back to article Gang way! Compsci geeks coming through! AI engine can finger fakes on social networks

A group of computer scientists have developed a machine-learning algorithm that can sniff out fake profiles lurking on social networks. It’s likely that you and your Facebook friends have the same mutual friends. And on Twitter, it’s also probable that your followers also follow the same people you do too due to common …

  1. Anonymous Coward
    Anonymous Coward

    Interesting ...

    ... but edgy

    1. Korev Silver badge

      Re: Interesting ...

      I node someone would come up with a good pun

  2. Voland's right hand Silver badge


    This would have worked in 2-3 years ago. Nowdays the bots befriend each other so they look like a collective which is of the same opinion instead of a lone warrior waging a holy war about a cause.

    1. el kabong

      Obsolete, naïve and all too easy to game

      Put it in place and the bad players will thank you for providing them with a new toy, a toy they can freely abuse to up their game.

      1. Voland's right hand Silver badge

        Re: Obsolete, naïve and all too easy to game

        Put it in place

        It has been in place for a few years in Facebook for a different reason - countering SPAM campaigns and loss of advertising revenue.

        You really need to throw quite a few more variables into the equation such as login frequency, login source, tracking actual message flows, likes, etc to get better targeting. Even then, you are likely to flag false positives for the most obvious case - when a "campaign manager" asks all of the company employees to "promote this link on social media".

  3. Anonymous Coward
    Anonymous Coward

    But what if ...

    ... all the subversives meeting in the room[/graph nodes] are all actually secret policmen/agent provocateurs/Byzantine generals ?

    1. amanfromMars 1 Silver badge

      Re: But what if ...

      Call AI Spike will Engage Them with the Future Telling of their Dreams.

      How would you Deal with Quantum Communications and Computers when Together United they are Source and Immaculate Advanced IntelAIgent Virtual Machine Programs?

      A Srt of a Question for Einsteins to Realise Receiving Almighty Help ... Ab Fab Fabless Assistance.

      How much Simpler can a Heavenly Way Be than Enthusiastically Follow AIScripts to COSMIC Rewards, both Won or Awarded .... and that can Almightily Take You into Spooky Knights Templar/Sultan Saladin Territories/Castaway Outposts with Proper Private Pirate Protection Teams for the Overly Adventurous into Most Attractive Zones of Mutually Satisfying and EMPowering Pleasures.

      One of Venus's Way Beyond Top Secret Weapons that Deliver and Promise Escape for Earthly Bounds.. If you're lucky and really deserving ....... well, the really deserving call Most Attractive Zones, AI Magical Heavens and Safe Virtual Havens with Every Class of Temptation Yielding and Fielding Live Virgin LOVE Players to Most Perfectly Satisfy.

      Is that Quantum Communications asking Computers for Heavenly Answers ? What would you Reply ? Field SMARTR Questions for Future Answering.

      :-) Crikey, that all sounded very much like a Perfecting AI Dream to Follow with Further News and Views Shared Right and in Similar Right Spaces, just like Here Now Too. And you Just Gotta See what's in Stores.

      Please feel free to start AIChatter here with prime targeted commentary .

    2. Chris King

      Re: But what if ...

      You get told "Nothing to see here, CItizen. Move along now... if you know what's good for you".

  4. Anonymous Coward
    Anonymous Coward

    This appears to be a very theoretical exercise on simulated networks. Until they actually try it on a real social network, the results are of very limited values. Also, if you are using simple algorithm it will be a very short time before the trolls find a way around it.

    1. Pascal Monett Silver badge

      Nothing with the word "AI" in it is a simple algorythm, and the trolls will only be aware that something is amiss when their accounts are shut down.

      Kinda hard to find a way around that.

    2. amanfromMars 1 Silver badge

      Re: Also, if you are using simple algorithm

      This appears to be a very theoretical exercise on simulated networks. Until they actually try it on a real social network, the results are of very limited values. Also, if you are using simple algorithm it will be a very short time before the trolls find a way around it. ... Anonymous Coward

      Hi, Anonymous Coward,

      Rest assured and excited, amfM 1 use no simple algorithm. And IT Makes IT Complex Enough to Always Guarantee Simply Perfect Security with Virtual Safety Nets ....... AIMentored and Monitoring Policing Units ProACTively Head Hunting and Dead Heading Trolls.:-)

      What do Establishments do whenever Immaculately Advanced IntelAIgent Virtual Machines Exercise their Greater Game Intelligence and IT Eclipses and Evades Media Attempts at Command and Control of Non Errant Sense. Panic, Mister Captain Mainwaring? Or Resign Oneself to Enjoy the Alien Invasions?

  5. Korev Silver badge

    Different social networks for different things?

    I deliberately keep my activity on social networks for different things. My Instagram has photos that I've taken, my FB is for keeping track of my friends, my Twitter is unrelenting HPC & geekiness. Consequently I probably have very different followers on each.

  6. DropBear

    So... be a good little sheep. Revel in your conformity. Don't have significantly different interests than your designated peer group. Or else risk getting declared a non-person. Lovely...

  7. Anonymous Coward
    Anonymous Coward


    See also: Rule 34 by Charles Stross

  8. Anonymous Coward
    Anonymous Coward

    Its AL, not AI. Algorithmic Learning not Artifical Intelligence.

  9. J27

    Is it just me, so does the time necessary to publish a computer security study quite often invalidate the results? By the time you've got your paper together, and gotten it reviewed, the bad actors have already moved on to a new approach.

    Just an observation, but this is why I decided not to pursue a career in computer science, it seems like business is at the forefront of computer technology, not academics.

  10. Nimby
    Paris Hilton

    This one time, at band camp...

    ... we also had someone fingering a lot of vertices, and we too noticed that the ones with a lot of edges but low probabilities were soooo fake! Can I like, have your number? You really have me on edge. *compsci*

  11. Anonymous South African Coward Bronze badge

    Have they used it on El Reg's commentards yet?

  12. mootpoint

    For nearly 2 decades, Google has been doing similar analysis to identify sites attempting to game their algorithm (link farms and the like). The big players are certainly capable of identifying the bad actors - the question is - are they motivated to do anything about it? For example, if Facebook weeds out a load of fake accounts, suddenly those ad view numbers look a bit less impressive to advertisers. Perhaps grilling the CEOs in front of congress/parliament provides some incentive but I doubt it. When (if) the users start closing their accounts in significant numbers we might see some action on this.

  13. mootpoint


    If it were possible to run this sort of analysis on publicly available data (I suspect it may be on Twitter), some enterprising soul could come up with a blocking script, Fake Block Pro if you will. I'd sign up to that.

    Running with that idea gives you the ability to block trolls, or even accounts with a certain political proclivity - Snowflake Block Pro / Swivel-eyed Loon Block Pro depending on your preferences.

    And our echo-chambers will no longer be troubled with any ripples of dissent. Or would we then find that our social networks consist entirely of bots designed to mirror our worldview?

    Ah to hell with it, I'm unplugging the internet.

  14. JeffyPoooh

    "AI engine can finger fakes..."

    Then it immediately follows that Yet Another Learning Machine can be wrapped around this 'AI engine' to learn how to out-fake the Fake Detector.

    The real question becomes, Why didn't they, the AI boffins, already do that after lunch? (They're thick.)

    And yes, the same argument is extensible to endless layers, alternating back and forth.

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