Presentation Master's thesis - Martin Ilić - psychological research methods

Abstract reasoning, the ability to solve complex problems by taking away the unnecessary

details (Clement et al., 2007) in order to derive a rule used in solving similar, novel tasks, is

an essential intelligent behaviour that AI deep learning models are generally not yet capable

of yet appears early on in humans. This thesis investigates whether the Emergent Symbol

Binding Network (ESBN; Webb et al., 2021) is a possible candidate for studying the

mechanisms that underlie how abstract visual reasoning (AVCR) ages and develops in

humans. By manipulating ESBN’s architecture when performing two AVR tasks – identity

rules and distribution-of-three - we tested if it could simulate two main developmental

phenomena, i.e. – that higher working memory capacity and improved inhibition control

promote AVR development. Results showed the ESBN failed to simulate the working

memory phenomenon, while the inhibition control phenomenon could not be tested due to the

model’s near-perfect task accuracy. This makes the ESBN an inadequate model for explaining

AVR development, a finding further research should corroborate.