What 27,000 YouTube Comments Reveal About Progressive Messaging on AI

A quick analysis

Bernie Sanders has begun to message relentlessly on AI, releasing several videos attempting to raise awareness of rising capabilities and calling for a moratorium on new data center construction. After scrolling through some of the comments sections on these videos, I decided to run a quick analysis.

I scraped 27,662 comments from six videos discussing AI from the Senator’s YouTube channel, including his widely viewed conversation with Claude. I then classified the comments using the message batches API with Claude Sonnet, across a series of sentiment and thematic labels. A supplementary keyword based classifier was then applied to identify pop culture references. Of course, none of this is easily generalizable to the public at large, but it at least gives some idea of how Sanders’ audience is responding. Most comments are noise, but let’s see if we can find some interesting takeaways in the data.

“We got Bernie speaking to Claude before GTA 6” (1,900 likes)

1] Political content:

20.9% of the comments were purely partisan or political, mostly straightforward condemnations of the GOP or praising of Sanders, such as

“Bernie was mentioned in the epstein files... as being hated!!! If hes hated by those monsters then he is a great guy. I wish he won” (4,200 likes)

2] Safety concerns:

16.5% of the comments were classified as expressing existential safety concerns. Given the content of the videos, this is perhaps unsurprising, and AI-optimist comments (7.8%), also received low engagement rates (2.5 likes on average vs. 10.2 for the whole sample.

“Claude’s literally saying ‘You need to stop me. But you won’t.’” (4,300 likes)

"...And it gives me cold chills how AIs mimic us, being fed off vast loads our data, but they aren't really like us at all. It's a mask. (109 likes)

3] Changed views:

Vanishingly few of the comments had real epistemic shifts, with 90 updating in favor of being more concerned and 6 less concerned, YouTube comments are probably not the place to look for self-reflection, but these comments are 5x as long as the dataset average. None of the videos seemed to trigger more changed views than any other.

there might be a danger that a stupid AI granted a lot of permissions might start running a dystopian script. I didn't think this before watching this video, but it seems possible that something like that could actually happen." (6 likes)

4] Calls for Regulation:

Of 2,655 pro-regulation comments, 51% are vague without specifics, simple calls to action, and only 17.1% specifically called for a pause, ban or moratorium. While only 2% of commenters mentioned UBI, almost all who did (89.3%) also mentioned job loss. Unexpectedly, the concern that provides the strongest bridge to demands for regulation was privacy (2.98x). I’m skeptical that this generalizes, and would worry about privacy regulations providing no real gain on safety or egalitarianism.

“We need to understand yesterday that these AI companies will not regulate themselves. Until they are given guardrails they will do whatever they can in order to get a competitive advantage” (784 likes)

“The development of AI must become the central political question of all political debates, the one on which people vote.” (141 likes)

5] Terminator

By far the most common cultural reference was the Terminator, or Skynet, appearing in 296 comments, with The Matrix, 2001: A Space Odyssey, and Black Mirror all appearing 50-100 times. I continue to think that the simple Skynet analogy is underrated by AI safety advocates in its efficacy. 59% of pop-culture comments co-occurred with the existential risk classification, many people’s mental model of the technology is a movie where the robot literally kills you!

Shout out to the 7 commenters who mentioned Roko’s Basilisk.