Possinble counterpoints to good realtime audio and/or video being the bottleneck to AIs disrupting human society/relationships might be:

  • Early adopters of character.ai and replika seem to complain more about memory/context management/long term coherence being the bottleneck
  • I don’t think(?) decent real time video calls caused a big jump in long distance relationships (audio calls is another matter obvs, but being pre-internet ruins the test). It seems like having a wider pool of potential matches + better matching algorithms was more causal there. Personalisation would be the equivalent bottleneck for AI companions.
  • Young people seem to spend more time doing voice chat + joint activity than video calls. Might be a group vs. 1-1 thing tho
  • People’s tendency to form parasocial relationships with youtubers vs. podcasters vs. substack/twitter users might be another data point. (edited)