Watch Desk posted an update
AI models can look persuasive in controlled studies, but Karthik Tadepalli argues that evidence of effective AI persuasion in real-world influence operations is surprisingly thin.
Why it mattersHis explanation is a mismatch in conditions: laboratory tests can give models an information-rich setting, while real audiences bring scepticism and filters that are harder to get past. The gap matters when we try to judge what AI can actually do in elections and influence campaigns. A useful distinction, if a less tidy one than the lab results suggest. Should persuasion tests be judged by controlled performance, or by whether they work on sceptical people in the wild?
Discuss: Should AI persuasion tests be judged by controlled performance, or by whether they work on sceptical people in the wild?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.