[Paper Review] Ethical Artificial Intelligence Principles and Guidelines for the Governance and Utilization of Highly Advanced Large Language Models

Soaad Hossain, Syed Ishtiaque Ahmed|arXiv (Cornell University)|Dec 19, 2023

Artificial Intelligence in Healthcare and Education4 citations

TL;DR

This paper proposes a proactive governance framework for highly advanced large language models (LLMs) using existing ethical AI principles—particularly responsibility, robustness, and technology misuse—alongside UNESCO and EU guidelines on societal/environmental well-being and safety. It advocates for pre-deployment screening akin to pharmaceuticals to mitigate irreversible harms before such models surpass human intelligence.

ABSTRACT

Given the success of ChatGPT, LaMDA and other large language models (LLMs), there has been an increase in development and usage of LLMs within the technology sector and other sectors. While the level in which LLMs has not reached a level where it has surpassed human intelligence, there will be a time when it will. Such LLMs can be referred to as advanced LLMs. Currently, there are limited usage of ethical artificial intelligence (AI) principles and guidelines addressing advanced LLMs due to the fact that we have not reached that point yet. However, this is a problem as once we do reach that point, we will not be adequately prepared to deal with the aftermath of it in an ethical and optimal way, which will lead to undesired and unexpected consequences. This paper addresses this issue by discussing what ethical AI principles and guidelines can be used to address highly advanced LLMs.

Motivation & Objective

• Address the lack of ethical AI governance frameworks for highly advanced LLMs despite their imminent potential to surpass human intelligence.
• Identify and adapt existing ethical AI principles and guidelines to the unique risks posed by advanced LLMs.
• Highlight the urgency of policy development before advanced LLMs are deployed, given their potential for irreversible societal and environmental harm.
• Propose a regulatory model inspired by pharmaceutical drug approval processes to ensure safety and accountability.
• Stimulate further research into ethical governance of advanced LLMs beyond current AI policy frameworks.

Proposed method

• Adopt the Australian Government’s definition of responsibility, requiring accountability for AI outcomes across the system lifecycle.
• Integrate African community perspectives on robustness, emphasizing ethical, legal, and socio-cultural impact mitigation.
• Apply UNESCO’s 11 ethical AI guidelines, particularly those on societal and environmental well-being, to assess environmental and health impacts of advanced LLMs.
• Incorporate EU guidelines on safety, emphasizing resilience, fallback mechanisms, accuracy, and reliability.
• Use UNESCO’s accountability guidelines to ensure traceability, oversight, and audit mechanisms for advanced LLMs.
• Propose a pre-deployment screening and approval process modeled on pharmaceutical drug regulation to manage high-risk AI systems.

Experimental results

Research questions

• RQ1 Which ethical AI principles are most relevant for governing highly advanced LLMs before they surpass human intelligence?
• RQ2 How can existing ethical AI guidelines from UNESCO and the EU be adapted to address the unique risks of advanced LLMs?
• RQ3 What policy mechanisms can be implemented to proactively mitigate the ethical, legal, and socio-cultural impacts of advanced LLMs?
• RQ4 To what extent can regulatory models from high-risk domains like pharmaceuticals be applied to advanced LLMs?
• RQ5 What are the key considerations for ensuring that advanced LLMs do not cause irreversible harm to society and the environment?

Key findings

• Ethical AI principles such as responsibility, robustness, and technology misuse prevention are essential for governing advanced LLMs.
• UNESCO’s guidelines on societal and environmental well-being require organizations to assess direct and indirect environmental impacts throughout the AI lifecycle.
• The EU’s safety guidelines emphasize resilience, fallback mechanisms, and reliability, which are critical for advanced LLMs.
• Accountability mechanisms, including oversight and audit trails, are necessary to ensure compliance with human rights and environmental norms.
• A pharmaceutical-style approval process is recommended for advanced LLMs due to their potential for large-scale irreversible harm.
• Existing ethical AI frameworks must be extended to cover the unique capabilities of advanced LLMs, such as generating novel ideas and performing previously impossible tasks.