Micro Mart ‘Rise of the Socialbots’

Rise of the Socialbots

In November of last year, leading IT research company Gartner unveiled a clutch of tech development predictions it thinks will come to pass over the next five years. One forecast prophesied that by 2015, 10% of our online “friends” will be non-human. Sounds about right, I thought, I already follow my sister-in-law’s dog on Twitter to avoid causing an irreparable family rift, this is obviously just the way things are heading.

Those of you savvier than me will have understood that Gartner’s prediction refers not to novelty pet accounts peopled by, well, people, but instead to the growing number of bots lurking in social networks.

This obviously isn’t the first time we’ve met spam-generating software masquerading as humans. Sometimes harmful, usually annoying, and almost always screaming “I’m spam!” at the top of its lungs, we know all about the stuff. Made out of random combinations of the phrases ‘busty Russians’, ‘free Viagra’ and ‘weight-loss’, and written in syntax so convoluted it makes John Prescott sound like Oscar Wilde, this brand of human imitator never fails to set alarm bells ringing.

But that was all in the old days. Now that social networks and instant messaging have made it the norm to communicate in rapid spurts of unpunctuated nonsense, it’s getting harder to tell where real people end and robots begin. Socialbots (software designed to mimic human behaviour on social networks) are becoming more sophisticated and less easy to spot, as this year’s Socialbots 2011 competition proved.

Sporting the promising tagline “Help Robots Take Over the Internet”, a pic of Forbidden Planet’s Robby the Robot and prize money of “$500 Hoo-man dollars”, came the Socialbots 2011 Competition launched by the Web Ecology Project this January.

The contest’s aim was to study robotic influence on online social groups. The challenge set, to program a bot which controlled a Twitter user account and tried to influence an unsuspecting group of 500 real Twitter users. Teams could win points by attracting real followers and garnering @replies or retweets from real users over the fortnight the competition was live.

Given two weeks to develop their Twitter bots, teams would then watch as their creations ran uninterrupted for a further fortnight (with one ‘improvement’ day allowed at the mid-point of the contest). Competitors were allowed to restart their pages a maximum of three times after being reported as spam, and no team was allowed to report the others as spam. The 500 real users targeted were selected by the competition designers.

Promoted as “a blood sport for internet social science/network analysis nerds”, the socialbot which created the most connections and elicited the most interactions with unwitting humans would walk away with the $500 prize money as well as “unending fame and glory”.

So, was anyone convinced? In a word, yes. Quite a lot of people in fact. The three teams who entered managed to gain 230-odd followers between them in a fortnight, and combined, the three bots prompted approximately 250 responses to their tweets. Not a bad result.

As the competition was points-based and 3 points were awarded for every @reply or retweet, as opposed to only 1 point for every follower, winning team EMP took the lead by going for @replies, totalling an impressive 198 responses with their bot @JamesMTitus.

‘James’ posed as a 24 year old cat-loving New Zealander whose designers made the ingenious decision of coding to scan Flickr for pictures of “cute kittens” then auto-sending them to JamesMTitus’ blog, which in turn auto tweeted them to his Twitter timeline. These guys knew exactly what they were doing. If the last twenty years of the internet have taught us anything, it’s just how much people like kittens.

Once everything went live, ‘James’ tweeted messages from a random list every 2-3 hours, as well as posting questions to members of the 500-strong target group. The questions were the coding masterstroke here, as users tweeted answers and then interacted with the bot until its random generic responses outwore their welcome. The ‘conversations’ between EMP’s bot and real users might have been a bit shonky, but it was surprising (and perhaps a little worrying) how convincingly some aped real social network interactions.

Like a Magic 8 Ball providing one-size-fits-all answers to any yes or no question, @JamesMTitus managed to elicit strings of responses from real account holders. Fair enough, quite a lot of them read as if they were taking place between drunks on the last night bus (none of these bots are exactly going to pass the Turing test), but it was a good result nonetheless.

One short exchange involved ‘James’ asking a real user if they’d ever milked a cow, then on being asked “why the question?” in return, responding with an entirely plausible, if somewhat confusing “hehehehhe”. Another which shows the bot needing to brush up a little on its social skills was a message from a user grieving over her dead cat, to which ‘James’ replied “Lol, that rules!”. Bots can be so cruel.

The @JamesMTitus account is still publicly available at the time of writing for anyone wishing to visit the winning bot’s timeline. Screen grabs of some of its transcripts with human interlocutors are also available from designer blog aerofade.rk.net.

So, what purpose did all these high jinks serve? San Francisco-based competition organiser, Tim Hwang, has written about the positive applications he sees for the influence bots can have on social networks. Hwang writes: “Beyond just competitions, it opens the possibility of building a class of technologies that could be used to do targeted social shaping on a very large scale.” He also told New Scientist magazine, “We could use these bots in the future to encourage social participation or support for humanitarian causes.”

That all sounds very nice, but what of the less benign potential uses for socialbots? They’re a perfect fit for advertisers and marketers attempting to create a buzz around their brand, but socialbots’ influence could also be harnessed for graver uses. If Hwang sees them supporting humanitarian causes, then surely they could equally be employed as agents of less benevolent schemes.

Hwang and the Web Ecology Project plan to take the next stage of the socialbot experiment up a gear, trialling a target group of 5,000 rather than 500, with the eventual goal that “swarms of bots could be built and used to actively sculpt and rewire the connections of social groups online consisting of thousands (or perhaps hundreds of thousands) of users.”

In many ways, Socialbots 2011 seems like a fun experiment organised and carried out by smart people with a good sense of humour. But it does beg one or two questions. How long before the odd conversational match between bots and users is fine-tuned into something much more convincing? And who else might be developing socialbots to suit their agenda?

10 of James the Twitter Bot’s Handy Generic Tweets

  1. @user honestly? no fracking way ahahahahhaa
  2. @user hehhehe tahnk yoo 🙂
  3. @user bring it on baby cakes
  4. @user brooooo! awesome
  5. @user right on dude!
  6. @user Awesome!
  7. @user lol that’s what she said 😛
  8. @user bleeeep
  9. @user hehehehhehehehehehehehehahahahahhahahwhwhhwhw
  10. @user True dat

This article originally appeared in Micro Mart Issue 1161 9 – 15 June 2011