content. New Ideas in Psychology, 1–9.
  • 2015. Explanatory completeness and idealization in large brain simulations: a mechanistic perspective. Synthese.
  • 2015. Evaluating Artificial Models of Cognition. Studies in Grammar, Logic, and Rhetoric, 40(1), 43–62.
  • 2014. Social intelligence: how to integrate research? A mechanistic perspective. In A. Herzig & E. Lorini (Eds.), Proceedings of the European Conference on Social Intelligence (ECSI-2014) (pp. 117–127).
  • 2014. Perspektywy ewolucjonistyczne w badaniach społecznych. In M. Gdula & L. M. Nijakowski (Eds.), Oprogramowanie rzeczywistości społecznej (pp. 185–208). Warszawa: Wydawnictwo Krytyki Politycznej.
  • 2014. Is the mind a Turing Machine? How could we tell? In A. Olszewski, B. Brożek, & P. Urbańczyk (Eds.), Church’s Thesis. Logic, Mind, and Nature (pp. 305–333). Kraków: Copernicus Center Press.
  • 2014. Computational Mechanisms and Models of Computation. Philosophia Scientiæ, 18(3), 215–228.
  • 2014. Computation and Multiple Realizability. In V. C. Mueller (Ed.), Fundamental Issues of Artificial Intelligence. Berlin – Heidelberg: Springer.