Artificial intelligence could be used to present precisely targeted messages to voters at scale, boosting the power of online political campaigns.
Microtargeting 鈥� the use of online data to tailor messaging to individuals 鈥� has been used to powerful effect in past elections and became the focus of a 2018 scandal when journalists exposed the work of political consultancy firm Cambridge Analytica in the US and other countries.
鈥淲hen people are microtargeted based on their personality in real political ads, it is more persuasive to them than being not targeted by personality,鈥� says at Ben-Gurion University of the Negev in Israel.
Advertisement
But despite the wealth of personal data available on social media platforms, microtargeted political ads don鈥檛 always perfectly match the user鈥檚 personality. And while spending more time on tailoring political messaging to specific individuals can make these campaigns more effective, it requires more resources.
Generative AI systems, which churn out images and text, could drastically reduce the time, effort and money required to create a successful political messaging campaign, says Simchon.
To explore how effective such technology could be, Simchon and his colleagues used a large language model (LLM) to analyse the content of 10 real political ads shown to UK Facebook users between December 2019 and December 2021. The researchers asked more than 1200 people to rate these ads on persuasiveness and complete a questionnaire assessing their own level of openness, one of the 鈥渂ig five鈥� personality traits.
The team then used ChatGPT, another LLM, to rephrase the ads into two alternative versions: one appealing to people with a high openness score and one appealing to people with a low openness score. Around 1600 people who viewed a selection of these ads rated them on persuasiveness and completed the openness personality test.
The adverts aimed at more open personalities were rated as being more persuasive by people with high openness scores, and the same was seen for those with less openness with respect to the adverts targeted at them.
Based on their results, the researchers estimate that 2490 people could be persuaded by an AI-generated ad for every 100,000 people exposed to political messages.
Simchon acknowledges that the size of this effect is small, but AI-assisted microtargeting could have a big impact if it is applied on a larger scale, he says. 鈥淚n the paper, we use the simulation for 100,000 participants, but potentially you can microtarget populations,鈥� he says.
This is concerning for those focused on ensuring election integrity. 鈥淢icrotargeting has appropriately been raised as a concern, particularly because of its potential effects on specific local election races,鈥� says , a former director of civic integrity at Twitter and board member at the OSET Institute, a non-profit organisation researching election technologies. 鈥淲hile many are focused on how AI might affect 鈥榖ig contests鈥�, like the presidency, microtargeting can be used to spread disinformation targeted at particular marginalised communities.鈥�
Perez points out that microtargeting without the use of AI was used to raise doubts about the safety of covid-19 vaccines . He worries that AI could amplify the scale of issues he and his colleagues face in the year to come.
PNAS Nexus
