Niamh Cashell
Politics PhD candidate on the Digital Campaigning in Electoral Democracies project at the University of Manchester. Digital Good Network Research Intern in the BBC R&D Responsible Innovation team. Special thanks to the BBC for hosting me for a 3-month project evaluating AI-generated images around the General Election.
Email: niamh.cashell@postgrad.manchester.ac.uk
UK Election 2024
Section 6: The digital campaign
62. Local news and information on candidates was insufficient (Dr Martin Moore, Dr Gordon Neil Ramsay)
63. The Al election that wasn’t – yet (Prof Helen Margetts)
64. Al-generated images: how citizens depicted politicians and society (Niamh Cashell)
65. The threat to democracy that wasn’t? Four types of Al-generated synthetic media in the General Election (Dr Liam McLoughlin)
66. Shitposting meets Generative Artificial Intelligence and ‘deep fakes’ at the 2024 General Election (Dr Rosalynd Southern)
67. Shitposting the General Election: why this campaign felt like one long meme (SE Harman, Dr Matthew Wall)
68. Winning voters’ hearts and minds… through reels and memes?! How #GE24 unfolded on TikTok (Dr Aljosha Karim Schapals)
69. Debating the election in “Non-political” Third Spaces: the case of Gransnet (Prof Scott Wright et al)
70. Which social networks did political parties use most in 2024? (Dr Richard Fletcher)
71. Facebook’s role in the General Election: still relevant in a more fragmented information environment (Prof Andrea Carson, Dr Felix M. Simon)
72. Farage on TikTok: the perfect populist platform (Prof Karin Wahl-Jorgensen)
Fears that AI-generated images, video and audio would bring an information apocalypse of misinformation ahead of the General Election did not materialize. Instead, the use of AI-generated images varied widely across tools, platforms and communities.
Despite blocks on generating presidential candidates in the US, Midjourney (a widely used commercial image-generating tool) allowed users to create images of UK politicians leading up to the election, albeit with unrealistic depictions of the politicians requested. Citizens largely used these tools in playful ways, creating memes, commenting on current events and inserting politicians as characters in their favourite films. Liam McLoughlin’s chapter highlights how AI videos created by citizens formed part of a participatory culture, while official campaigns used AI tools to generate campaign content and even create AI candidates.
Although it was often difficult to tell which images are AI-generated in the wild, on Midjourney Rishi Sunak was undoubtedly the most generated politician of the campaign as citizens generated images and memes relating to current events. After his rain-drenched election announcement, images of Sunak speaking outside a flooded Downing Street were shared widely across social media platforms. After Sunak stated he went without Sky TV as a child, Generative AI images showed him staring at a blank television screen, crying and begging for Sky. Facebook events dedicated to Sunak’s leaving drinks were flooded with satirical images following the events of the campaign.
Behind Sunak, Starmer and Farage were the next most generated party leaders. Few images were generated of Liberal Democrat leader Ed Davey, perhaps because the real photographs of him bungee jumping and falling off a paddleboard were exciting enough. On X and Reddit, some users shared images of Farage as a heroic and patriotic figure, for instance riding a lion while wearing a Union Jack suit. Images were used to mock or deify political leaders, and most focused on a politician’s personal characteristics rather than policy issues.
One of the most popular uses of Midjourney was to generate illustrations of politicians. These included Midjourney’s default painting-style images, as well as more political content such as cartoons and caricatures, which can serve as tools of political engagement and expression. A similar proportion of images related more explicitly to meme culture, taking and exaggerating current events. A category of images inserted politicians into popular franchises such as films (Dune, the Matrix), television shows (Fallout), and video games. Few images created with Midjourney appeared to be political and realistic in a way that could spread misinformation, potentially due to blocks on this content, although some showed politicians having coffee with each other or meeting other world leaders.
Alongside this more lighthearted content, AI-generated images shared amongst the far-right on X generated harm through depictions of dystopian futures. Such images visualized conspiracy theories of Muslims ‘taking over’ London. Amongst these X users, images of politicians wearing Muslim dress were created and shared. For example, an image of Keir Starmer wearing a pink hijbab was picked up and shared by GB News presenter Darren Grimes.
These insidious images have the potential to cause harm through their blatant Islamophobia, and gendered and racialized constructions of Muslims as an outgroup in society. Muslim women were portrayed as victims, with full-body covering Islamic dress being a recurring theme amongst the far right. Muslim men were depicted en masse, as a large group and with faces hidden, a depiction which a social semiotic approach informs us creates fear and anxiety. WWII soldiers were a recurring theme amongst these images, invoking nostalgia for a time when white men were revered as heroes. Lions, Union Jacks, the St George’s Cross and figures such as Winston Churchill all construct the in-group as white, British and masculine.
Notably, many of these images were not realistic. Instead of fake news and realistic misinformation, they caused harm through their clear and divisive constructions of in-groups and out-groups. They perpetuated harmful stereotypes against minorities, sowing division and hatred between groups in society. This emphasizes the importance of not giving in to hype on issues like misinformation and taking a theory-led approach to AI-generated synthetic media and its dangers for society.
This election showed that AI-image generating tools are largely used in fun and playful ways. The post-election challenge lies in combatting the more harmful content in a holistic way. Technical initiatives such as labelling AI-generated images and including embedded image provenance are important and necessary steps to enable citizens to understand where an image has come from. However, we also need to adapt our approach to combat the harms of images which are more noticeably AI-generated and attack minorities.
Approaches could include regulations for platforms to remove harmful images, which may be more difficult to detect than text-based abuse, media organizations building media literacy, and academics developing theory around the unique impact of visuals on emotions and behaviour of citizens.
This article is based on early empirical findings of a content analysis of AI-generated images conducted in collaboration with the BBC R&D Responsible Innovation team, funded by the ESRC Digital Good Network.
