Research Summary: Why we need an AI-resilient society- Profiling Large Language Models

The evolution of Artificial Intelligence (AI) has progressed through three distinct generations, culminating in Large Language Models (LLMs) which convert natural language into a programming interface. This development profoundly reshapes societal processes such as knowledge generation, decision-making, and governance. LLMs introduce systemic risks, building upon earlier concerns like deepfakes from generative adversarial networks, necessitating a focused analysis of their characteristics and potential societal impact.

Why it matters

The emergence of Large Language Models (LLMs) fundamentally alters the human-computer interaction paradigm and introduces new complexities for societal functions. Understanding these shifts and their associated systemic risks is critical for long-term stability and effective strategic planning across various sectors. This development necessitates proactive measures to build societal resilience against potential disruptions.

Key insights

• AI's evolution is categorized into three generations: explicit logic programming, neural networks learning from data, and Large Language Models (LLMs) using natural language as a programming interface.
• The shift to LLMs extends AI's influence beyond computer science, impacting how societies create knowledge, make decisions, and govern.
• LLMs introduce a new class of systemic risks, differentiating from and adding to issues like deepfakes and synthetic media associated with generative adversarial networks.
• The report employs a 'mindhunting' approach, utilizing a forensic-psychology profiling methodology to characterize AI based on documented features like 'hallucination'.

Source

arXiv — Computers and Society — https://arxiv.org/abs/1912.08786