Cross-modal Turing test and embodied cognition: agency, computing

Deep learning paradigm allowed computer scientists to take a fresh look at the format of knowledge representation and assimilation. Studies of artificial analogs of neurons and synaptic connections of the brain have indicated many significant regularities in the opposition of cognitive "medium&...

Full description

Saved in:
Bibliographic Details
Published in:Procedia computer science 2021, Vol.190, p.527-531
Main Author: Leshchev, Sergey V.
Format: Article
Language:English
Subjects:
artificial intelligence
embodied cognition
Turing test
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Deep learning paradigm allowed computer scientists to take a fresh look at the format of knowledge representation and assimilation. Studies of artificial analogs of neurons and synaptic connections of the brain have indicated many significant regularities in the opposition of cognitive "medium" and cognitive information. This understanding gave a new impetus to the previously developed concept of embodied cognition in various branches of artificial intelligence. "Embodiment" is usually understood as a combination of cognitive and substrate components. At the same time, there remain world-systemic connections that involve a broader context in the dynamics of correlation between the subject of cognition (cognitive agent, bounded rationality) and the environment. The concept of embodied cognition assumes a clash of the range of cognitive systems, built upon different infogenesis and infotectonics (for example, different computing platforms and degrees of agency). The cross-modal Turing test is supposed to be a universal communication interface that allows “message” and “medium” of embodied cognitive agent to test each other. The use of reciprocal, environments and systems will allow a sequential cross-modal Turing test for two competing modules. Such an approach may turn out to be decisive in cyber-physical systems, which are born at the junction of diverse technical-scientific engineering solutions, as well as in systems that require a high learning rate and model correction. In neural network practice, this approach can be effective in the field of transfer learning, in which a pre-trained fragment of the network can be correlated with fundamentally irrelevant (for a neural network) data.
ISSN:1877-0509
1877-0509
DOI:10.1016/j.procs.2021.06.061