[ryandrake]

The constant request for more and more "context" is textbook Sealioning [1]. When faced with an overwhelming video evidence of misconduct, the only viable way to deny it is to endlessly seek out some kind of magical exculpatory "missing context". That's pretty the only way misconduct-deniers can sow doubt at this point. It's bad faith argument and disheartening to see here on HN.

1: https://en.wikipedia.org/wiki/Sealioning

[lki876]

Unless it's true. Look at any court case for example. If you only heard one side you'd think that side was obviously right.

[WealthVsSurvive]

Your argument is the essence of "sea-lioning" itself. There is indisputably a tremendously large dataset of unassailable evidence (video, cross-referenced personal accounts, audio), things we consider meeting our "beyond a reasonable doubt" in the court of law. And, yet, here you are, saying "maybe we don't have all the facts." At what point are you simply wrong? Never?

[throwvid19]

While it's true that there is tremendous data of many events of police overreach--corroborated by many sources--certainly not all events in this data set are corroborated. How many? We can't know without enriched data.

Given that this is a resource for events of police brutality, it should be no surprise that contributors are likely to be biased to report events favorable to the assertion of brutality. The general premise asked, "what wrong thing was done to you/your people?" which is likely to result in emotionally biased response. If it was a dataset of reports of interactions between two different ant colonies, for example, the general bias is likely to be significantly lower.

If the goal is to understand the relationship between a population and their police, then obtaining data from both sides would be ideal. Of course, that's not what this data set is, which is why some people are raising concern of bias. As a data set, it's use is limited to support one aggrieved side. This is not much different from training ML models on, say, only Caucasian faces: it may work if the intent is to recognize or generate Caucasian faces, but it is by no means general purpose. As such, it seems reasonable to question the fitness and intent of this data set.

No data set is perfect, and we'll never have "all the facts." But I don't think upholding inherently-prone-to-bias data as "good enough" is a reasonable response to questions about its bias. We cannot achieve perfection, but that doesn't justify denial of bias in the data.

On the matter of "sea-lioning", I've never heard the term before, and I'm not sure if I've ever been exposed to this type of trolling because to recognize it would require me to be able to read minds. However, I understand the forum guidelines charge us to assume the best interpretation of any comment, so I am disinclined to assume that people here asking for more data are trolling. The essence of claiming "sea-lioning" appears, at least at face value, to be an alternative to saying "I don't have to explain myself to you" while maintaining the illusion of taking the high road.

People have brought up the very reasonable concern that the data is extremely prone to bias, and those people are being silenced. It seems this is because it's not a popular idea to challenge the aggrieved party, not because the data is somehow unbiased and they're asking for something unreasonable. This seems unusual for an otherwise truth-seeking community.

Perhaps I am wrong, and I recognize everyone has their own bias and not everyone always acts in good faith. I have just come to expect more from this community than what I've seen in these comments. It seems good faith is not assumed in many cases among these many conversations.