Dan Luu
patreon
How do commenters coordinate their independent "obviously" wrong answers?
🕑 Added 2022-09-09 09:53:34 +0000 UTCI keep a list of questions that I'm wondering about in my head and, when I'm talking to someone who I think might have an answer, I'll ask them a relevant question. The per-ask hit rate on this is very low but, over time, most of my questions get answered.
One that's been on my mind for a while is, how do people know to write & upvote basically the same wrong comments on different sites? I asked this question on Twitter a while back ( https://twitter.com/danluu/status/1503095391557881857 ) and didn't find any of the answers even remotely plausible, which is typical of what happens when I ask one of these questions.
One of the reasons this was mysterious to me was that the wrong answers would typically be ones that I'd expect a person with no particular expertise in the area to realize are wrong if they spent less than a minute thinking about it. An example of a question that elicited such answers across multiple sites where the same obviously wrong answers were the most liked/upvoted was when Jeff Kauffman asked the question, why are there so few six door cars?
The most popular answer was that it was due to cost because doors are so completed, have lots of moving parts, have seals, etc. But without knowing anything in particular about cars, anyone who's shopped for a car or even just look at car ads on TV would know that new car prices are roughly $15k to $120k for normal-ish cars, with prices going up to the millions for supercars. What fraction of the cost of the car could doors be on a $15k car? Surely not even close to 100%, but even if it was, given the dynamic range in pricing of cars, it doesn't make sense that it wouldn't be possible to buy, as an option, 6 doors, due to the marginal cost of doors. Additionally, someone who spend about 15 seconds searching for this information (I just timed this) would see that most of the cheapest cars are 4 door cars and not 2 door cars. And there are about six similar paths one could take that would take about 15 seconds to see that a pair of doors isn't marginal cost prohibitive.
The next most popular answer is that it wouldn't be possible due to extra doors causing cars to have poor strength and failing crash tests. But thirty seconds of searching would reveal that the actual impact tests that are done should not be majorly impacted by adding extra doors to the rear of the vehicle because that's outside of the zone of impact of side impact tests and of course even further from the impacting front crash tests.
The other common answers are just as obviously wrong, where anyone who knows a bit about cars will immediately see that they're wrong and anyone who doesn't can trivially see that the answer is wrong with a single web search.
And this isn't particular to the six-door question. On almost any topic, you'll see similar comments on HN, lobsters, LW, etc.; the top answers are usually totally wrong in a way that's easily apparent.
If people independently generated the same correct answer, that wouldn't be a mystery because correct answers are highly constrained. And if people had similar "almost right" answers, that wouldn't be mysterious either for the same reason, but it was a real mystery to me how people independently came up with the same wrong answers that were so obviously wrong that no one who thought about it could think that the answer was even within the realm of plausibility.
After about a year of asking about this mystery and mostly getting answers that were just as obviously wrong as the answers to the six-door car question, I asked Pam Wolf this question and she gave me what seems like the correct answer: people don't evaluate answers for plausibility at all and just throw out any salient thing they can think of. When it comes to cars, cost is the most salient thing, and safety / crash tests are the next most salient, which is why those come up, and this goes for the other top answers as well. Her explanation was that my mistake was thinking that people even attempt to evaluate their answers for correctness. Once I relax that constraint, the answers people give make sense.
When I mentally checked her answer against the list of other examples in my head of "six door car"-like questions, I found that her explanation worked for basically every case I could see of popular bad comments that weren't due to incorrect memes getting passed around.
Misha Yagudin was the one person I asked who also had an answer that had some explanatory power, which was that "[c]ould it be that peoples truth seeking module is actually convincing others module". This does seem like part of it, in that once people come up with some obviously wrong explanation, rather than spend 5 to 30 seconds checking whether or not their answer is correct, which would quickly show that it's incorrect, they often spend significantly more time explaining to other people why their answer is correct.
This phenomenon of people throwing out answers without spending basically any time at all thinking about whether or not their answer could be correct relates to something else people often object to: when I say that there's some obvious that people get wrong, internet commenters will object that, if it was obvious, people wouldn't get it wrong. I think this can be plainly seen to be untrue for many particular examples, e.g., the answers to the six-door car question, but I didn't have an response to the general complaint. However, a corollary Pam Wolf's answer to the above mystery also supplies an answer to the general complaint; something can be obvious to anyone who puts seconds of effort into answer the question, but most people won't do that, so they'll still get it wrong, but that doesn't make it non-obvious.
Comments
I don't know if this applies to the general case. But for FAQs, there's often this pattern: 1. Newbie joins a community 2. They read answers to a question, perhaps that they asked 3. As they hang around, they read the same answer to the same question over and over again 4. Excited that they can now contribute, newbie begins to answer question with the answer they've seen a lot 5. Newbie #2 shows up and sees many "experts" all agreeing on an answer, including Newbie #1. Surprise! None of them actually have any first-hand knowledge of the question and have been parroting something that entered the local lore a long time ago. I've definitely done this myself _several times_, but also have seen it repeatedly (forums, reddit, discord, any place where people come to ask for advice from a community of people interested in that subject). Sometimes the answer is correct; sometimes it's not. Most often, it's somewhere in-between, lacking nuance or complete accuracy that a true expert would be able to provide - or it's outdated and now incorrect. I do think Pam Wolf's hypothesis happens a lot as well, though.
This hypothesis could be tested (or at least "tested") with a text prediction network. You could say that these things answer with the most salient thing without evaluating the answer for correctness by design. It would be interesting to ask a few "novel" questions on internet forums and compare the consensus human answers to the net's answers. It would seem that the most-salient-thing hypothesis predicts that people will tend to agree with the net. (The rigor of this test is debatable and could be an example of me answering with a quite salient thing without evaluating correctness...)
