Is a newly learnt sense immediately combined with vision?

Abstract

In many perceptual tasks, observers minimise random error by reliability-weighted averaging (e.g. Ernst & Banks, Nature 2002). Late development of this ability in childhood (e.g. Nardini et al, Curr Biol 2008; Dekker et al, Curr Biol 2015) suggests that it requires either considerable experience with specific cues or maturation of the nervous system. To study the role of experience, we asked if adults learning a new sense would immediately combine it with vision. Ten adults judged the distance to a fish hidden in a virtual "sea". In 75 initial trials they were trained in using a new sense, echolocation – clicks and their echoes, coming from a virtual surface at the distance of the fish. In 175 subsequent trials they localised the fish using echoes, a noisy visual cue (bubbles), or both. The visual cue's variance varied trial-to-trial, tracking on average the echo cue's estimated variance. The fish was shown after each trial. 7/10 subjects were above chance at using the new cue. Their data were analysed by Bayesian model comparison. For most (5/7), single-cue models (switching or vision-only) were better fits than cue combination models – including those that do not reweight by reliability or that systematically mis-weight – by a large margin (Bayes factors > 20). Author JN and one final participant were best fit by an optimal combination model (BF > 100 over the best single-cue model). Most participants who learned a new sense did not immediately combine it with vision. This suggests that extended experience with specific cues underlies human cue combination abilities. However, optimal combination after very little, or relatively little (author JN) experience was also seen. In future research we will ask which aspects of single-cue learning (e.g. bias) predict cue combination, whether with practice more participants learn to combine cues, and what intermediate stages exist. Meeting abstract presented at VSS 2016