PERSPECTIVE

Deep Learning algorithms utilize "black box" techniques to imitate intentionality by discovering arbitrary functions from a trained set rather than prioritize models of consciousness that accurately represent the fundamental components of life, such as phenomenology. Given that current AI is not neuroscience but mostly within software engineering, techniques, i.e., “black-box”  approaches, may prove futile as quantum information biology or intrinsic information is subjective physicalism and cannot be predicted with Turing computation. However, the potential of conscious AI, with its functional systems that surpass automation and rely on elements of understanding, is a beacon of hope in the AI revolution. The shift from automation to conscious AI, once replaced with machine understanding, offers a future where AI can comprehend without needing to experience, thereby revolutionizing the field of AI. The proposed dynamic Organicity Theory of consciousness (DOT) is a promising approach for building artificial consciousness that is more like the brain with physiological nonlocality and diachronicity of self-referential causal closure. However, the limitations of the black-box approach in deep learning algorithms present a significant challenge. This perspective suggests that deep learning algorithms effectively decode labeled datasets but not hidden data due to unlearnable noise, and encoding intrinsic information is beyond the capabilities of deep learning. New models based on DOT are necessary to decode hidden information by understanding meaning and reducing uncertainty (lacking information). The process of “encoding” entails functional interactions in evolving informational holons, forming informational channels through thermodynamic constraints that reduce informational redundancy (also referred to as intentionality). Therefore, there is a possibility that advances will be made through observer-independent non-Turing computation, but this will require leapfrogging singularity and replacing machine learning with machine understanding.

​

Keywords: Non-Turing computation, non-metacognitive approaches, deep learning-neural decoding, nonlinear time, self-referential causal closure, functional dynamics, informational channels, dynamic organicity theory, science of consciousness.  

​

Conflict of Interest

The author/s declare that they were an editorial board member of JMN, at the time of submission. This had no impact on the peer review process and the final decision.

​

This article belongs to the Special Issue                

Multiscalar brain adaptability in AI Systems             

      Lead Editor:  Dr. Shantipriya Parida
     Senior Scientist
      Silo AI,  Helsinki, Finland

​

Copyright: © 2024 The Author(s). Published by Neural Press.

This is an open access article distributed under the terms and conditions of the CC BY 4.0 license.

​

Publisher's note: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors, and the reviewers. Any product that may be evaluated in this article, or claim that made by its manufacturer, is not guaranteed or endorsed by the publisher.