Knowledge Applied to New Domains: The Unconscious Succeeds Where the Conscious Fails

Document Type: 
ASSC Conference Item
Article Type: 
Artificial intelligence
Unconscious States Processing
Unconscious knowledge, Unconscious transfer, Familiarity
Deposited by: 
Ryan Scott
Contact email:
Date of Issue: 
Ryan Scott, Zoltán Dienes
Event Dates: 
5-8 June 2009
Event Location: 
Event Title: 
The 13th Annual Meeting of the Association for the Scientific Study of Consciousness
Event Type: 
ASSC Conference
Presentation Type: 
Publish status: 
There is a common view that consciousness is needed for knowledge acquired in one domain to be applied to a novel domain. We present evidence of precisely the opposite; where the transfer of knowledge from one domain to another is achieved only in the absence of conscious awareness. Adopting higher order thought theory and exploiting subjective measures of consciousness we examine knowledge transfer using the artificial grammar learning (AGL) paradigm. In the standard non-transfer AGL task participants are initially exposed to strings of letters which, unbeknown to them, conform to complex grammar rules. Subsequently participants are informed of the existence of rules and required to distinguish the grammaticality of new letter strings. In this context participants reveal both conscious and unconscious knowledge; judgments made with confidence and those attributed to random selection both show above chance accuracy, with conscious knowledge being more reliable. Knowledge in all categories appears to derive from familiarity (Scott & Dienes, 2008). In the transfer variant of AGL, training strings are presented in one alphabet or modality and test strings presented in another. We examined three types of transfer: between one letter-set and another, between tone sequences and letter sequences, and between tone sequences and symbol sequences. For each test string participants reported whether they thought the string was grammatical, how familiar it felt, how confident they were in their grammaticality judgment, and their perceived basis for that judgment i.e. random selection, intuition, familiarity, rules, or recollection. An objective measure of fluency was also obtained using a timed perceptual clarification task. Results were consistent across all three conditions. Responses attributed to random selection showed above chance accuracy (60%) while those attributed to other categories did not (52%). Familiarity ratings were predicted by consistencies in the repetition structure of training and test strings and were hence related to grammaticality. Fluency, though increasing familiarity, was unrelated to grammaticality. While all judgments were predicted by familiarity ratings only those attributed to random selection showed a significant additional contribution of grammaticality. It appears that in knowledge transfer, as in visual perception (Marcel, 1993), the unconscious may outperform the conscious.
Berlin 2009.pdf159.76 KB