Hacker News Reader

Comment by ChrisMarshallNY | Hacker News Reader

I noticed they are all white people. Does the algorithm have trouble with other skin tones?

That is actually a serious question, and it has nothing to do with the personal bias or motivations of the author.

I was once shown an algorithm for removing red-eye from photos, and noticed that every example they showed, was of drunk white Irish people (usually with light eyes). Turned, out, the library didn't do so well, for other skin tones.

[EDITED TO ADD] I actually enjoyed the heck out of it. Many of the comments, below, explain why it seems that way. Also, it may just be "the luck of the draw." I suspect that most images are from a common demographic, because that was who the photographers that supplied the images, came from. The tech, itself, seems to be quite basic, and is a fun idea.

View on HN

IIRC this wasn't made by an algorithm, the pointing locations were mapped by hand and the list of images is fixed.

So I think in this case it is because of personal bias of the author (regardless of how unintended it might be)

Since “Implicit bias” could not have been proven to actually exist your charge is a direct personal one based on nothing concrete at all.

I really love this project but that is a super on point observation. I should have noticed that. And I think the answer to your question is probably obviously yes. Like I know what I'm talking about...

Also very curious on the source for these images. They seem like scraped Facebook photos from a very specific time range. Maybe early 2000s?

Anyway whatever this is its art and I love it and it should be at the MoMA.

I agree. It is super fun. Kudos!

There are other skin tones but you're right it's mostly white people. However it's worth baring in mind that this is an old site. Wayback machine has it at 2012. I'm sure I've read an explanation that said each image was manually curated and then the angle of the pointing manually plotted. So I wouldn't say there's any algorithm bias beyond whatever images the author had access to at that point (no pun intended) in time.

Shrug, the first image that came up for me was people with dark skin.

I had several people with darker skin tones.

Me too.

Snap!

Me too.

My guess is that the location of fingers were manually identified so it’s more an issue of where the photos are sourced from.

I watched the video mentioned in another comment just to figure out exactly that question. From my understanding Voronoi diagrams are not used to find finger position, it is most likely preset, probably by human.

You're observation is generally correct. Here's an exception that proves the rule: https://pointerpointer.com/images/678.jpg

Strange, I did get photos with black people in the first tries.

I noticed in the photos I got, the person pointing was usually white, sometimes asian, and then plenty of black people were standing in some of those photos.

So? Ask yourself, would you have made the same comment if it was predominantly people of any other skin color?

What do we gain by forcing token inclusiveness into everything except making it feel artificial. What's wrong with the author just having mostly white friends to source pictures from?

This whole idea that you need people of your own skin color or gender or whatever because you can't identify with others is absurd.

Sigh …

Listen, sorry I asked. I originally asked, because I was under the impression that the photo selection was the result of an algorithm, and I actually shared an example, from my own life, that established a context for the question.

In my testing, I went through dozens of photos, and, in my testing, they all came up white, so it’s not surprising that I would have the idea that there was some … homogeneity … to the dataset.

I quickly found out, due to comments, that the selection was done manually, and that the dataset was not completely homogenous, rendering my question moot. In fact, I even added a comment, to that effect.

But by that time, it was too late. The torches and pitchforks were out. As it is standard practice, on the Internet, to jump to wild, cartoonish, hilariously wrong, conclusions about people, I set myself up for that.

Basically, I made the same mistake as you: an inaccurate assumption, based on insufficient data.

I won’t make this mistake again.

Have a great day.

FWIW I got 30%+ Asians. The rest was White.

I got asian and black people too. Not all of them white.

Reply on news.ycombinator.com