Medium Article: https://medium.com/@jsmith0475/building-human-facing-agentic-systems-the-psychology-and-sociology-of-super-intelligence-1571cf565d51

Dr. Jerry Smith's January 2025 Medium article explores the psychological, sociological, and ethical implications of "human-facing agentic systems," AI powered by large language models (LLMs). These systems, leveraging multi-headed attention mechanisms, aim to collaborate with humans, not merely serve as tools. The article emphasizes the potential for both empowerment and manipulation inherent in these systems, highlighting the crucial role of self-interest, emotion, and identity in human-AI interaction. Smith concludes with a call for responsible development guided by transparency, inclusivity, and justice, warning against the potential for these systems to become instruments of control.

Building Human-Facing Agentic Systems: The Psychology and Sociology of Super Intelligence

Medium Article: https://medium.com/@jsmith0475/flat-facts-curved-beliefs-a-geometric-hypothesis-for-transformer-cognition-5ad6f850ebd5
The article, by Dr. Jerry A. Smith, proposes a geometric hypothesis for transformer cognition, suggesting that beliefs might operate within a curved, hyperbolic mathematical space, unlike factual information which likely resides in a flatter, Euclidean space. This theory attempts to explain why opposing concepts, like "love" and "hate," appear artificially close in traditional, flattened visualizations of transformer's internal representations. The author suggests that different "attention heads" within transformers may specialize in different geometries, with some handling stable facts in Euclidean space and others managing nuanced beliefs in hyperbolic space, which naturally accommodates hierarchies and divergent ideas. The text outlines potential experiments to test this hypothesis, such as measuring geodesic distances between beliefs in a hyperbolic model and analyzing the "tree-like" quality of attention head graphs. Ultimately, this perspective implies that transformers have independently discovered the need for varied geometries to fully represent the complexity of meaning, moving beyond the limitations of simply increasing Euclidean dimensions to accurately model human-like understanding.

Flat Facts, Curved Beliefs: A Geometric Hypothesis for Transformer Cognition

In-Depth Explorations of Neuroscience-Inspired Architectures Revolutionizing AI.

Flat Facts, Curved Beliefs: A Geometric Hypothesis for Transformer Cognition

Dr. Jerry A. Smith

Related tracks