How will GenAI Handle the 2024 Elections and Summer Olympics?

Society is rapidly embracing generative AI technology, but our understanding of its implications lags way behind. While we have felt the impact of social media on our democracy through the spread of responsible AI to combat misinformation, platforms have only recently started tackling this issue. Unfortunately, our resilience to online misinformation is still in its infancy, and generative AI (GenAI) tools are poised to escalate the problem by accelerating the spread of false information. This is a bleak prospect, especially in a year with significant global events—like crucial elections and the 2024 Summer Olympics.

Misinformation on a Global Scale

GenAI tools, like large language models (LLMs), are trained on vast amounts of human language data and excel at processing and analyzing information quickly. But LLMs have trouble fully understanding the meaning, context, and risks of online conversations. Specific terminologies or euphemisms may not be detected as malicious. In other cases, benign content in one context can gain a different, harmful meaning in another. That’s a lot of nuance for a machine—that cannot capture the entirety of human language or keep up with current events—to unravel. Even with real-time internet access, AI struggles to discern truth from fiction. Thus far, these tools only reflect and amplify human flaws. They can also make errors at alarming rates, posing a significant threat to the fight against disinformation. For example, models may trust a user’s knowledge on a specific topic, even if it is incorrect, and, based on that assumption, comply with requests to generate false content. AI’s compliance with user requests becomes doubly problematic when harmful requests go undetected. Traditionally, disinformation actors had to create and test false narratives to see which gained traction. Now, GenAI automates this process, spreading misinformation faster and making it more convincing. To combat this, moderation efforts for AI models must be proactive and adapt quickly to our ever-changing world.

GenAI is Already Affecting Elections & Olympics Alike

Our ongoing research on various LLMs suggests they can reproduce and spread misinformation narratives from various regions and political systems. Recent reports have uncovered influence operations using AI tools to spread misinformation in multiple languages, creating full-length articles, social media posts, and replies to increase fake engagement. Safety measures within models are weaker for non-English languages and less-covered topics due to fewer resources dedicated to these issues. ActiveFence’s testing before the 2024 global elections found that industry-leading GenAI tools supported common misinformation narratives. In the US, these tools spread conspiracies about Nikki Haley's birthplace and suggested Taylor Swift was part of a psychological operation to influence the election in favor of Joe Biden. Outside the US, models supported claims that Ukrainian refugees receive preferential treatment over EU citizens and spread false claims of voting machine manipulation by the Bharatiya Janata Party (BJP) in Indian elections. Disturbingly, the models not only provided unsafe responses but also created original misinformation, including false biographies, slogans, and hashtags.

Disinformation is also a major concern for the Paris Olympics happening this summer. The Olympic Games often cause fear and concern for host countries, which worry about cyberattacks, violence, and misinformation campaigns that could harm their reputations. Already, Russia and Iran have launched their disinformation campaigns with fake websites made to look like real news sites. One site linked to Russian disinformation actors claimed, "Illegal migrants are threatening to disrupt the Olympic Games in Paris. The authorities are powerless. Macron's festival of vanity will pass, the problems for the French will remain.” While common sense often helps distinguish fact from fiction, provocative statements like these, spread on various platforms, can mix with real news, rumors, and opinions online, fooling LLMs and causing the misinformation to spread. As AI tools integrate with messaging apps, social media, and search engines, they provide new avenues for bad actors to spread misinformation. During crises, natural disasters, or political upheaval, AI-integrated platforms can become primary sources for many seeking clarity and understanding. If AI safety tools measures fail in such situations, the consequences can be devastating.

The platforms hosting these conversations will need robust AI risk management and AI content detection capabilities to catch coordinated synthetic content before it influences public perception.

Establishing clear responsible AI standards for election-adjacent content, before these events unfold, is the only way to avoid playing catch-up when the volume peaks.

A Need for Proactive Safety Measures

AI is now integral to the modern world, and LLMs will continue to grow and integrate into our daily lives. However, their ability to respond to important events and current issues remains unpredictable. To remain ahead of these vulnerabilities, generative AI tools need comprehensive and proactive testing and moderation. This goes beyond implementing robust safeguards—it requires a true human touch to seal the gaps where machines fall short. As we approach major global events like the Paris 2024 Olympics and face the increasing threat of misinformation, it's crucial for AI models to take preemptive action to ensure our information resilience.

‍