Hacker News new | ask | show | jobs
by zxcmx 2198 days ago
It seems like the possibility that photos-of-undress are promoted by organic engagement is not discussed? (aka people seem to be clicking on this).

Confession: this is the first thing that sprang to mind for me. Not that algorithms suck or companies suck (though both of those are true) but that people suck.

13 comments

That was my first idea too! Not that people suck (not sure why you must suck if you enjoy looking at good-looking people), but that it's organic. Instagram just prioritizes what people click on, and people are attracted to attractive people, so they click on that.
Follow-up questions:

- Does Instagram prioritize scantily-clad photos of attractive people over scantily-clad photos of relatively less-attractive people?

- Is the prioritization based on a machine analysis of the photo, or on the response of earlier users to the photo?

- "While [the skew towards nudity] was consistent and apparent for most volunteers, a small minority were served posts that better reflected the diversity published by content creators." Are the majority people who have clicked on plenty of attractive scantily-clad photos? Are the minority people who have been presented with such and avoided clicking them?

> Is the prioritization based on a machine analysis of the photo, or on the response of earlier users to the photo?

I would bet the latter. Why go to the trouble of analysing difficult stuff about body structure, when you can just let users "decide".

I think we should formulate a law, something like "every internet imageboard, if left to user-moderation, will eventually turn into pr0n".

It's likely that FB uses user response to posts to decide how to prioritize them. However, several patents describe systems where they analyze (using CV) and make decision on the importance of pictures before they are published.
Remember kids, just because a Megacorp has a patent on something, doesn't imply that they are using an approach like this anywhere in production.
We didn't have enough data to test these hypotheses. If more people contribute their data, we'll be able to test that: https://algorithmwatch.org/en/instagram-algorithm/
> - Does Instagram prioritize scantily-clad photos of attractive people over scantily-clad photos of relatively less-attractive people?

If my theory of "It's not Instagram that prioritizes attractive people, people prioritize attractive people" is right, then yes, less attractive people are less attractive and therefore less prioritized.

I'm sure this is exactly the case. I bet far more people interact (regardless of whether they 'like' it) with the swimwear/sexy pose shots than they do the photo of last night's dinner.

I like a lot of photos, but in my own 'explore' timeline I've noticed categories of photos that appear from time to time, despite me never liking them. Until I did the whole "I do not want to see content like this" thing and made them disappear, I was seeing a lot of fishing videos, or bizarrely, people slipping their feet into shoes filled with foam (?). I'm fairly certain that as you scroll down the explore timeline just paying over a playing video - or particularly opening something (even if it's really odd or gross) just because you're like "what the hell is that" seems to register it as interest. I think if that topic is pretty "niche" you start seeing it a lot more pretty quickly.

I guess a bit like reading a single story on Quora about going to prison, or surviving an aircrash - you can pretty much expect your daily digest to be full of those for the next 3 weeks.

Point being, I assume few people will admit to it, but I bet their eye is drawn to the scantily clad photo every time. They open it/pause on it - and it's probably registered as 'most interesting'.

> I guess a bit like reading a single story on Quora about going to prison, or surviving an aircrash - you can pretty much expect your daily digest to be full of those for the next 3 weeks.

Man, this but on Youtube. There is a definite chilling effect, where I don't click on ANY ads or anything that looks targeted or served solely because I don't want to get spammed for for the next month. You click one video about Joe Rogan and then it's all you see.

When Jordan Peterson got big I broke down and watched a couple of his vids to see what he was about. And holy crap I got nothing but conservative videos spewed at me for weeks, PragerU ads for months. Even now I'm hesitant to watch any BLM, COVID, or anything that's trending because the algo will just spew more crap at me.

While it's probably based on engagement, it's still a flaw of the algorithms I feel. Just because other people engages more with the content, doesn't mean it's the correct content for a given hashtag / following.

For instance, I mostly follow cycling and running stuff on IG. The suggestions / explore feed is therefore full of women in short running shorts and bicycle jerseys with the zipper down. It's not really that relevant content considering what I actually follow, but I think the engagement those things get push all other cycling / running related content away.

Have you tried the GCN app for browsing cycling photos?
Didn't know they had their own app. I regularly watch them and the triathlon version on YouTube, so will check it out, thanks.
Of course it is. I'd even say that what we found is probably similar to the issue of offensive suggestions by search engines. A minority of Instagram users see the platform as a free source of soft porn images and their behavior is probably picked up by ML systems, amplified, and pictures of nudity are pushed for all users, in a vicious cycle. Just like search engines spread far-right conspiracies by suggesting them to millions of users after a few thousands searched for them.
but as opposed to "offensive suggestions", there is nothing wrong with showing-skin.
Unless your don't want to, but you are obliged to use IG because your business focuses on a demographic on which IG has a monopoly (the 15-25 demographics in the EU).
That's where models, brand ambassadors, and influencers come in. It's not like lifestyle business just found out that sex sells.
I'm sure there are worse jobs than using Instagram
There's a difference between using Instagram and having to wear a Bikini on Instagram even if you don't want to.
Nothing worse than owning a car repair shop and having to force your mechanics to wear a bikini for Instagram ads. Imagine having to have that conversation with a bunch of short tempered hairy dudes
Nobody is "obliged" to use it.

You can also buy ads if you want to show your posts to people, like any other business does.

To be fair, the linked article does suggest that this effect occurs for ads as well as organic posts.
Where's the suggestion that it applies to ads? The only coverage is 2 un-sourced instances of the ad creative review algorithm falsely flagging the content.
How confident are you that Instagram’s algorithm isn’t just optimizing for that particular user? And that it’s just that most users of Instagram seem to engage with scantily clad photos more than other photos?
The technical term for these is "thirst trap" ( https://www.urbandictionary.com/define.php?term=Thirst%20Tra...).
The article points to a Facebook patent for an "engagement metric" created at time of posting (that is, before user interaction) which specifically calls out "state of undress" as a factor to consider:

http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=H....

...and this metric is used to determine what gets shown in people's feeds.

The patent link was not working for me, here is an alternative: https://patents.google.com/patent/US8929615B2/
They can patent an architecture?
People suck because they go online to look at attractive people?
Algorithms suck because they take this as an engagement metric. Or rather people who design algorithms in this way.
If you’ve ever visited the profile of a woman with a very sexualised or glamorous LinkedIn profile, you’ll see the “people also viewed” results are almost entirely other women with glam photos. The normal case is that “people also viewed” is others at the same company or with similar industry/job type.
> "but that people suck."

It's in the definition of "attractive", things which attract attention and interest. Do people suck for wanting to eat food instead of mud? For wanting to listen to pleasant music instead of screeching noises? It seems like the least sucky creature in the world implied by your words would be Buridan's Ass[1].

Hyper-stimuli are one of the big problems of our time; hyper-palatable food, hyper-comfortable indoors sedentary pursuits, hyper-engaging gambling and gaming environments, hyper-engaging edited pictures and videos (and adverts and films and TV shows). The highlights reel, on steroids.

[1] https://en.wikipedia.org/wiki/Buridan's_ass

People suck? For being sexually attracted to members of the opposite sex? That mechanism is the very reason you are here. Your parents were attracted to each other and mated to produce you. Don't demonise sexual attraction.

What's actually happening here is some people are exploiting this natural quality for their own personal gain. Giving men access to countless pictures of naked women is like giving an alcoholic access to a brewery. It destroys people and it destroys relationships.

This brought to mind a quote from Roger Ebert's book "Great Movies":

> Of all the arts, movies are the most powerful aid to empathy, and good ones make us into better people.

Even if people suck, could Instagram tune their recommendation algorithm into recommending content that made them into better people?

And who gets to decide what is 'better'?
Here’s my guess: there is an ML model running at Instagram to prioritize content shown to users in the service of some target: engagement, revenue, shares, etc.

This model takes many of the features of the image into account: the account posting, the account’s previous likes, the number of existing likes and the time frame they were received in, the GPS location of the poster, etc... and possibly... deep learning features from the image itself.

If the model has the deep learning features... it will take all of 10 minutes of IG-scale data to overpromote hot naked people. It will also learn all of their userbase’s biases, and underpromote content from black people that isn’t “stereotypically black,” overweight women, short men, etc.

With the rise of blackbox deeplearning it's entirely possible that this is actively reinforced by instagram and not just an organic process.