back to article This won't end well. Microsoft's AI boffins unleash a bot that can generate fake comments for news articles

As if the internet isn’t already a complicated cesspool full of trolls, AI engineers have gone one step further to build a machine learning model that can generate fake comments for news articles. The eyebrow-raising creation, known as DeepCom, was developed by a group of engineers at Beihang University and Microsoft, China. “ …

Page:

  1. Richard 12 Silver badge
    WTF?

    If you could try *not* breaking the Internet

    That'd be great.

    There's already too many spam posts being made that are a partial copy-paste of earlier comments, or bits of the article with a few dodgy links, without making the filter (and moderators) have to work even harder.

    There is literally no useful purpose for this. Not one. It might even be illegal in some jurisdictions - misleading etc

    1. jake Silver badge

      Re: If you could try *not* breaking the Internet

      Bots didnl't break BBses, nor did they break IRC or Usenet. What makes you think they have the power to break the entire Internet? Not even the combined evils of goophabet.tube, instaface, twitter and spammers have managed that yet!

      1. jake Silver badge

        Re: If you could try *not* breaking the Internet

        Odd. 33 downvotes in 12 hours, and I wasn't even trying. And yet not a single commentard has stated why they downvoted my post ... C'mon, folks, share! Us trolls want to know what pushed your buttons.

        Perhaps you think bots actually did break BBSes, IRC or Usenet?

        Perhaps you think a trivially filtered bot can break all of The Internet?

        Perhaps a passing flock of goophabet.tube, instaface, twitter and/or spammer lovers passed through and the downvotes are an artifact of this event?

        Perhaps you think one or more of the trivially filtered goophabet.tube, instaface and twitter have already b0rken the Internet? (spammers aren't as trivial to filter, but filterable they are, at least for the most part; regardless, they haven't managed to break the Internet).

        Speak up! Inquiring minds want to know! :-)

        1. Not also known as SC

          Re: If you could try *not* breaking the Internet

          @Jake,

          I've only just read this article and can't see anything wrong with your comments so I'm going to up vote you because I agree that bots won't break the internet. They might break parts of the internet, for instance social media etc, but that will just be the internet evolving.

        2. Tom Paine Silver badge

          Re: If you could try *not* breaking the Internet

          It's really pretty obvious, if you look back at your comment and have another think about it.

          (Edit - I didn't downvote you myself)

        3. Geezheeztall

          Re: If you could try *not* breaking the Internet

          Perhaps the bots disagreed.

    2. jmch Silver badge

      Re: If you could try *not* breaking the Internet

      "If you could try *not* breaking the Internet"

      a bit pedantic but this won't 'break' the Internet. It will render parts of the internet useless. Hopefully it will lead to sites exerting more control over what posts they allow.

      (from their claim)"“Automatic news comment generation is beneficial for real applications... " bollocks

      "There is literally no useful purpose for this." spot on

    3. bombastic bob Silver badge
      Meh

      Re: If you could try *not* breaking the Internet

      when you consider the activist trolls, it might as well be bots anyway.

      'Teh Intarwebs' will survive so long as gummints don't try to LEGISLATE it. Something about free people exercising their freedom, actually BEING free, not restricted on phrasing, linguistics, expressiveness, or use of words that make snowflakes cry, etc. Some places STILL DO exist without moderation, and USENET is one of those. And most of the time the mods aren't brown-shirt fascists. Some are, over on Fa[e]ceb[itch,ook] and Tw[a,i]tter, but that's another topic.

      And who said someone ELSE didn't already have a "fake news" bot - or a room full of paid activists, same thing - gumming up the works and trolling everyone/everything on Tw[i,a]tter and Fa[e]ceb[ook,itch].

      1. RM Myers

        Re: If you could try *not* breaking the Internet

        DeepCom, is that you?

      2. Palpy

        Re: Don't legislate, leave the Internet to...

        ...rooms full of paid activists [and bots] "gumming up the works and trolling everyone/everything...."

        Logical disconnect: no regulation, ie a true laissez-faire Internet policy, results in domination of said Internet by Alphabet, Facebook, Amazon, troll farms, malware-slingers, and bots. Lets abbreviate the whole cesspool as AFATFMSB.

        How then, Bob, do you keep AFATFMSB in check? Rely on the wise morality of Mark Zuckerberg? Count on the gentle restraint of the paid trolls in Macedonia? Trust Alphabet to limit its intrusion into the privacy of netizens out of benevolence and concern for humanity over profit?

        Didn't we just see an article on the worthlessness of companies self-certifying themselves as "good advertisers," the upshot being that a large percentage didn't bother to actually abide by the rules they themselves agreed? Don't we already know that, given a laissez-faire system, corporate and private bad actors will, as you write, gum up the Internet and troll everyone and everything?

        Really, Bob, I'm serious. What do you see in the real world as we know it which actually counteracts AFATFMSB and all the rest -- without legislative teeth involved?

  2. jake Silver badge

    Old news is old.

    In 1972, ELIZA (as "The Doctor", at BBN (tenex?)) and PARRY (at SAIL) had a conversation at the first ICCC ... Well, they had a conversation that was followed over the ARPANET during the ICCC. It was immortalized in RFC 439.

    More leftovers from SAIL here. Not much has changed in 47 years ... right, amfM?

    1. Ugotta B. Kiddingme Silver badge

      Re: Old news is old.

      "So you say Not much has changed in 47 years ... Tell me more."

  3. Blockchain commentard Silver badge

    Open dialogue allows people to discuss their opinions and share new information. Coming from China that's rich.

    And this comment was generated by a human (which is what I'm programmed to say!!!!).

    1. jake Silver badge

      Yeah, sure, right.

      A human would have made the ubiquitous pseudotypoe "!!1!", as any ful no.

    2. LDS Silver badge
      Big Brother

      "Open dialogue allows people to discuss their opinions and share new information. "

      That's why China wold like to bury it under a tons of auto-generated useless comments...

      1. Ian Emery Silver badge
        Black Helicopters

        Re: "Open dialogue allows people to discuss their opinions and share new information. "

        This is going to save China a lot of money, but cost over a million jobs.

        The jobs of the million+ they currently employ to post propaganda every day.

        Just look at some of the wonderfully dishonest stuff posted by "Peking Duck" on Quora. Apparently the CCP has peaceful and loving relationship with its neighbours, never gets into an argument with them, and the only threat to the region comes from the US.

        Oh bugger, there goes my next Chinese tourist visa.

    3. Anonymous Coward
      Anonymous Coward

      And this comment was generated by a human

      I must confess, I already no longer know whether I'm a bot or just a bot's bot. Let's chat about it somewhere private, eh?

      1. Anonymous Coward
        Anonymous Coward

        Re: And this comment was generated by a human

        "Let's chat about it somewhere private, eh?"

        You want to go somewhere private and chat about your "bot" to someone you just met on the internet? OK, you're human...

        1. Anonymous Coward
          Anonymous Coward

          Re: And this comment was generated by a human

          well, isn't it bot-designers' dream to make bots behave like humans?

      2. herman Silver badge

        Re: And this comment was generated by a human

        You must be a Canadian bot, eh.

  4. pavel.petrman Bronze badge

    In sweet memory of Tay

    the AI tweet robot from Microsoft.

  5. Shadow Systems Silver badge

    No thanks, we've already got one...

    Isn't that true, A Man From Mars, Bob, & Shadow Systems?

    *Blink blink*

    HEY! I am not a bot! I'm a Human, I swear! Fuck your CAPTCHA with a power drill damn it, I am TOO Human!

    *Crosses arms & hurrumphs grumpily*

    =-)P

    1. A Non e-mouse Silver badge

      Re: No thanks, we've already got one...

      Amanfrommars's output is far too complex (& almost coherent) for an AI to generate.

      1. Dave 126 Silver badge

        Re: No thanks, we've already got one...

        Bug powder dust and mugwump jism. Aman is inspired by William Burroughs through a Gibsonian 'invisible literature' filter. Class 1 Laser Product.

        In contrast, Shadow Systems often makes reasoned comment on the accessibility of systems and services by those visual impairment, as well as on many other topics.

        My cousin makes $50,000 a week working from home. You can too!

        1. Doctor Syntax Silver badge

          Re: No thanks, we've already got one...

          Is this talk like a bot day?

          1. Psmo Bronze badge
            Terminator

            Re: No thanks, we've already got one...

            Bleep bloop take me to your leader.

            I, for one, welcome MemeNotFoundError at line 94

            1. Rameses Niblick the Third Kerplunk Kerplunk Whoops Where's My Thribble? Silver badge

              Re: No thanks, we've already got one...

              +++MELON MELON MELON+++

              +++Error At Address: 14, Treacle Mine Road, Ankh-Morpork+++

              +++Divide By Cucumber Error. Please Reinstall Universe And Reboot +++

              +++Oneoneoneoneoneoneone+++

              1. Michael H.F. Wilkinson Silver badge
                Coat

                Re: No thanks, we've already got one...

                Quick!! HEX needs to work with the FTB enabled!

                Sorry, couldn't resist

              2. Someone Else Silver badge

                Re: No thanks, we've already got one...

                Close B Close Mo on Deputy Dan....

                1. Anonymous Coward
                  Anonymous Coward

                  Clippy helps you out

                  I see you're trying to write a comment on a news item. Would you like Microsoft to help you with that?

        2. Tom Paine Silver badge

          Re: No thanks, we've already got one...

          Leave Justin Warfield out of this, please

    2. I ain't Spartacus Gold badge

      Re: No thanks, we've already got one...

      Shadow Systems,

      But are you sure your screen reader isn't using your forum log-ins while you're away from your computer and posting comments all of its own?

      I'll join you in fucking CAPTCHA with power tools though. The latest ones which show you images from something like Google Street View and ask you identify traffic lights are particularly annoying. If they were decently sized images - that would at least be a tiny bit less annoying. Though still as annoying as fuck - but given how visually impaired unfriendly they all are, it's amazing that the audio CAPTCHAs that I@ve tried have been totally impossible. And I've got perfect hearing - but am forced to rely on my far from perfect eyesight anyway.

      1. Shadow Systems Silver badge

        At IAS...

        This is his screen reader. Shut up about me using his computer when he's not around. It's the only way I get to have any fun! =-D

      2. Ian 55

        Re: No thanks, we've already got one...

        Or expect you to know US English: "crosswalk".

  6. Roml0k

    the commentards are going to have a lot of fun in this article .

    1. macjules Silver badge

      The comments it generates are short - on the order of tens of words - and aren’t complex enough to incite much reaction.

      So, only the Daily Mail or the Daily Telegraph (aka The Mouth of Sauron Downing Street) commentards then.

  7. revenant Silver badge

    "The paper didn’t mention any potential malicious applications of the technology"

    No need. The number of worthwhile uses is so low that it is safe to assume that it would be used mostly for malicious purposes.

    There is no pressing need for this sort of technology in spaces where the primary purpose is for people to communicate with eachother.

    One day, when an AI is sophisticated enough to be classifiable as a 'Person', then of course he/she should be allowed to comment. Otherwise, best keep them out of spaces that are already minefields of piss-poor human interaction.

    1. Mike 137 Bronze badge

      "One day, when an AI is sophisticated enough to be classifiable as a 'Person'"

      It never will be. Masses of current neurophysiological and neuropsychological research emphasise that the brain is primarily an interface between the world and your body. It's also vastly complicated, much more so than any conceivable piece of IT. Consequently, the model of AI is fundamentally irrelevant.

      The brain's capacities for pattern matching and weighted reasoning (what AI relies on) are not what it's for - they're just part of how it accomplishes what it's for - orienting the person in the environment with survival and perpetuation of the individual and the species as primary drives. See Damasio "Descartes Error" for a very readable account.

      1. revenant Silver badge

        Re: "One day, when an AI is sophisticated enough to be classifiable as a 'Person'"

        I think 'Never' is a bit strong, but on the whole I agree with your sentiments. I don't believe AIs are anywhere near being 'people', and they may never be (but if they can achieve that state then they should be allowed to join in).

        Of course, that isn't the issue here as Microsoft and others who would misuse the AIs will be satisfied with something that merely passes for human in the narrow context of a comments section.

        1. amanfromMars 1 Silver badge

          Re: "One day, when an AI is sophisticated enough to be classifiable as a 'Person'"

          I don't believe AIs are anywhere near being 'people', and they may never be (but if they can achieve that state then they should be allowed to join in). .... revenant

          How about the reverse/converse/obverse, revenant? ....... certain people being AI clones and/or drones?

          Are you anywhere near being sure that may never be believable ..... or are you realising there are very strange powerful forces and sources at their work constantly and you are practically little more than just a spectator until you get your act together and Go with the Flow/Run with the Program .... and AI in Remote Virtual Command and Relatively Anonymous Autonomous Control with Computers and Communications of CHAOS [Clouds Hosting Advanced Operating Systems]?

    2. Teiwaz Silver badge

      Re: "The paper didn’t mention any potential malicious applications of the technology"

      Otherwise, best keep them out of spaces that are already minefields of piss-poor human interaction.

      There's one possible beneficial use case - Youtube.

      As a 'minefield of piss-poor human interaction', it's got a bigger problem than Somalia (that's regarding 'minefields', not piss-poor.... Ok, that as well probably). Letting a few thousand loose on the Youtube comments might actually dilute the god-botherers, racists and plain 'ole ignorant and ill-informed pontificating some.

      Certainly couldn't make the Youtube comments sections any worse.

      1. I ain't Spartacus Gold badge

        Re: "The paper didn’t mention any potential malicious applications of the technology"

        The AI in 'War Games' had the Youtube comments problem sorted out years ago. The only way to win, is not to play.

      2. J. Cook Silver badge
        Joke

        Re: "The paper didn’t mention any potential malicious applications of the technology"

        Certainly couldn't make the Youtube comments sections any worse.

        And it just might make them better.

  8. Kabukiwookie Silver badge

    Skewing opinion

    The only obvious world application for this would be to skew opinion one way or another by vastly spamming opinions that align with the bot operator.

    Reviews of your movie overwhelmingly bad?

    We can 'fix' that.

    This is already being done, but fake reviews can usually be fairly easily spotted. If 'positive' reviews would be longer and somewhat coherent, they'd be much harder to spot.

    Don't like 'The last Jedi'? Looks like you're no longer part of the majority.

  9. A-nonCoward
    Mushroom

    REALLY obligatory XKCD

    I mean, who needs to read the article (I didn't) when everything you need to know about this topic is already in https://xkcd.com/810/

    1. Ian Emery Silver badge
      Boffin

      Re: REALLY obligatory XKCD

      Ah, but are you a bot programmed by XKCD ??

      1. A-nonCoward
        Gimp

        Re: REALLY obligatory XKCD

        how did you know?

    2. Dave 13

      Re: REALLY obligatory XKCD

      Post of the week. You win the internet.

  10. Filippo

    I thought this had been done and deployed widely a long time ago. Sure feels that way.

    1. Paul Kinsler
      Joke

      I thought this had been done and deployed ...

      Indeed.

      But I wonder if you could use it as a filter; i.e. test user-generated posts against outputs from the bot, and if the match is too good ask the user to improved their post ...

      "Your proposed content is equivalent to something generated by a low-quality machine-learning network. Please improve your post and try again"

      :-)

Page:

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

Biting the hand that feeds IT © 1998–2019