Old Problems, New Tricks

Elections at Risk

Caroline Sinders, a machine-learning/design researcher who has been examining the intersections of technology’s impact on society, artificial intelligence, abuse, and politics in online conversational spaces, says that our fear of what’s possible with White supremacists and AI is a fear that researchers have been sounding the alarm about for over a decade. “[Far Right actors] have always been here. I think the reason we’re starting to see it more is because Twitter (now X) has changed hands, and a lot of folks who had been removed from that platform have been invited back. We’re seeing the gutting of trust and safety teams.” Trust and safety teams on the social media platforms we use every day are critical because they’re responsible for developing and maintaining a safe and fair environment for end users.  They moderate content, manually review flagged cases, identify and address fraudulent activity, and ideally escalate issues of violence and discrimination so these events do not run rampant on a platform. Many tech companies, like Twitter/X, Meta, and Amazon have been firing their Trust & Safety teams in large numbers since 2023.^[11]

Dr. Nikki Stevens, a software engineer and postdoctoral researcher at MIT who researches how trans activists and communities use data in their activism and lobbying work, shared their thoughts on the re-imagining of an age-old White supremacist issue: “All AI is doing is masking the fundamental problems with our electoral system under the guise of computational sorcery. We also know as recently as the 2000 presidential election, we were having basic computational problems,” they said.^[12] This “computational sorcery” produces AI-enabled apparitions that are already being used to create a reality that doesn’t exist, one where BIPOC and LGBTQIA+ people are the villains and White supremacist cis-heteropatriarchal fascism is the solution.^[13]

In 2024, U.S. voters are becoming increasingly concerned about AI being used to manipulate presidential elections. According to a 2023 poll by the Associated Press, nearly 58 percent of adults think AI tools that can specifically target political audiences, mass produce political messages, and generate deepfakes will increase the spread of disinformation and misleading information during the 2024 presidential election.^[14] Across political affiliations, 61 percent of Democrats, 44 percent of Independents, and 61 percent of Republicans said that AI being used to spread false information would increase during the 2024 presidential election.

And it’s begun globally. In January, the New Hampshire Attorney General’s office was investigating a series of robocalls that used artificial intelligence to mimic President Joe Biden’s voice, attempting to push Democrats to vote in November and skip the primary.^[15] In December of last year, Moldova dismissed a deepfake of President Sandu expressing support for a Russian-friendly political party.^[16] Last March, YouTube suspended accounts in Venezuela using AI-generated news anchors to promote disinformation in favor of President Nicolás Maduro.^[17] In Bangladesh last year, users were using deepfakes of politicians and news anchors to generate AI avatars for as little as $24 a month.^[18]

What makes this problem especially alarming are the chasms of digital literacy that exist, and all the more so in historically marginalized communities due to systemic exclusion from technological advancements. Last year, Community Tech Network reported that 70 percent of Black Americans were “unprepared with digital skills, affecting their employability, and, as more work moves to remote formats, they estimate that Black and Hispanic workers, without intervention, could be locked out of 86% of jobs by 2045.”^[19] Daniel Greene’s research shows that the U.S. “digital divide” was constructed through racist tech policies since the Clinton administration.^[20] What Greene calls the “access doctrine” tried to “solve” the problem of poverty through technological tools and skills—notably, AI also claims to solve problems of access and representation. Writ globally, the chasms around digital literacy continue to worsen, especially when we consider the fact that, for example, 40 percent of Black Americans did not have access to high speed internet in 2020, and more than a third of the world’s population have never used the internet.^[21][22] Countries with the highest numbers of people most lacking internet connections are India, China, Pakistan, Nigeria, Ethiopia, Bangladesh, and Indonesia.^[23] No access to high-speed internet equates to little to no access to technological advancement and the skills-building needed to not be left behind.^[24]

Large Language Models (LLMs) are AI systems that process and generate text. They use machine learning to determine probability, like the percentage of probability that a person might end an email with Best Regards, use very specific terminologies that would make an AI-generated anchor believable to millions of voters in Bangladesh, or call a Democratic voter in New Hampshire. These language models predict the most likely next word in a sentence based on the previous words entered. Language models then make predictions from these pre-existing patterns. Without existing datasets, namely human beings’ writing and communication styles, language models could not exist – at least not as we know them today. Not everyone is convinced that the use of AI by leftists and progressives, particularly in its current unregulated form, is a good answer to AI-enabled voter disinformation.

In fact, the new media, policy, and cultural concerns around AI recalls the past decade’s anxiety around big data, e.g. our personal data collected for profit. Craig Johnson, who owns a slew of AI companies and is the founder of Unfiltered Media, a boutique digital communications firm that seeks to raise the bar on progressive digital strategy, believes that that world is possible. “We have [a] product [where] we do online polling. But instead of unobserved online polling, instead of people asking people what they believe, we just listened to what they believe… I can train a model to know how a leftist talks, someone on the right talks, and someone in the center talks or someone in the media talks, or a specific group of people talks. I can evaluate a million tweets at a time.” Ever since we lost control over our data, some LLMs and other forms of AI are made to seem like an obvious answer to the same injustice AI produces in everything from Instagram and TikTok to our text autocorrect and email predictions.

Caitlin Chin-Rothmann, a fellow at the Center for Strategic and International Studies, became interested in tech policies after the scandal surrounding Cambridge Analytica. She shared her concern about the lightning-speed adoption of AI to advance political targeting: “One of my biggest concerns is that technology companies are really able to collect and share such a vast amount of personal information. And this is very detailed and intimate information as well and includes our geolocation records. So, where we go on a daily basis, and from that, you can infer where somebody lives, who we hang out with, what we do. So, this is all incredibly invasive information [and] there’s very few limitations on how mobile app developers or how websites or social media platforms are able to share and sell this content.” Our data—like algorithms—flows across the international data and processing centers of international corporations. Thus, our data privacy, software access, and algorithmic regulation laws are also geographic and deeply uneven: from U.S. states (when Montana banned TikTok in 2023) to nation states’ policies (Iceland’s ranking as having the strictest data privacy protections), or a collective’s position (the EU’s European Digital Media Observatory (EDMO) of 28 countries producing synced fact checking and policy recommendations.^[25]These uneven and disparate policies and approaches (and at times subterfuge) create extensive gaps in protections that work to safeguard those most vulnerable to the manipulation of data, images, and information.

Taking Action

BIPOC, queer, trans, disabled, poor, refugees, and other marginalized people don’t have time to wait on technology to catch up to their calls for protection. What do our communities do in the meantime?

Global organizations, social movements, and elected officials working to protect our digital rights and digital privacy teach us just how connected these movements are to interventions in far-right absolutism. For example, in the U.K., the Online Safety Act 2023 requires online platforms to remove illegal content when they become aware of it.^[26] In 2018 the BBC published a report based on extensive research of nationalism in India, Kenya and Nigeria and how citizens engage with and spread mis- and disinformation on the internet.^[27] The World Economic Forum released its Global Risks Report for 2024 analyzing challenges countries will face in the coming decade. The report places mis- and disinformation at the top of the list in posing a risk—and ranked India first as the country facing the highest risk in the world.^[28]

Experts interviewed for this article recommend tools like Nightshade, which poisons large datasets so our images cannot be scraped by AI, and the messaging platform Signal, which can be used for the most encrypted communication. It’s important to take the time to refuse access to your data on each site you land on, and use virtual private networks (VPNs are built into some browsers like Opera) to prevent sharing private information and data on the internet. Beyond specific tools, healthy skepticism and clear boundaries are mandatory. There’s also the time-consuming and essential work of researching audio and video messages, calls from political figures, and other news to ensure that they have been accurately reported on by reputable sources. Given this, labor and social movement leaders likely need to designate entire new staff and create training to support their bases.

At the same time, historically marginalized communities have been relying on circles of community and support for centuries. It’s important to remember that AI is a tool that is not inherently bad or evil and AI has not created alt-Right or White supremacist ideologies, but AI is being used by bad actors to further that radicalization and normalization of violence and bigotry all over the internet. It is always important to demand tangible, continually up-to-date legislation that protects users and voters from AI-enabled disinformation, laws and policies that can only be passed by a government that values our privacy and safety.

Because we’re still in the infancy stages of understanding how these technologies interact and should be permitted to interact with our society, it is too early to tell exactly how far-right AI abuse will impact the result of the global elections. Now is the time to brace and prepare, because it will be substantial.