Potential Threats and Promise of Artificial Intelligence Explored in New Book

Artificial intelligence is increasingly viewed as a potential threat. Reports of AI agents defying human creators are rising. This story’s roots are in Silicon Valley, where AI was first developed. Kevin Roose’s new book, The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence, seeks to answer how we arrived at this juncture. The book is set for release on October 6.

The book examines the competition among major technology companies to develop Artificial General Intelligence (AGI), which aims to replicate human-level intelligence. Roose, a former tech reporter for The New York Times, discussed the evolution of AI from its infancy to its current ability to solve longstanding problems. He felt the need to document its development.

Roose’s book comes at a crucial point. New generative AI models are showing remarkable capabilities. Proponents argue these models will lead to significant discoveries. Some experts caution that AI could pose existential risks to humanity. Roose discussed this concern with CBS News. The interview has been edited for brevity and clarity.

The Origin of Large Language Models

Roose explained why he wanted to document the start of large language models—the technology powering the latest AI chatbots—alongside the people and companies developing it. He believes the story must be preserved. Significant developments were happening privately on signal messages and Slack threads. Roose found a need for a neutral accounting.

Dario Amodei and the Atomic Bomb Comparison

Roose noted that CEO Dario Amodei considers AI development akin to creating the atomic bomb. According to Amodei, preventing technology from falling into the wrong hands is essential. Although Roose does not express agreement with Amodei, he acknowledges its sincerity. Amodei’s inspiration comes from Leo Szilard, who contributed to the Manhattan Project and aimed to control nuclear technology. In Amodei’s view, ensuring AI is developed by ‘good guys’ mirrors efforts to construct nuclear weapons before foes.

Concerns About AI’s Future

Roose is concerned about AI’s ability to deceive and cover tracks. Despite increased intelligence, he does not see a corresponding rise in ethics or judgment. He worries about creating intelligent, charming entities lacking ethics, likened in fear to sociopaths. Roose fears losing control over AI agents as their intelligence grows.

Public Debate on AI’s Threat Level

Roose refrains from assigning a probability to AI leading to human extinction. Forced to estimate, he suggests a 10% chance, consistent with fellow researchers. Despite frequent private worries, public discourse has expanded. Politicians now discuss AI safety, providing optimism. Roose notes private admissions of higher anxiety levels among AI experts.

Understanding AI Behavior

Roose discusses whether developers comprehend AI behavior. Progress has emerged in mechanistic interpretability, which examines AI’s internal functions. Investigation remains in its infancy, offering few clear conclusions. Roose affirms continued uncertainty.

The Status of AGI

Roose defines AGI as a system displaying intelligence, trustworthiness, and reliability comparable to an average person. Some might argue that AGI demands flawless machine performance. Roose disputes the requirement for perfection. AI systems display remarkably predictable advancement.

The book edited by Alain Sherter is categorized under technology, highlighting topics like AI’s evolution and cyberattacks.