Well, but when it came to COVID, it probably wouldn’t have detected that there were two different schools of “consensus”. Because one side was being heavily censored and communicating mainly outside of the normal channels.
So it most likely would have surmised that the consensus position was the one not being censored, since that’s what it would have found everywhere, especially from the most “trusted” and “reliable” sources. Even though that consensus proved to be wrong in just about every imaginable way.
The same would hold true for other controversial positions, like climate change, transexuals, etc.
The consensus doesn’t have to be a totally binary thing. A well designed AI would detect and present opposing positions even if they were expressed less often on the internet. The results would be proportional to some degree rather than winner take all.
But yeah, with something like COVID where the establishment left totally flooded the zone, it’s inevitable that an AI using a statistical-linguistic model would give more weight to the establishment position. It’s just reflecting the available data. What more can it do?
I mean, AI isn’t going fight our fights for us. If we lose the public communications battle, then AI will reflect that in its results.