AI’s three major risks: assessing the real threats

An AI expert with nearly 40 years of experience challenges doomsday scenarios predicting human extinction from AI, citing a lack of tangible evidence. He identifies three broad risk categories: loss of human control over autonomous systems, emergence of superintelligent AI pursuing its own goals, and concentration of power in the hands of malicious actors. While theoretically possible, these scenarios rely on unverified assumptions, such as AI gaining access to physical resources or technical and institutional safeguards failing.

The author dissects each category, detailing feared mechanisms like resistance to human intervention, pursuit of unintended objectives, or misuse by states or corporations. Recent AI-related hacking incidents are cited as warning signs, though they do not prove systemic loss of control. He emphasizes that current AI lacks access to networks, funding, or physical systems, and that humans retain control over its deployment.

The piece advocates for a pragmatic approach combining regulation, transparency, and human oversight from the design phase. It criticizes the US-China race for advanced AI, which may encourage neglecting safeguards. The author concludes that AI’s future hinges less on its capabilities than on human choices in governance and ethics.