×
Why thinking of AI as human might actually help us understand the risks
Written by
Published on
Join our daily newsletter for breaking news, product launches and deals, research breakdowns, and other industry-leading AI coverage
Join Now

The collision of human and artificial intelligence frameworks presents a thought-provoking paradox in how we understand AI risks. While experts frequently warn against anthropomorphizing AI systems, there may be legitimate value in comparing AGI dangers to human behaviors. This counterintuitive approach could actually accelerate public understanding of genuine AI risks by leveraging our intuitive grasp of human capabilities, rather than treating AI dangers as abstract or hypothetical threats.

The big picture: AI systems share many concerning properties with humans that make them potentially dangerous, suggesting our intuitions about human behavior might be useful for understanding AI risks.

  • Humans already demonstrate situational awareness, goal-directedness, general intelligence, unpredictability, deceptiveness, and instrumental convergence—all properties that make advanced AI systems concerning.
  • These parallels could make anthropomorphism a useful tool for public understanding rather than merely a dangerous misconception.

Why this matters: The coming transition to widespread human-like AI agents could trigger a necessary shift in public perception about AI capabilities and risks.

  • Current public understanding significantly underestimates how rapidly AI systems are approaching human-like capabilities in ways that could pose existential risks.
  • Accelerating the development and deployment of human-like AI agents while base models remain somewhat limited could help catalyze public concern before truly dangerous AGI emerges.

Counterpoints: Anthropomorphizing AI carries significant risks despite potential benefits in raising awareness.

  • Assuming AI systems have human prosocial instincts or constraints could lead to dangerous miscalculations about AI safety.
  • Human-like behaviors in AI might mask fundamental differences in how AI systems actually operate, creating false confidence in our ability to predict or control their actions.

Reading between the lines: The article posits that accelerating toward human-like AI applications might paradoxically improve safety by forcing earlier public reckoning with AI risks.

  • This approach suggests that confronting AI dangers through widespread exposure to increasingly capable but still-limited systems could build consensus for safety measures before superintelligent systems emerge.
  • The author acknowledges this position remains provisional and invites counterarguments to this acceleration thesis.

Where we go from here: A potential shift in public opinion toward “rational alarm” about AI progress appears inevitable but may come too late without deliberate efforts to accelerate understanding.

  • The development of increasingly human-like AI agents could serve as a visceral demonstration of AI capabilities that abstract discussions of AI risk have failed to convey.
  • Timing this shift in public perception before truly dangerous AGI proliferates represents a critical challenge for AI safety advocates.
Anthropomorphizing AI might be good, actually

Recent News

Somerset Council trials AI to speed up special educational needs reports

AI automation allows Somerset caseworkers to reduce paperwork and spend more time directly helping children with special needs while maintaining human oversight of final decisions.

Surfshark report reveals alarming data collection by AI chatbots

Popular AI chatbots collect vast amounts of personal data, with Meta AI harvesting 90% of possible data types while others collect significantly less for the same functionality.

Pátria commits $1B to build AI-powered data centers in Latin America

Brazilian asset manager commits $1 billion to build renewable-powered, high-capacity data centers across Latin America, leveraging the region's regulatory stability and geopolitical neutrality to attract global AI computing workloads.