Identifying who inflames or defuses discussions is only part of the puzzle, and there are systems which eschew algorithmic content selection (including the Fedi

Identifying who inflames or defuses discussions is only part of the puzzle, and there are systems which eschew algorithmic content selection (including the Fediverse from which my example is taken) where that doesn't really hold. Admins might benefit from such tools, however. But the larger problem remains that people really can't and don't grasp the distinction between truly-global-scale-public and intimately private within online discussion contexts, and they've kept on not grasping this for decades now. You can go back to the 1960s with Joseph Weizenbaum and the ELIZA Effect as an example of how not only how much people will disclose personal and intimate details online, but with how engaged they get with such systems and the discussions. Another possibility for AI is, of course, to provide an interactive discussion mechanism itself, much explored in fiction and film. Any number of issues could arise there. One issue I'd identified years ago is what I call the Pennyworth Problem. Bruce Wayne has a private personal butler. AI and centralised computing services could give us all Pennyworths (and we do already have Alexas, Siris, Cortanas, and, hey! Google!). But where Pennyworth might hypothetically betray Wayne specifically, PennyworthAI could betray all of Gotham, all of the United States, and all of Earth, whether to the legitimate operators of the service, to a government demanding control over it, or to some third party identifying an opportunity and availing themselves of it. And do so with disclosure rates and fidelity hugely greater than that of any human. See: <https://news.ycombinator.com/item?id=15513270 https://news.ycombinator.com/item?id=15513270> Which would be another dimension of the general intimacy problem.