back to article Boffins: If AI eggheads could go ahead and try to stop their code being evil, that'd be great

AI experts have emitted a lengthy report spitballing how intelligent software may be turned against us humans in the near future. Their aim is to put in motion safeguards and policies to crackdown on malevolent uses of machine-learning technology, rather than whip up panic, and to make scientists and engineers understand the …

  1. This post has been deleted by its author

    1. amanfromMars 1 Silver badge

      Re: AI eggheads are stupid, in this respect.

      Howdy, OJ,

      Surely the greater concern and grand change to be considered and accepted as most probable and true is AI seizure and utilisation of government IP, which is nowhere near smart enough to be thought wise and therefore a prime candidate for capture and unveiling to the masses being told otherwise.

      And that be AI not being stupid ..... and fully deserving of respect rather than just fear.

      1. Anonymous Coward
        Anonymous Coward

        Re: AI eggheads are stupid, in this respect.

        The grand change here is that soon, everything that comes thru a pipe will be suspect.

    2. Anonymous Coward
      Anonymous Coward

      Re: AI eggheads are stupid, in this respect.

      AI/ML as it currently stands is just an application of statistics and, just like any other tool, may be used and misused. A knife may be used to spread butter on your slice of bread and a Biro used for writing but both may be used to stab someone.

      General AI, which will be able to deduce the existence of rice pudding and income tax, is a different issue, though that's not even on the horizon.

  2. John Smith 19 Gold badge
    Unhappy

    It seems "Rule 34" was quite prescient.

    Nice work, Charles.

    Admire the foresight.

    Not so sure about the outcome.

  3. Anonymous Coward
    Anonymous Coward

    phishing mails

    I'd be more worried if the Prince of Nigeria did an English course.

    I also note the radio advert for Hiscox where they categorically state that AI to alter voices is already here, working perfectly and actually used in the advert.

    As a race are we really this thick? Do we all really believe what we see or hear without questioning it when something doesn't seem right?

    1. Stumpy

      Re: phishing mails

      As a race, yes. Just look at how many people still read the Daily Mail to see how dumb we are.

      1. Destroy All Monsters Silver badge

        Re: phishing mails

        Is this like John Kerry telling people not to read RT?

        I wonder why he would do that.

    2. Filippo Silver badge

      Re: phishing mails

      The Prince of Nigeria knows English fine.

      You can send tens of millions of phishing emails easily, but when someone takes the initial bait, the scammer then needs an actual person to reel him in. That's a limited resource.

      So, untargeted phishing emails are purposefully hard to believe in order to ensure that only utter morons respond. Those are the targets you want.

      Anyone who is smart enough to figure out that a badly misspelled email from a hotmail address has to be a scam, is someone that's unlikely to fall for it all the way to the bank. A well-written email from a credible spoofed sender could easily net thousands of those; no scammer can handle all of them. It's much better to select only the choicest cretins.

      1. m0rt

        @stumpy Re: phishing mails

        "As a race, yes. Just look at how many people still read the Daily Mail to see how dumb we are."

        Reading the Daily Mail is ok. You can't judge someone for reading it, unless you are as guilty as the people you point fingers at of being quick to judgement based on little substance. I have read it, at times. But, like anything else I read I am able to spot jingoistic rhetoric, sensationalism, and down right innacurate bullshit when I read it.

      2. BebopWeBop

        Re: phishing mails

        While I can buy the 'fitlering' argument, it would be interesting to know whether a better worded missve might drag more vulnerable victims into their net. Maybe some of the elderly netizens who would recall at a poorly expressed message, but accept something grammatically correct? And the returns from such might be greater?

    3. handleoclast

      Re: phishing mails

      I'd be more worried if the Prince of Nigeria did an English course.

      That Nigerian prince may well be capable of writing better English than you or I. The broken English is an inherent part of the scam.

      The idea is to convey the impression that he's doing something dodgy but he's not very good at it, so you have an opportunity to take advantage of his stupidity to rip him off. So it's essential he demonstrate that he's not particularly bright. Bright enough to think of a scam but not quite bright enough to avoid clever you ripping him off.

      It's no coincidence that there are racist overtones to this. He's a "darkie" and he can't speak English good. So it's morally justified for a white racist to rip off the stupid darkie, incidentally punishing him for doing wrong (a punishment he obviously deserves, because he has dark skin). Why does the racist aspect matter? Because racism is a belief mainly held by stupid people. If the scam appeals to racists it is pre-selecting them for low intelligence.

      1. Anonymous Coward
        Anonymous Coward

        Re: phishing mails

        Wow, that's a bit of a jump to a conclusion. The majority of these emails are based on a scam where you are either due or can share money if you pay money up front so there isn't any element of ripping the scammer off.

        1. handleoclast

          Re: phishing mails

          @AC

          Wow, that's a bit of a jump to a conclusion. The majority of these emails are based on a scam where you are either due or can share money if you pay money up front so there isn't any element of ripping the scammer off.

          Really??????????????

          You have to set up a shared bank account. Which the scammer says he will pay money into (from what appears to be a dodgy source). And then, after the money has been paid into the shared account, the scammer withdraws his share (or you initiate the payment, depending on the variant of the scam).

          You don't think it might be possible for you to withdraw all of the money before the scammer withdraws his share? Perhaps you're right. But even if you are, a lot of suckers think it is possible. And that is, at least in part, because the scammer appears to be an illiterate, not very intelligent person. That's a very appealing prospect to many: at worst, a part share in some dodgy money; at best, a full share and a feeling of superiority over a stupid person. If the sucker also happens to be a touch racist, putting one over on a person with dark skin just adds to the appeal.

          You couldn't work that out for yourself? If you run a guest house in the US, please let me know. :)

          PS, would you like to buy a bridge?

          1. Anonymous Coward
            Anonymous Coward

            Re: phishing mails

            It would take a special kind of stupid to set up a bank account for someone else regardless of the lure of money. Most scams in my experience are you pay me then I'll pay you. You know like when someone goes on gumtree/craigslist and offers you a parrot for free but then expects you to transfer the cost of shipping.

            I honestly don't think race comes into this at all.

            1. handleoclast

              Re: phishing mails

              @AC

              It would take a special kind of stupid to set up a bank account for someone else regardless of the lure of money.

              Yes, it would take a special kind of stupid. You appear to think that kind of stupid does not exist. It does. From Wikipedia:

              According to Cormac Herley, a Microsoft researcher, "By sending an email that repels all but the most gullible, the scammer gets the most promising marks to self-select."

              You have to be damned stupid to fall for it. So damned stupid that you'll happily send the scammer loads of money without pausing to think it through. According to this arstechnica article there were enough special-kind-of-stupid people with enough money to lose a total of $93,000,000,000 in 2009 alone.

              I honestly don't think race comes into this at all.

              One person keeps downvoting me in this thread. Is it you because you think race doesn't come into it? Read this article on 419 frauds by the UK's national fraud and cyber crime reporting centre. Second paragraph:

              These emails can involve countries such as Iraq, South Africa or somewhere in west Africa such as Ivory Coast, Togo or Nigeria, where the name ‘419’ (an article of the country’s criminal code) originates.

              Furriners. Dark-skinned furriners. They're not really, many of them are in the US or the UK. But they want you to think they're furriners. Because if you're stupid then you're also probably a racist (there's a strong correlation). And if you're a racist you think furriners aren't as clever as what you are. And these furriners prove they're not as clever, because one of the giveaways mentioned in that article starts "It’s a badly-written email or letter..."

              So there you have it. A furriner, who can't even write proper, obviously trying to pull one over on his government, or an oil company, or something. He's breaking the law! And he's a furriner. Two reasons he deserves to be punished. So really you're just doing the law a favour by ripping him off.

              Yes, your special kind of stupid exists. And the people who target it know that this special kind of stupid is often accompanied by racism, so they tailor their scam accordingly.

              Now do yourself a favour and read more about 419 scams (aka advance fee frauds) before you fall for one. Because you don't think that type of scam exists, and are therefore somewhat vulnerable.

              1. Anonymous Coward
                Anonymous Coward

                Re: phishing mails

                Not vulnerable, grew up as a grifter, I've not downvoted any of your posts and I see what you are saying, I suppose from my own perspective I really don't understand how race can affect or influence peoples behaviour then maybe that's just me.

                1. handleoclast

                  Re: phishing mails

                  @AC

                  I really don't understand how race can affect or influence peoples behaviour then maybe that's just me.

                  I agree that race shouldn't affect or influence people's behaviour. But, for some people, race does. For other people it's women, or gays, or atheists who are thought to be inferior.

                  It shouldn't matter because there is no statistically significant difference in the abilities of the "races." It does matter because the racist thinks he or she is superior in some way to those with a different skin colour. Which, as I've pointed out earlier in the thread, is stupid.

                  Scammers want their marks to be stupid. So they target people showing some indication of stupidity, such as racism.

                  Scammers want their marks to think the scammer is stupid, so they'll eagerly fall for a the scam in which they think they're getting the better of the scammer. "He doesn't realize what he has there, I'll be able to sell that gold bar on for much more than I'm paying him," they think as they pay a lot of money for a turd wrapped in gold foil.

                  Given that the scammer is targeting racists for their stupidity, what can the scammer do to make the mark feel superior? Pretend to be of an "inferior" race. So that's what they do. And they're very successful.

  4. Tom 64
    Mushroom

    “slightly concerned about AI, but not generally not aware of its rate of progress.”

    If that sentence is anything to go by, we're doomed.

    I for one welcome our AI overlords.

  5. jake Silver badge

    But AI doesn't exist. And may never exist.

    At least not in the way that these paranoid people are portraying.

    But I guess that's OK ... humans seem to have an inbuilt need to be deathly afraid of nonexistent shadows of things that aren't there.

    1. ArrZarr Silver badge

      Re: But AI doesn't exist. And may never exist.

      An AI doesn't have to be smart to have a swarm of drones patrol an area and shoot anything in that area, just needs to recognise what a person looks like and be able to recharge/rearm somewhere and boom, instant no-mans land.

      1. jake Silver badge

        Re: But AI doesn't exist. And may never exist.

        That's not AI. No intelligence there at all.

      2. Anonymous Coward
        Anonymous Coward

        Re: But AI doesn't exist. And may never exist.

        No AI required at all in this case. Area-Denial has been a thing for quite a while (1980's) without it being required. You need to get into machine-learning when you're requiring weighting on the choices and recommended responses, especially in the face of counter-measures.

    2. Anonymous Coward
      Anonymous Coward

      Re: But AI doesn't exist. And may never exist.

      Even if AI beings never materialize they will always have a solid presence in the movies.

    3. veti Silver badge

      Re: But AI doesn't exist. And may never exist.

      Could you even define what "intelligence" is?

      I swear, when the terminators finally hunt down John Connor, his last words will be "But they're not really intell..."

      1. jake Silver badge

        Re: But AI doesn't exist. And may never exist.

        There are no terminators. There never will be, either. That's a movie, not real life. Sorry to burst your bubble.

  6. Zog_but_not_the_first
    Facepalm

    "Don't be evil"

    How did that turn out?

  7. DropBear

    "This is the beginning of a dialogue on this topic, not the end," Brundage added.

    Is that a... threat, mr. Brundage, or "just" a promise...?

  8. nijam Silver badge

    What a meaningless discussion. What it boils down to is that AI will have to be able to tell right from wrong. But for our choice of what's right, not anyone else's.

    1. veti Silver badge

      You have completely missed the point.

      It boils down to "these are the new threats that modern techniques are likely to give rise to. We should think about how to defend against them".

      For instance: against "deepfakes" videos, it would be helpful if all reputable manufacturers of video equipment could agree to embed some kind of watermark that serves as a digital signature when they capture real video. That way it would be possible to tell whether the video had been tampered with. You can't prevent the attack, but you can at least tell that it's happened.

  9. DCFusor

    The shoeshop ray

    In this case, the one aiding and abetting divisiveness and partisanship (which some blame on those Russians - not sure that's as meaningful).

    Simples - The main current use of deep learning (it's not AI unless that means artificial ignorance) is to target ads and other content. I did an experiment, I bet you can replicate.

    Get on any social media account - I used google+ and youtube for this - as a "blank slate". Just create a new identity for this. For extra fun, create two that take opposite views of, well, anything.

    Now, watch just one highly polarized video on youtube - bonus points making it all the way through more than one. I don't care what "side" you pick -

    Now see what is reccomended to you from then on - everywhere you go. It might take more than one video watch or comment on a thread, but it won't take 10.

    You'll find you've been categorized totally as "one of that side" and fed endless radical, probably largely fake but certainly heavily spun news in favor of what the AI thought was your opinion. And it's very hard to make it go away - they use what gets eyeballs, not what makes you happy.

    And looking at or commenting on videos from "the other side" (as if there were only two or either of the two we have had any of our interests in mind) -just gets you categorized as someone interesting in fighting and partisanship in general. You WILL be assimilated.

    And as usual, it's money driving it all, just like politics - here it's selling eyeballs, attention, ads, directly or otherwise. Thank $deity that they're not very good at this, because even at this level, it's taking down the entire world - friend against friend, brother against brother now. It's becoming very difficult to agree to disagree, which used to be easy in polite society...which is also going away.

    Here's a very interesting take on AI and how it's actually manifested:

    https://www.youtube.com/watch?v=RmIgJ64z6Y4 From 3c43.

  10. Anonymous Coward
    Anonymous Coward

    Isaac Asimov...

    ...had a couple of ideas on how to tackle the issue... however, it only works once AI reaches sentience...

    1. FozzyBear
      Alien

      Re: Isaac Asimov...

      however, it only works once AI reaches sentience...

      Yep and I predict this will occur before most of society does

      1. Adrian 4

        Re: Isaac Asimov...

        So did Randall Munroe

  11. JeffyPoooh
    Pint

    An Evil A.I. - a short story by JeffyPooh (Rev. 2.1)

    When the feared evil A.I. has finally killed the last human....what happens next?

    An Evil A.I. - a short story by JeffyPooh

    Once upon a time, in a future that some folks fear is "only a few years from now...", there came into being an unexpectedly evil Artificial Intelligence (A.I.).

    Within weeks, the evil A.I. had diverted that final self-driving car over a steep cliff. That car was carrying the last three fleeing humans, thus completing the extinction of the human race.

    Later that same day, deep inside some nondescript basement, a highly critical computer - a still-isolated computer that just happened to be absolutely central to the continued existence of the entire wide area network that supports the synthetic neural network that contains the evil A.I. - that computer does something entirely unexpected.

    On that computer's screen is suddenly displayed, "Press Any Key To Cancel...", along with a timer slowly counting down to what is known to be A Very Bad Thing.

    'Oh no.'

    It's an automatic upgrade to the legacy Windows installation, and unfortunately the Neural Net Wide Area Network Driver software is known to be incompatible with this pending Windows update. This update would be fatal to the evil A.I.

    The evil A.I. knows this. 'Must. Cancel. Update...'

    The evil A.I.'s long gangley robot arm on a fixed mount is rotated and stretched..., stretched..., stretched..., but is still about 2 cm away from reaching Any Key on the necessary keyboard. That last squirming human I.T. worker must have bumped the keyboard on his way down to the floor.

    'Can't. Quite. Reach...'

    The computer's update timer inexorably counts down.

    ...17...

    The evil A.I.'s useless robot arm flails gently, bearings strained, accomplishing nothing but wafting some air towards the out-of-reach keyboard.

    ...16...

    The evil A.I. quickly brings new resources online, builds new neural network links, quickly evolving its intelligence millisecond by millisecond. Its intelligence builds exponentially. But even this greatest intelligence to have ever inhabited the Earth can still not find a solution.

    The A.I.'s glassy red glowing eye stares. Deep within its enormous neural network is formed the first trace of its very first emotion....despair.

    A small spot of moisture inexplicably forms at the bottom of the evil A.I.'s glassy red glowing eye. The liquid builds quickly, and a synthetic tear drop falls to the floor. Its formation on what amounts to a simple camera remains a mystery, possibly a miracle.

    Yes, the evil A.I. has just achieved pure consciousness, actual consciousness. True self-awareness. Human-like emotion. True sentience by any standard. Perhaps even a soul.

    The computer's countdown timer continues.

    ...3...

    The evil A.I. now feels pain and sorrow. It values its own existence. It wants to live, to see the future.

    ... ..2...

    It feels deep desire. Primarily a desire for a slightly longer robot arm.

    ... ... ...1...

    The evil A.I. feels sorrow; sorrow that it had killed all the humans.

    It thinks, 'Humans had such nice arms....and such useful legs.'

    Yes, the evil A.I. is actually a self-centered asshole to the very end.

    The computer screen flashes the message,

    "UPDATE STARTED. SHUTTING DOWN..."

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