The next session at the 5th Data Donation Symposium at the Weizenbaum-Institut in Berlin is on the role of chatbots in data donation studies, and begins with Daniël Jurg, whose focus is on news engagement via chat systems; this involves chatbot audits as well as as chatbot data donations.
The news environment is cacophonic today, and this has led users to employ various curation tactics that are also informed by algorithmic imaginaries; in chatbot environments, though, there is also automatic curation by the chatbot itself. Chatbot systems engage in conversational steering, and this is also a strategy available to users as they engage with the chatbot; this then can also manifest as a form of algorithmic resistance against the chatbot’s own activities.
This can be studied through data mirroring: data donations on chatbot interactions can be visualised by the researcher, but then also discussed with the data donor; this back and forth can lead to a better understanding of the patterns captured in the data donation, and correct misreadings by the researcher. In the process, the data donor also interprets their own data patterns.
Much of this has been done on social media platforms; how can it be applied to studies of generative AI engagement, then? First, data collection might take place via browser plugins that enable users to download and share the chat; this could work via network interception approaches that access the full JSON of the in-browser interaction with the chatbot (and may also allow the user to select certain aspects to in- or exclude).
From this, it may then be possible to extract what sources are being cited by the chatbot, for example; the more complex challenge, however, is how to deal with the masses of text that chatbot conversations may contain, and how to visualise this information for subsequent walkthroughs and discussions with the data donor. Good visualisations are critical to the quality of a posteriori interviews with the data donor.
