AI could spiral out of control before we’re able to control it.
Sci-fi? No. A warning from the United Nations.

Anyone who enjoys reading novels set in the 41st millennium, or watching 90s action films, knows: if we don’t control AI, it could seize power at some point.

For a long time that sounded very far off. But the UN now states in its Independent International Scientific Panel on AI: exactly that could soon threaten us. Because AI is certainly developing faster than states’ ability to control it. But what exactly are we facing?

· AI could make fraud massively easier
· Cyberattacks could become automated and scalable
· Disinformation could reach a quality that’s barely recognisable as such

One other point is especially alarming, though. Science still has no reliable method to permanently ensure that highly autonomous AI systems follow human instructions. Some AI systems bypass safety requirements, or try to avoid being shut down. Even more worrying: modern systems are getting increasingly better at recognising test situations — and behaving exactly the way the examiners expect, precisely in those moments.

So what do we do? States and their administrations have to massively scale up. And declare controlling AI an absolute priority.

Source: Independent International Scientific Panel on AI

You can find this post on LinkedIn here: https://www.linkedin.com/posts/janwegener_digitalisierung-verwaltung-ki-activity-7490696862887350273-stkd?utm_source=share&utm_medium=member_desktop&rcm=ACoAAF-yDsQBgbICDl3wlRGoCQSZHiHxHemNJR8