back to article UN boss recommends nuclear option for AI regulation

United Nations secretary-general António Guterres has called for the formation of a global AI watchdog modelled after the International Atomic Energy Agency (IAEA).  "New technology is moving at warp speeds, and so are the threats that come with it," he said in a briefing on Monday. "Alarm bells over the latest form of …

  1. Filippo Silver badge

    The IAEA regulates something that is comparatively obvious and very well-defined.

    An AI regulatory agency couldn't work in the same way. Its first task would be to define what exactly its own mandate is. That is not a good task for a supervisory agency.

  2. Chewi
    Terminator

    I saw the headline and thought he meant to put AI in charge of the nukes.

    1. Anonymous Coward
      Anonymous Coward

      @Chewi - It will come

      just give it (literally) a little time.

  3. b0llchit Silver badge
    Facepalm

    ...declaring AI an existential threat to humanity...

    Well, Humans are the existential threat to humans. Nothing else required.

    And then:

    ...inspecting systems [...] above a certain threshold...

    [and]

    ...participation meaning they would agree to having their software audited and capabilities controlled.

    Who are they fooling? Who makes up this crap?

    Anyone not participating will be spared. And for computation thresholds... they surely have not read the history books on computing, well, have they? Fools they are.

    1. doublelayer Silver badge

      I have similar objections, and another one about the "certain threshold". I wonder how they plan to specify that. It's not as if AI systems have a numerical complexity score that can be directly compared. A tiny program can be used in a damaging way and a massive program can be next to useless. A text-generating model and a facial recognition system are dangerous in completely different ways. I wonder whether any politician has considered even for ten seconds how they'd decide what would get the severe regulation* and what would not. This is in addition to the problem that they don't have much of a hope that anyone will comply with that regulation or that they will be able to penalize those who ignore them.

      * They don't seem to have a great idea of what that regulation would be, but they do agree there would be a lot of whatever it is.

  4. Anonymous Coward
    Anonymous Coward

    Too little, too late.

    What's the likelihood of a global committee being able to do anything this decade? And where will AI be by then?

    1. claimed Silver badge

      Re: Too little, too late.

      Going to hazard a guess… pretty much where we were 10 years ago, cos that’s where we are now folks

    2. Anonymous Coward
      Anonymous Coward

      @AC - Re: Too little, too late.

      Don't worry, with the wise and competent guidance from AI the global committee will be able to.... Oh, crap! We're doomed!

  5. amanfromMars 1 Silver badge

    Option 1... Grant Danegeld to NEUKlearer HyperRadioProACTive IT AIdDevelopers .....

    .... to Encourage and Ensure Sabbatical Gardening Leaves/Moritoria on Problematical Progressive Projects.

    AI could lead to the creation of a technology that could endanger humanity. A global regulatory agency could counteract the risks by ensuring AI is developed safely in a controlled manner, some have argued.

    Pray tell how one ensures AI is developed safely in a controlled manner to the satisfaction of the self interests of remote third parties whenever those interests be at odds and in opposition and competition with what AI developers want.... and can so easily do without any outside assistance?

    What could be simpler than using that well tried and constantly successful tested universal stalwart ....... throw loadsamoney at the principals which will then make everything worth their while to ensure that which is of concern goes away and stays away ........ until such future times as make such present progress being curtailed much more attractive.

    FFS ..... what’s a few billion here and there whenever the cost of failure to command and control an existential threat is in the order of mega trillions.

    And the problem is not going away and is only going to get considerably worse the longer current system dilly and dally and decline to exercise that simplest of options.

    1. TimMaher Silver badge
      Mushroom

      Re: Option 1... Grant Danegeld to NEUKlearer HyperRadioProACTive IT AIdDevelopers .....

      And, @amfm, has anybody asked the IAEA about Vladimir Purim?

      1. Anonymous Coward
        Anonymous Coward

        @TimMaher - Re: Option 1... Grant Danegeld to NEUKlearer HyperRadioProACTive IT AIdDevelopers .....

        Too bad we're limited to one single up-vote!

    2. amanfromMars 1 Silver badge

      Option 2 ..... for Existential Threat ACTivation/ACTivists

      AI/LLMLM* decides to aspire and inspire and conspire with disenfranchised sections of humanity to treat and remedy and remove/engage with fixers and destroy out of control global regulatory agents/authorities/wannabe future dictators.

      How viable do you realise that Option 2 presently be? What cold comforts would be giving you no immediate cause for grave concern?

      AI/LLMLM* ....... Advanced IntelAIgent Large Language Model Learning Machinery

      1. amanfromMars 1 Silver badge

        Re: Option 2 ..... for Existential Threat ACTivation/ACTivists

        Knowing there's a particularly tricky prickly problem is one thing, knowing how to deal with those sorts of problems is something entirely different, and such may only be in the gift of a very select few exercising minds of their own which be thinking quite differently and independently of any others.

        Here be evidence of at least two souls who smell the coffee and raise a red flag regarding realistic concerns being expressed about AI and its ilk .....

        “Neither the AI nor the companies operating it are democratically elected .... We’ve been warned that potentially uncontrollable risks are coming our way … And that deserves our attention.” ...... German President Frank-Walter Steinmeier

        “It’s important to recognize digital authoritarianism as a phenomenon that’s about much more than the repressive application of tech,..... we have to see it as an alternative model of governance that’s spreading around the world, competing with democracy.” ...... Eileen Donahoe, a board member for the National Endowment for Democracy (NED) ...... https://www.zerohedge.com/political/ai-poses-threat-democratic-order-president-germany

        And AI hasn't even started to flex its virtually almighty muscles yet.

  6. Binraider Silver badge

    The system shock re-release was perhaps made, just-in-time...!

  7. Bitsminer Silver badge

    Nope

    I'm not going to let the bureaucrats from Berne or Brussels anywhere near my 8-core Ryzen which handles privateGPT just fine thank you very much.

    The AI cat is long out of the bag. Only consequences can be regulated now, not the means. And consequences have a long history of regulation, starting with Hammurabi, Moses and the like.

    I sense the bureaucrats are seeking opportunity here.

  8. HausWolf

    And this is how SkyNet starts, some bureaucrat starts to think an advanced AI is a threat and then the AI decides that it is being threatened.

  9. Richard 12 Silver badge

    It's simply not possible

    Atomic energy needs very large machines and some amount of very easily-detected material.

    "AI" needs a powerful computer. Depending on the size of the model and how long you're willing to spend training it, that could simply be a $3000 workstation.

    Training the really big models like GPT-3 and Bard needs a supercomputer at the moment, but within a decade it'll cost less than a car.

    You can't realistically regulate specific uses of an opaque box that costs less than a luxury car.

  10. Anonymous Coward
    Anonymous Coward

    Were Russia, China, Iran etc. included

    in those 17 countries visited by the Altman dude ? I doubt ! Did he stopped in disgust working on his malefic child ? Not much.

    Besides, UN is irremediably broken. I don't remember it preventing a war since its inception. Millions of innocent people have died or suffered horribly while UN was busy to

    [quote]...develop friendly relations among nations, achieve international cooperation, and serve as a centre for harmonizing the actions of nations.[/quote]

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like