AI is a double-edged sword for democracy. On the one hand, traditional analytical tools can be used to analyze data and spot trends in certain issues that would be tough for human policymakers to see, and to help make better decisions about policy. The emergence of large language models (LLMs) has also been a game changer to empower ordinary citizens to participate more in the democratic process. For example, LLMs can be used to help us formulate and express our perspectives and policy position as well as making political arguments more cogent and informed (whether in social media, letters to the editor, or comments to rule-making agencies in response to policy proposals).
 
But obviously, the ‚creative‘ power of LLMs can also be dangerously used to spread misinformation and disinformation, which can undermine trust in democracy. In fact, we’ve already seen concerning deepfake videos of politicians saying or doing things they never actually did. In addition, we’ve also seen AI-powered recommendation algorithms being used to track user activities (e.g. on social media) and target them with political ads that could potentially skew their political views and sway their voting decisions. Indeed, the combination of biased recommendation algorithms and deepfake political content is incredibly dangerous to democracy. To solve this, we need a combination of technological solutions (e.g. anti-deepfake detectors), regulatory solutions (e.g. transparency obligations in algorithms and restrictions around political use of AI) and culture/educational changes (i.e. educating citizens to not take every content at face value and to do more due diligence into any political related content).