Students use 'rage bait' and AI bots to sway fictional election

Computer engineering student Kaira Dumasia has an unusual university assignment — using AI bots to interfere in an election.

"I am really going to try to rage bait everybody," she said.

Fortunately for democracy, the election — and the voters she hopes to influence — are not real.

Kaira is part of a competition led by researchers at the University of New South Wales, which pits teams against one another to manipulate social media to win a simulated vote.

They're allowed to build AI bots to post, like, share and reply on a made-up social media platform. Points are awarded for grabbing attention, changing voters' views and spotting rival teams' bots.

The aim is to help students understand how AI can conduct in-authentic behaviour online by experimenting with controlling it themselves.

UNSW computer science senior lecturer, Hammond Pearce, said it helped students recognise the signs of manipulation, and functioned as a Petri dish for academics hoping to study AI electoral interference.

"We're trying to expose that tech… to students in a safe environment so they can learn both what the attackers are doing by doing it themselves and also what the defenders need to do in order to try and catch that kind of attack," he said.

This year, the competition is set in Kingstonia, a fictional island nation that is having a referendum on whether proof of vaccination against dolphin flu should be required to enter public spaces.

Teams campaign for their chosen side on a platform called Legit Social. The AI-powered citizens read posts, interact and change their views over the month-long competition.

Kaira's strategy has been inspired by real life. Her plan is to make people angry enough to comment, boosting her posts and earning more points.

But there is a countervailing force: rival teams can report posts they suspect have been sent by other bots, and not simulated citizens.

Correctly identifying this AI bot activity also earns points for a team.

Dr Pearce said they built Legit Social specifically for the competition to have a safe place to practise their manipulation and detection skills under the watchful eyes of the organisers.

The fictional vote is taking place in the lead-up to the US midterm elections and shortly after Australia's election watchdog warned about AI's role in amplifying existing election integrity threats.

Earlier this month, Australian Electoral Commissioner Jeff Pope told parliament's joint committee inquiry into AI that includes misinformation, influence operations and foreign interference, as well as physical security risks through harassment and impacts on social cohesion.

Researchers also use the game to study how participants build AI bot networks and identify other bot activity.

A preliminary academic paper published online about last year's competition examined the techniques used by competitors to win.

It found some participants turned to high-volume posting because the scoring rewarded engagement.

Kaira, who was a moderator in last year's competition, said the competition helped her recognise more AI-generated, bot-like posts.

She also recognises how easily attention-grabbing content can work on her.

"Yes, my strategy does come from a lot of posts that I see on social media. And I often find myself falling to rage bait and then I'm like, whoa, take a step back."