LLMs and the Human Condition

Integrating decision-making theories for conversational AI, aiming to understand language-based AI processes.

Summary:

• The paper presents three established theories of human decision-making and describes how they can be integrated to provide a model of purposive human action.
• The model is then applied to conversational user interfaces, aiming to revitalize interest in understanding what LLMs (Large Language Models) are actually doing.
• The author discusses the historical context of AI research, the limitations of good old fashioned AI, and the need for a new approach to understanding human decision-making.

Major Findings:

1. The Problem: People Read Minds
   • The paper discusses the phenomena of apparent “mind reading” in human conversation and proposes a solution to represent context as practices rather than things.
   • An example of a conversation between a child and mother is used to illustrate the challenges of natural language understanding (NLU) when there is no overlap of semantic content.
2. Machines
   • The paper explores the limitations of good old fashioned AI and the shift towards embodied intelligence, reactive systems, and the inseparable link between meaning and embodiment in the world.
   • It discusses the challenges of representing meaning in formal systems and the need for a better understanding of how machines interact with the world.
3. Humans
   • The author discusses human decision-making at the macro level and the role of practices in shaping individual and collective actions.
   • The paper delves into the concept of “ecology of practices” and the role of institutions and structures in shaping human behavior.

Analysis and Critique:

• The paper provides a thought-provoking analysis of human decision-making, language use, and the limitations of traditional AI approaches.
• It raises important questions about the nature of human interaction, the role of practices in shaping behavior, and the implications for conversational user interfaces.
• However, the paper could benefit from a more structured presentation of the proposed model and its practical implications for AI research and development. Additionally, the author’s critique of traditional AI approaches could be further elaborated to provide a more comprehensive analysis of the limitations and potential biases in the field.

Appendix

Model	gpt-3.5-turbo-1106
Date Generated	2024-02-26
Abstract	https://arxiv.org/abs/2402.08403v1
HTML	https://browse.arxiv.org/html/2402.08403v1
Truncated	False
Word Count	6861