Scoop has an Ethical Paywall
It's not possible to control a super intelligent AI
Science fiction writers have worried about the idea of artificial intelligence getting out of control for years.
Now scientists at the Max Plank Institute in Germany are concerned.
A group of artificial intelligence researchers have published a paper that says once AI starts working at levels beyond the scope of today’s programmers, it won’t be possible to set limits on what it does.
This may seem like an idea that is far off, but there are already AI systems in use today that are making decisions and running things that the programmers who built the systems don’t fully understand.
Part of the problem is that while we could limit the capabilities of AI systems to make sure their capability doesn’t snowball out of control, the whole point of AI is to solve problems that are beyond us using conventional techniques.
I should point out these researchers don’t have any comforting suggestions.
It's not possible to control a super intelligent AI was first posted at billbennett.co.nz.
Bill Bennett
New Zealand technology news
Bill Bennett publishes technology news and features that are directly relevant to New Zealand readers.
Covering enterprise and small business computing, start-ups, listed companies, the technology channel and devices. Bennett's main focus is on New Zealand innovation.
Bill Bennett stories are republished on Geekzone and Scoop.
Stories published on this site are available to publishers for a fixed fee or a monthly subscription.
Binoy Kampmark: Warring Against Encryption, Australia Is Coming For Your Communications
Gordon Campbell: On Fast Track Powers, Media Woes And The Tiktok Ban
Binoy Kampmark: Censorship Wars, Elon Musk, Safety Commissioners And Violent Content
Gordon Campbell: On The Public Sector Carnage, And Misogyny As Terrorism
Ramzy Baroud: NATO’s Never-ending War: The 75-Year-Old Bully Is Faltering
Eugene Doyle: Joining AUKUS Not In NZ’s National Interest