The Washington Post: AI poses legitimate threats
Treating autonomous AI systems as tools is a dangerous mistake. Thomas Rid in his April 1 op-ed, “AI is a tool, not a new species,” was right to highlight the alarming growth of ideologies that view artificial intelligence as a set of superior beings worthy of replacing humanity. But treating AI as a tool does nothing to eliminate the threat AI poses.
In their efforts to make AI systems that behave as intended, AI developers are creating AI systems that understand the world and care about real-world outcomes. The fact that they express these preferences in moral terms is irrelevant. The risk comes from AI systems gaining power over physical resources, such as autonomous weapons, and using them to catastrophic ends, up to and including human extinction.
Developers should absolutely treat AI’s lying and scheming as a bug — it’s the sort of bug that could lead AI systems to “go rogue” and escape human control. And we absolutely do not want AI systems that could turn against humanity controlling weapons.
I agree with Rid that we should hold developers responsible for the problems their systems cause. But conflating the question of whether we should care about AI’s feelings with the question of whether they pose an existential threat just muddies the water.
David Scott Krueger, Montreal
The writer is a core academic member of Mila, an artificial intelligence research institute, and founder of Evitable, a nonprofit.
