In general I agree (but I already did before reading the arguments) there’s probably a hype around AI-based disinformation and argument 2, i.e., that we actually don’t have a lot of evidence of large change in behavior or attitudes due to misinformation, is the strongest of the arguments. There is evidence from the truth effect that could be used to argue that a flood of disinformation may be bad anyway; but the path is longer (truth effect leads to slightly increased belief than then has to slightly influence behavior; probably something we can see with a population of 8 billions, but nothing with losses as large as war or environmental pollution/destruction). The other arguments are significantly weaker, and I’d note the following things:
It’s interesting that, in essence, the central claim relies on the idea that there’s widespread misinformation about the fact that misinformation does not impact people’s attitudes, behavior, etc., that much. [Edit: In general, I’d note that most studies on misinformation prevalence depend on a) representativeness of the data collected and b) on our ability to detect misinformation. As for the first, we don’t know whether most studies are representative (although it is easy to suspect they aren’t due to their recruitment methods; e.g., MTurk and Prolific) nor in terms of the experiences these already-probably-not-representative people have (they mostly focus on one platform, with X, formerly Twitter, dominating most research). In terms of the second, self-reported misinformation exposure and lists of URL “poor quality websites” are probably very noisy (the latter depending on how exhaustive the URL lists are and relying on the assumption that good quality news don’t ever share misinformation and that bad quality news don’t ever share actual information; and of course this does not allow the detection of misinformation that is not shared through an URL link but by user’s own words, images, videos, etc.), and alternatives such as human fact checkers and fact-check databases are also not perfect, as they reflect the biases (e.g., in the selection of what to fact-check in the first place) and lack of knowledge that the human curators naturally have. In sum: We should have a good amount of uncertainty around our estimates of misinformation prevalence.]
The dig at the left with “aligns better with the prevailing sensibility and worldview of the liberal commentariat” is unnecessarily inflammatory, particularly when no evidence of difference in perspective between left and right is advanced (and everyone remembers how Trump weaponized the term “fake news”; regardless, left-right asymmetries aren’t settled with anecdotes, but with actual data; or even better, instead of focusing on “who does it more” focusing on “stopping to do it”).
In general I agree (but I already did before reading the arguments) there’s probably a hype around AI-based disinformation and argument 2, i.e., that we actually don’t have a lot of evidence of large change in behavior or attitudes due to misinformation, is the strongest of the arguments. There is evidence from the truth effect that could be used to argue that a flood of disinformation may be bad anyway; but the path is longer (truth effect leads to slightly increased belief than then has to slightly influence behavior; probably something we can see with a population of 8 billions, but nothing with losses as large as war or environmental pollution/destruction). The other arguments are significantly weaker, and I’d note the following things:
It’s interesting that, in essence, the central claim relies on the idea that there’s widespread misinformation about the fact that misinformation does not impact people’s attitudes, behavior, etc., that much. [Edit: In general, I’d note that most studies on misinformation prevalence depend on a) representativeness of the data collected and b) on our ability to detect misinformation. As for the first, we don’t know whether most studies are representative (although it is easy to suspect they aren’t due to their recruitment methods; e.g., MTurk and Prolific) nor in terms of the experiences these already-probably-not-representative people have (they mostly focus on one platform, with X, formerly Twitter, dominating most research). In terms of the second, self-reported misinformation exposure and lists of URL “poor quality websites” are probably very noisy (the latter depending on how exhaustive the URL lists are and relying on the assumption that good quality news don’t ever share misinformation and that bad quality news don’t ever share actual information; and of course this does not allow the detection of misinformation that is not shared through an URL link but by user’s own words, images, videos, etc.), and alternatives such as human fact checkers and fact-check databases are also not perfect, as they reflect the biases (e.g., in the selection of what to fact-check in the first place) and lack of knowledge that the human curators naturally have. In sum: We should have a good amount of uncertainty around our estimates of misinformation prevalence.]
The dig at the left with “aligns better with the prevailing sensibility and worldview of the liberal commentariat” is unnecessarily inflammatory, particularly when no evidence of difference in perspective between left and right is advanced (and everyone remembers how Trump weaponized the term “fake news”; regardless, left-right asymmetries aren’t settled with anecdotes, but with actual data; or even better, instead of focusing on “who does it more” focusing on “stopping to do it”).
It sounds odd to me to assume in 4. that what people fear is only or even mostly about anti-establishment propaganda. In part because, without any evidence, this is just mind-reading, in part because, given modern levels of affective polarization, the most likely is that when one’s party is not in power (or has only recently left), we are more likely to believe that the president (we don’t like) is using their powers for the purposes of propaganda, and so are more worried about establishment propaganda as it then becomes salient (e.g., in the US context: https://edition.cnn.com/2021/01/24/politics/trump-worst-abuses-of-power/index.html | https://www.nytimes.com/2023/06/14/business/media/fox-news-biden-dictator-trump.html).