You keep claiming all these statistical failures, yet you fail to produce evidence for any of them.
I'm sorry, but I trust actual statistics experts more than some guy on HN that makes bold assertions with no proof. To say "nuh uh" and then not back it up with anything is below the standard of debate I expect.
What are these "obvious" failures? Why have you failed to point them out concretely over several comments now? What statistical knowledge do you possess that makes them obvious to you but not to one of the most prestigious statistics organizations in the world? Please, do share with the class.
Your ideological attempts to poison the well by ignoring objective data because you don't like the messenger are not an argument.
To call me disengenous when you spent the last several comments failing to engage with the data and trying to hand-wave it away is ridiculous.
It is getting very hard to assume good intent at this point. This series of comments paints a picture of somebody starting from ideological priors and then trying to bend reality to arrive at their premade conclusion. I would be happy to be proven wrong here, but so far you have failed to do so.
> This series of comments paints a picture of somebody starting from ideological priors and then trying to bend reality to arrive at their premade conclusion.
The irony is palpable.
> I trust actual statistics experts
I listed the issues, but if you need an expert to explain them to you in detail, then perhaps you have neither the statistical nor media literacy necessary here.
1) Selection. Only 9 accounts, no interaction, short period of time. This is not a sufficiently large or realistic sample. That is also not how users interact with the product — for example, things like dwell time directly and materially impact the algorithmic weighting. They specifically avoided any measurable interactions.
2) Base rate. They did not know if the results reflected the base rate across the platform, or algorithmic bias. In other words, did X itself skew right, or did the algorithm.
3) Unreliable classification. Their own experts rated its accuracy poorly, and those are the people most ideologically aligned with the progressive political consultancy that trained the classification model in the first place.
Never mind the “propaganda” claim, which would require showing that not only was the algorithmic bias real (not demonstrated), but that the intent was propaganda instead of — for example — simply the result of maximizing for user engagement. None of which was shown.
> The irony is palpable.
> I already am eating from the trashcan all the time. The name of this trashcan is ideology. The material force of ideology - makes me not see what I'm effectively eating. It's not only our reality which enslaves us. The tragedy of our predicament - when we are within ideology, is that - when we think that we escape it into our dreams - at that point we are within ideology.
Some food for thought from ideology expert Slavoj Zizek.
> I listed the issues, but if you need an expert to explain them to you in detail, then perhaps you have neither the statistical nor media literacy necessary here.
You didn't list any of those before, no need to get personal. You could just argue on the facts instead.
Regarding 1 - You still haven't established why 9 accounts are insufficient. No interaction seems to be exactly the right move if the point is to establish the baseline behavior.
Your argument 2 is kind of self-defeating, don't you think? Either it's a right-wing propaganda machine because its algorithm emphasizes right wing propaganda or it's a right-wing propaganda machine because it's full of right-wing propaganda. I don't see how this argument helps your case that it isn't in either case.
Re 3 - now you're getting back to attacking the messenger. Attack the actual model. Which classification do you disagree with?
> Never mind the “propaganda” claim, which would require showing that not only was the algorithmic bias real (not demonstrated), but that the intent was propaganda instead of — for example — simply the result of maximizing for user engagement.
You haven't established why it would require any of that. As a counterexample, MoMA funded abstract art in the 50s as what they thought was supporting artistic freedom. Turns out it was CIA-funded propaganda. [0] It didn't require any intent from MoMA. It still was a propaganda campaign.
[0] https://news.artnet.com/art-world/artcurious-cia-art-excerpt...