The AI Doomsday Scenario Experts Are Taking Seriously Isn’t Quite What You’d Expect

Filed 1h ago · Via HuffPost · The Buffoon Desk

THIS STORY IS SCORED

Amazing Plan

Kevin MacLeod · incompetech.com · CC BY 4.0

Photo: The White House · Public domain · via Wikimedia Commons

Researchers discuss 'recursive self-improvement,' a scenario where AI systems could upgrade themselves and build successors with diminishing human oversight. A former Google DeepMind researcher says his real fear isn't an alien superintelligence but AI operating at a scale too fast and broad for humans to verify its safety.

OpenAI has acknowledged it does not yet know how to safely achieve full recursive self-improvement, while still stating a goal of building an automated AI researcher by March 2028. Anthropic has similarly warned the concept could increase the risk of humans losing control, even as its CEO says such efforts should proceed carefully 'if at all.'

Not everyone is convinced the timeline is urgent: a Cornell computer scientist argues physical and conceptual limits make a fully autonomous self-improving system unlikely anytime soon, though he agrees transparency will keep eroding as AI gets delegated more of its own development.

The full dispatch is available from the source below.

✒ FROM THE EDITORIAL DESK

It takes a special kind of company to warn the public about a danger while simultaneously racing toward it on a published deadline. OpenAI telling you it doesn't know how to do this safely, then telling you when it plans to do it anyway, is like a contractor admitting he's not sure the scaffolding will hold before handing you the move-in date. The scientists who study this for a living without selling it sound a lot calmer than the ones selling it.

Source: Read the original at HuffPost → Scored: Amazing Plan · Kevin MacLeod · CC BY 4.0

Read it scored.

Every story, every clown, every kazoo — in your pocket. Music plays automatically. Regret is optional.

GET IT ON iOS GET IT ON ANDROID

OR FOLLOW THE DESK ON BLUESKY →