Objective
How do human infants develop complex cognition? We propose that artificial intelligence (AI) provides crucial insight into human curiosity-driven learning and the development of infant cognition. Deep learning—a technology that has revolutionised AI—involves the acquisition of informative internal representations through pre-training, as a critical precursory step to learning any specific task. We propose that, similarly, curiosity guides human infants to develop ‘hidden’ mature mental representations through pre-training well before the manifestation of behaviour. To test this proposal, for the first time we will use neuroimaging to measure the hidden changes in representations during infancy and compare these to predictions from deep learning in machines. Research Question 1 will ask how infants guide pre-training through directed curiosity, by testing quantitative models of curiosity adapted from developmental robotics. We will also test the hypothesis from pilot data that the fronto-parietal brain network guides curiosity from the start. Research Question 2 will further test the parallel with deep learning by characterising the developing infant’s mental representations within the visual system using the powerful neuroimaging technique of representational similarity analysis. Research Question 3 will investigate how individual differences in curiosity affect later cognitive performance, and test the prediction from deep learning that the effects of early experience during pre-training grow rather than shrink with subsequent experience. Finally, Research Question 4 will test the novel prediction from deep learning that, following perinatal brain injury, pre-training creates resilience provided that curiosity is intact. The investigations will answer the overarching question of how pre-training learning lays the foundations for cognition and pioneer the new field of Computational Developmental Cognitive Neuroscience.
Fields of science (EuroSciVoc)
CORDIS classifies projects with EuroSciVoc, a multilingual taxonomy of fields of science, through a semi-automatic process based on NLP techniques.
CORDIS classifies projects with EuroSciVoc, a multilingual taxonomy of fields of science, through a semi-automatic process based on NLP techniques.
- natural sciencesbiological sciencesneurobiology
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringrobotics
You need to log in or register to use this function
Keywords
Programme(s)
Topic(s)
Funding Scheme
ERC-ADG - Advanced GrantHost institution
D02 CX56 Dublin
Ireland