Why Biden’s ‘central control’ approach to AI and national security might not work

Biden’s new AI regulation plan: A step forward with limitations: President Biden has unveiled a plan to regulate artificial intelligence, focusing heavily on national security concerns while falling short in addressing transparency and responsible use.

The context of AI regulation: The White House’s approach to AI regulation appears to be influenced by concerns about potential threats, including scenarios reminiscent of science fiction.

• Biden’s plan comes after a previous memorandum that lacked substantial content, with the new version offering more detailed proposals.
• The plan mentions “national security” 68 times, while “responsible use” and “transparency” are mentioned only 18 and 2 times respectively, indicating a strong emphasis on security concerns.

Understanding AI as a tool, not an entity: The plan’s approach to AI regulation may be misguided in treating AI as a singular, controllable entity rather than a versatile tool.

• AI is not a character from movies like The Terminator, but rather a technology designed to assist humans in various tasks.
• Regulations should focus on how AI is used rather than attempting to control the models themselves.

Control and human oversight in AI systems: Biden’s plan emphasizes human control over AI, particularly in military applications.

• The plan proposes a “human in the loop” approach, where the President would decide when to use military AI.
• However, this approach may not always be practical or effective, especially in situations requiring rapid decision-making or when communication with human operators is impossible.
• The reliability of human decision-making based on AI-generated information is also questionable, given the potential for misleading or manipulated data.

Data ownership and AI development: The plan acknowledges the importance of data in AI development but fails to address critical issues of data access and ownership.

• The United States is ahead of some regions, like Europe, in recognizing data’s role in AI development.
• However, the plan doesn’t provide clear solutions for accessing and utilizing data for AI training.
• Different countries’ varying data privacy rules and aggressive data collection practices (e.g., China) could lead to imbalances in AI capabilities and economic advantages.

Transparency in AI systems: Biden’s plan falls short in addressing the crucial aspect of AI transparency.

• The plan only briefly mentions the need for transparency without providing specific measures to achieve it.
• Recent incidents, such as Google’s Gemini model controversy and studies showing bias in AI language models, highlight the importance of transparency in AI systems.
• A more comprehensive approach to monitoring and understanding AI outputs is necessary to address potential biases and unintended consequences.

Limitations of centralized AI control: The plan’s approach to regulating AI as a centrally controllable technology may be unrealistic given the widespread accessibility of AI tools.

• AI technology is becoming increasingly accessible and affordable, making it difficult to implement centralized control measures.
• While this accessibility can be beneficial in fields like healthcare, it also poses challenges in preventing misuse, such as the development of autonomous weapons.

Analyzing deeper: The need for a more comprehensive approach: While Biden’s AI plan represents progress in addressing the challenges posed by artificial intelligence, it falls short in several critical areas.

• The plan’s heavy focus on national security overlooks the importance of transparency and responsible use in AI development and deployment.
• A more balanced approach that involves both public and private partnerships for monitoring AI utilization is necessary.
• Drawing lessons from past experiences with opaque algorithms in social media, the government should prioritize creating a framework that promotes understanding and responsible management of AI technologies across various sectors.