The final speaker in this session at the ZeMKI 20th anniversary conference in Bremen is Giovanna Mascheroni, whose focus is on discussions with communicative AI systems about controversial and polarising political issues. This was explored by the use of serious games, with ChatGPT performing the role of a political journalist arguing first against and then for the radical Last Generation climate protest group. The switch from one position to the other was made once ChatGPT’s arguments for started to repeat themselves. Students then did the same, and also interacted with ChatGPT as they did so, and a jury judged who won the argument.
This provides an indication of the forms of public debate that might take place in conversations with communicative AI. It emerged from this that ChatGPT was able to set the boundaries of a civic conversation; it is designed to be civil as it interacts with its users, and to accommodate their points of view. It tends to prevent conflict – unlike humans –, and takes a non-aggressive tone.
But on doing so it might also downplay critical and urgent issues like climate change in pursuit of civility, and avoid taking a distinct position; in this, it is a machinic and non-autonomous interlocutor. It is clearly detached from lived and bodily experiences and does not share the common culture and shared knowledge of its users. Its use of historical examples (Gandhi, Martin Luther King) in contrast to current climate protests was ahistorical and clumsy, and its argumentation can tend to be very simple, using phrases from public debate and picking up on the commonsense arguments that one might find online.
As a serious game, though, this process clearly has great value: it stimulates discussion and prompts further thinking. The non-affective nature of ChatGPT might be seen as an antidote to affective polarisation, but at the same time the lack of shared knowledge and collective memory might also be a source of fragmentation and polarisation. It’s rhetorical strategies remain non-transparent forms of persuasion, and there is a ‘chat-chamber’ effect as the GenAI system aligns its responses with the inferred views of the current user.
However, these technologies are also evolving very rapidly, so their patterns of conversational engagement may well continue to evolve very considerably.
