The 2016 "runaway AI" vision, ten years later: the real risks are quieter
Understand this story
The author looks back at the HASI (Human-Centered Artificial Superintelligence) concept he proposed in his 2016 graduate school application, arguing that its biggest flaw was that it only provided the endpoint of "aligning goals" without explaining the implementation mechanism. What actually arrived ten years later are language models that have been used daily by millions of people since late 2022, and the risks manifest more quietly. The EU AI Act took effect in August 2024 and implements obligations in phases; the author says the nearer danger is that people are gradually abandoning the skills needed to supervise AI.
AI overview based on source reports · Updated 30 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIWhat I Wrote About Rogue AI in 2016, and What Changed
The author revisits a 2016 graduate-school proposal for HASI (Human-focused Artificial Super Intelligence), noting that its worry about autonomous weapons and pre-emptive guardrails held up while its terminology and lack of a mechanism did not. He pairs HASI with his RACE framework (Research, Advisory, Conceptualize, Educate) to govern how much autonomy AI systems receive, and argues the nearer risk is people losing the skills needed to oversee machines, citing a 2025 Microsoft and Carnegie Mellon survey of 319 knowledge workers in which 62 per cent reported thinking less critically when using generative AI. He concedes HASI cannot design for superintelligence and applies mainly to systems deployed now, illustrating the approach with a community pharmacy example.
Activity for this story
Not enough continuous observations to show a trend yet.