There’s little doubt we are living in Babel-land. Discourse seems to produce a set of irreconcilable camps more or less shouting at each other, hearing what they want to hear, supporting their teams no matter what, desperately clinging to some hard belief in an attempt to escape madness as history is reaching a breaking point. A look at some of the recent outrage cycles should be enough to make the point: Darryl Cooper questioning aspects of the WWII narrative, Douglas Murray debating Dave Smith over Israel, James Lindsay calling everybody and their grandma “woke right,” someone using the n-word and getting cancelled… All of these dramas have led to rivers of digital ink being spilled on analysis, with little persuasion across the tribes to show for. Clearly, good ol’ “facts and logic” don’t seem to convince anybody to change their tack about anything.
A recent study brings home the point even more dramatically. Researchers in Zürich let loose an AI to argue people out of their opinions on Reddit, comparing how well it does to human posters doing likewise — with depressing results: the most important of which is how rare it really is that someone’s mind is changed, AI or not. Humans were typically able to achieve “conversion” in only 3% of cases. The AI did a better job with a success rate of 9%-18% — still low, but magnitudes higher. What makes this even more depressing is that the AI did not achieve these success rates by brilliantly gathering facts-and-logic and providing sources, but essentially by emotional manipulation: the AI tailored its message to the recipient, pretending to belong to the same group (“as a conservative…”) and adapting its narrative framework accordingly. It also boldly stated unproven facts, hitting its human counterpart with dubious but authoritative and emotionally charged statements instead of well-reasoned arguments. Not a good look. It also drives home that short of emotional sophistry, there is little that can bridge the gap between opposed opinion-groups, and even sophistry rarely succeeds, and is probably not sustainable.
No comments:
Post a Comment