Arvind Narayanan

Floods might be the most under-recognized AI harm. By AI flood I mean overwhelming communication channels, forcing them to shut down or imposing a huge time cost on recipients. Three reasons we don’t talk about them much: 1) There are 50+ kinds of floods, so it registers as 50 different problems instead of one huge problem. https://www.cs.princeton.edu/~arvindn/ideas/ai-slop/ (And no one is affected by more than a few of these, so it’s hard to grasp how pervasive it is.) 2) The harms are usually diffuse. For example, people stop opening cold emails. There’s no news story there. 3) Similarly, the harms are harder to explain. AI agent swarm hacks are legible and vivid. But as I said in the quoted post, “there was something magical about the egalitarian potential of the internet, and we lose something when we go back to clubby hierarchies.” That’s a profound shift, but it only becomes clear after we sit with it for a bit. And again, that’s just one of 50+ floods. A semi-unified response to the “flood of floods” might be critical for protecting societal resilience.