About this blog

Welcome to The LLMs and Democracy Blog. This blog explores how Large Language Models such as ChatGPT by OpenAI can influence our modern democracy and political processes. In an age where technology rapidly shapes the way we communicate, information becomes available faster than we can capture, let alone check it. This development is enhanced by the newest developments of LLMs.

To get an impression on the relevance of LLMs in our society, let’s look at the results of a small survey we conducted. The first question (see figure 1) asked about the familiarity with LLMs such as ChatGPT.

Figure 1: Familiarity with LLMs.

It is obvious that there is a huge hype about this topic, since all of the participants had heard of LLMs. Most participants had even tried it, about 31.1% reported that they used it regularly. However, as figure 2 shows, the technique of language models and the inner workings are unknown to an equally large group of participants.

Figure 2: Knowledge on LLMs.

This shows that LLMs have a huge impact on our society and simultaneously what a mystery LLMs are for most.

When asked what they associated with LLMs in the context of democracy, most participants used optimistic words, such as “helpful” or “opportunity”. However, the scepticism about this new technology became evident through words like “risk” and “dangerous” (see figure 3).

Figure 3: Word cloud of associations with LLMs and Democracy.

It is clear that using LLMs for democratic processes is perceived as ambivalent and there is a need to look into this in more detail.

To shed light on this highly popular topic, this blog first explains how LLMs work on the Background posts. Then, five scenarios on how language models can impact our democracy are explicated. The scenarios first lay down an initial situation and then present a best case and worst case on how these LLMs could be used. Keep in mind that all scenarios assume state-of-the-art models, which also means that the models don’t have a consciousness or will on their own and are therefore not able to be malicious in itself. The worst case scenario happens either because of malicious human use or unintentionally implemented biases.

This blog should provide some food for thought on what chances and risks LLMs bring for our democracy and what precautions need to be taken by individuals and citizens to prevent a negative influence of these powerful models.

Have fun exploring this blog!