[feed] pefprints@pef.uni-lj.si | [feed] Atom [feed] RSS 1.0 [feed] RSS 2.0 |
  Logo Login | Create Account

Computer modelling of the influence of natural selection on perceptual veridicality

Tine Kolenik (2018) Computer modelling of the influence of natural selection on perceptual veridicality. MSc thesis.

Download (3325Kb)


    The thesis explores computer modelling and its value in cognitive science as natural epistemology. This exploration is realised on several levels of analysis in terms of abstractness. Cognitive science and epistemology are argued to be closely related, manifesting their overlap in what some proponents of this connection name natural epistemology. The latter is defined as the study of epistemological questions with scientific methods. Key elements of natural epistemology are identified and proposed, most importantly the loop of knowings between epistemological insights in cognitive science and epistemology of cognitive science, which characterises progress in natural epistemology. Cognitive science and epistemology both primarily wonder what the relationship between mind and world is, and perception is identified as one of the sources on the knowledge of the world. It is therefore chosen for investigating this relationship, taking the evolutionary perspective on the development of perception. Computer modelling with genetic algorithms is used to study whether it is isomorphic or non-isomorphic perception that is more beneficial for modelled organisms. Two computer models are introduced – a model presented by Donald D. Hoffman and his colleagues, which possesses cognitivist presuppositions, and a newly designed model, which builds on Hoffman’s model by replacing certain cognitivist presuppositions for enactivist ones, mostly focusing on the addition of a sensorimotor loop. The models both produce the same results, as they show that non-isomorphic perception is evolutionary more beneficial than isomorphic. However, the sensorimotor loop causes the newly designed model to evolve faster. Afterwards, computer modelling is presented in the light of cognitive science as natural epistemology, questioning the results’ validity. The value and role of computer modelling is shown to be historically monumental by placing it in the loop of knowings and showing its influence on epistemological insights in cognitive science as well as epistemology of cognitive science. Despite the influence, several problems are identified, especially the “PacMan Syndrome”, the problem of the designed agents being unable to self-determine their meaning, which is forced upon them by the designer instead. The value of the two implemented models is discussed in this light. Two essential questions are posed: What do the models tell us about cognition? What role does their modelling play, especially the approach of designing models with different (epistemological) presuppositions and discerning their influence on final results? The first question is addressed by evaluating the models in several areas. The models are found to be explanatory of a possibility of non-isomorphic perception evolving (as opposed to the prevalent thoughts on that not being possible), not predictive (as they are not meant to be), abstract and simple, which may hinder the approach of comparing the models for their presuppositions, as they might not be able to affect the results because of the simplicity. Regarding the role of genetic algorithms, their arbitrariness in certain elements is presented as problematic, but as even more problematic, the design of the fitness function is presented. The fitness function is identified as an instantiation of the PacMan Syndrome, as the fitness function dictates what is good and what is bad for the models’ agents. It is suggested that by making the fitness function evolvable phylogenetically and ontogenetically, the designer’s role in predictably forcing its own meaning onto the agent is diminished a bit. By making the models more complex, the approach of comparing them would be made more legitimate in this case, but it was argued that it was a useful approach, as it showed the value of the sensorimotor loop. Regarding the models’ value on learning about cognition, it is suggested that they offer a functional understanding of a possible occurrence of non-isomorphic perception. Finally, the models are placed in the loop of knowings, their possible influence speculated upon.

    Item Type: Thesis (MSc thesis)
    Keywords: cognitive science, computer modelling, enactivism, epistemology, evolution, genetic algorithms, perception
    Number of Pages: 100
    Language of Content: English
    Mentor / Comentors:
    Mentor / ComentorsIDFunction
    izr. prof. dr. Urban KordešMentor
    prof. dr. Igor FarkašComentor
    Link to COBISS: https://plus.si.cobiss.net/opac7/bib/12208969
    Institution: University of Ljubljana
    Department: Faculty of Education
    Item ID: 5468
    Date Deposited: 03 Dec 2018 11:14
    Last Modified: 03 Dec 2018 11:14
    URI: http://pefprints.pef.uni-lj.si/id/eprint/5468

    Actions (login required)

    View Item