Epistemic Autonomy: Self-supervised Learning in the Mammalian Hippocampus

Area of Science:

• Neuroscience
• Artificial Intelligence
• Cognitive Science

Background:

• Biological cognition relies on epistemic autonomy (self-supervised knowledge acquisition).
• Artificial neural networks (ANNs) typically lack this autonomy, depending on external criteria.
• Error backpropagation in ANNs drives AI advancements, prompting questions about integrating biological autonomy.

Purpose of the Study:

• To investigate if epistemic autonomy can be achieved using error backpropagation in biological systems.
• To propose a mechanism within the entorhinal-hippocampal complex for combining autonomy and error-based learning.
• To analyze the computational emulation of this principle for autonomous cognitive systems.

Main Methods:

• Examining the functional role of the entorhinal-hippocampal complex in cognitive processes.
• Proposing a computational model involving a modulatory counter-current inhibitory network in the hippocampus.
• Analyzing the system's ability to minimize prediction errors through internal mechanisms.

Main Results:

• Evidence suggests the entorhinal-hippocampal complex integrates epistemic autonomy with error backpropagation.
• A novel mechanism is proposed where the hippocampus minimizes signal discrepancies via a specific inhibitory network.
• The computational emulation demonstrates potential for autonomous learning in AI.

Conclusions:

• The hippocampus may reconcile self-supervised learning with error-based mechanisms.
• This principle offers a pathway for developing more autonomous artificial intelligence.
• Understanding biological learning can inform the design of future cognitive architectures.