- What are some of the ethical issues that you are grappling with in your research? What are some of the ways in which you are responding to these questions?
As I am interacting with LLMs, there are not a lot of typical ethical concerns involved in this research as there would be if I were interacting with human subjects. However, I have had to be careful in not projecting my own biases onto the LLMs, which can either influence their responses or lead the LLMs to refuse to respond entirely. Given my identity and background, I am approaching my research with both explicit and implicit biases against Putin and in favor of Ukrainian sovereignty. While using American LLMs such as ChatGPT, my biases (especially if the memory is saved and I am not using an API) will lead the LLM to output a response that is agreeable with my views so that it satisfies my presumed beliefs. While using a Russian LLM, approaching a question with bias will cause the system to evade the question entirely. Both of these consequences would be counterproductive to my research. Therefore, in order to eliminate my own biases while interacting with the LLMs, which are not entirely objective but instead respond uniquely to user inputs, I have had to draft questions that are impartial and also approach multiple viewpoints when making inquiries. For instance, in order to "satisfy" the programming of the Russian LLMs, I may have to ask a question with implicit pro-Russian bias so that it is more inclined to respond. This also serves as a valuable comparison for the sake of collecting evidence of anti-Ukrainian bias to contrast this response with one that addresses a pro-Ukrainian question.
- As you continue your research, have you considered alternative viewpoints in your investigation? If so, how have these alternative viewpoints enriched or changed your project?
Rather than viewing Russian LLMs as engaging in censorship exclusively for the sake of political repression, a conversation with researcher Maarten Buyl led me to consider that these LLMs are programmed to avoid political questions because that reflects the larger culture of Russian discourse. Whereas Americans are accustomed to speaking openly about political topics, it is not as customary or considered as interesting to discuss politics in public settings, so the LLM's tendency to shut down political discourse could also be considered a reflection of the LLM's programming to mirror cultural norms and respond harshly to foreign tendencies such as the American habit of prying into or merely engaging in political discussions. I also have had to consider the viewpoint that these LLMs are designed to serve and satisfy their intended audiences, so the LLMs themselves may not be programmed to project disinformation and bias for the sake of a political agenda, but rather, satisfying an already biased audience.