TL;DR: I still really like this text, but am unhappy I didn’t update it/work on it in the last year. There’s now research on the exact topic, finding attention spans have plausibly increased in adults.
Out of the texts I wrote in 2023, this one is my favorite: I had a question, tried to figure out the truth, saw that there was no easy answer and then dug deeper into the literature, producing the (at the time) state-of-the-art investigation into attention spans—sometimes often answered questions haven’t been actually checked, yet everyone spouts nonsense about them anyway. I’m happy this one got nominated, I like been positively reinforced for writing high-effort factposts.
Looking through the texts linking to this one (found with the backlink checker) I think my biggest mistake was not explaining the subscript probability in the abstract—it’s very often misread as attention spans declining by 65%, not that I believe that attention spans have been declining with a probability of 65%.
I’m also not super happy about the middle section: I wrote it first while going through papers, and it could be cleaned up a bit to be more readable. I didn’t run any of the experiments proposed in the later sections, and now I don’t have to: Andrzejewski et al. 2024 (h/t Marginal Revolution via Zvi) performed a meta-analysis very similar to the one outlined in option 3 in this section[1] with the d2 test[2], finding gains on attention span in adults over the last 30 years, and some worsening in children.
I have some wild guesses what this means, but didn’t predict an increase in attention spans. Maybe some Flynn-like-but-not-measured-by-IQ-tests gains in cognitive performance are still happening due to improved nutrition or less pollution, but children are more affected by a attention-sapping environment; or there is some amount of immunization or habituation happens to people in a distracting environment, but they have to go through that as children first; or there is some selection effect in the papers for the meta-analysis (I haven’t read it in that great amount of detail).
In the future, I would personally like to read Andrzejewski et al. carefully & work them into the text, as well as try and find some more work done in the meantime. I likely won’t find the time, but if I did I’d want to replicate their meta-analysis with the TOVA or the continuous performance task.
It is a bit crazy that was ahead of academia by about one year in noticing the discrepancy of research on this topic and producing a then-state-of-the-art investigation on it—a topic many people were talking about at the time! (And continue to talk about.) Another nail in the coffin for Cowen’s 2nd law.
Especially after reading the paper, I’ve become slightly less worried about declining attention spans, but not enough to let down my defenses against the onslaught of internet grabbiness. At the time I was writing the text I was quite worried about AI systems posing immediate & under-rated danger through distracting/addicting/convincing people in difficult-to-notice and hard-to-avoid ways, that has taken a bit longer than I expected. I wasn’t worried about deepfakes or political polarization as much as more subliminal & unintended-by-the-developers hacking of human minds by AIs. This is now finally starting to happen, but more covertly than I guessed it’d look like in 2022.
Finally, I liked the comment section on this post, especially the comment by Viliam about movies increasing in length.
Since this is my own post, I can’t vote on it, but I’m happy where it’s at, and just the fact it got nominated means I’ll update & edit it—it’s rough in a couple of places, especially the middle.
Indeed, their method is so similar to the one I propose that I’m tempted to speculate they did the analysis after reading my post—probably not true, but a fun option.
TL;DR: I still really like this text, but am unhappy I didn’t update it/work on it in the last year. There’s now research on the exact topic, finding attention spans have plausibly increased in adults.
Out of the texts I wrote in 2023, this one is my favorite: I had a question, tried to figure out the truth, saw that there was no easy answer and then dug deeper into the literature, producing the (at the time) state-of-the-art investigation into attention spans—sometimes often answered questions haven’t been actually checked, yet everyone spouts nonsense about them anyway. I’m happy this one got nominated, I like been positively reinforced for writing high-effort factposts.
The post was linked in a bunch of places: First on HN, where it occupied the top spot for a while, and later on mastodon, Stack Exchange Academia Meta, metafilter, HN comments, here on LW and a few blogposts. Multiple places link it just for one of the pictures :-). (I’ve also linked this post whenever someone made any claim about attention spans. Fun.)
Looking through the texts linking to this one (found with the backlink checker) I think my biggest mistake was not explaining the subscript probability in the abstract—it’s very often misread as attention spans declining by 65%, not that I believe that attention spans have been declining with a probability of 65%.
I’m also not super happy about the middle section: I wrote it first while going through papers, and it could be cleaned up a bit to be more readable. I didn’t run any of the experiments proposed in the later sections, and now I don’t have to: Andrzejewski et al. 2024 (h/t Marginal Revolution via Zvi) performed a meta-analysis very similar to the one outlined in option 3 in this section[1] with the d2 test[2], finding gains on attention span in adults over the last 30 years, and some worsening in children.
I have some wild guesses what this means, but didn’t predict an increase in attention spans. Maybe some Flynn-like-but-not-measured-by-IQ-tests gains in cognitive performance are still happening due to improved nutrition or less pollution, but children are more affected by a attention-sapping environment; or there is some amount of immunization or habituation happens to people in a distracting environment, but they have to go through that as children first; or there is some selection effect in the papers for the meta-analysis (I haven’t read it in that great amount of detail).
In the future, I would personally like to read Andrzejewski et al. carefully & work them into the text, as well as try and find some more work done in the meantime. I likely won’t find the time, but if I did I’d want to replicate their meta-analysis with the TOVA or the continuous performance task.
It is a bit crazy that was ahead of academia by about one year in noticing the discrepancy of research on this topic and producing a then-state-of-the-art investigation on it—a topic many people were talking about at the time! (And continue to talk about.) Another nail in the coffin for Cowen’s 2nd law.
Especially after reading the paper, I’ve become slightly less worried about declining attention spans, but not enough to let down my defenses against the onslaught of internet grabbiness. At the time I was writing the text I was quite worried about AI systems posing immediate & under-rated danger through distracting/addicting/convincing people in difficult-to-notice and hard-to-avoid ways, that has taken a bit longer than I expected. I wasn’t worried about deepfakes or political polarization as much as more subliminal & unintended-by-the-developers hacking of human minds by AIs. This is now finally starting to happen, but more covertly than I guessed it’d look like in 2022.
Finally, I liked the comment section on this post, especially the comment by Viliam about movies increasing in length.
Since this is my own post, I can’t vote on it, but I’m happy where it’s at, and just the fact it got nominated means I’ll update & edit it—it’s rough in a couple of places, especially the middle.
Indeed, their method is so similar to the one I propose that I’m tempted to speculate they did the analysis after reading my post—probably not true, but a fun option.
I vaguely remember seeing that Wikipedia article, but I’m unhappy I didn’t include it in the text as one possible measure of attention span.