Ask HN: good idea to create api for detecting porn? | Hacker News Reader

Ask HN: good idea to create api for detecting porn?

6 pointswsieroci15 comments

Hi,

I am wondering what do you think of idea of creating service which would detect porn for web 2.0 services using humans brains (they would be employees of this service). You would point to image url using API and service after some time (let say 5 minutes) would answer requesting (GET/POST) website url given by you with answer (YES/NO). Service would charge 1 cent by one image uploaded.

What do you think of this idea?

Regards, Wiktor

If you're going to offer a service like this, you want a few tiers of verification. You should have a lightning fast level 1 service that returns an automatic response. Perhaps level 2 would be a more computationally expensive automatic response. Level 3 might be human analysis, but the image would be just one of several shown on a screen at once. Level 4 would actually get a human's full attention for a few seconds. And Level 5 would be to handle strange cases, do CP reports and handle disputes.

Yes, such process could be perhaps most optimized, but the most problematic step is step 1.

Sure, it would be hard, there's some projects that might be able to help out however, e.g. https://github.com/pa7/nude.js. It's not an easy problem to solve but it's possible. You could also start the service with 2-5 of 1123581321's example or just 2-4 and wait to see what your customers request.

Existing flagging systems are adequate and 1 cent per use doesn't seem like a sustainable business model for anyone involved. It seems like mostly social networks would be your customer, but 5 minutes is a long wait and eliminates many use cases. A quick script hooked up to Mechanical Turk (or TaskArmy) does the same thing and could provide results in a minute or less.

You'd also be associating yourself with porn, which comes with all sorts of weird consequences.

What makes you think that this is a problem that needs solving? And if it needs solving, what makes it worth it to you? Are you uniquely qualified to solve this problem? What is your end goal?

Yes I think this problem needs solving because I do not see any good reliable API service like this on the web right now.

I'd suggest the reverse.

Because there is no existing API this cannot be in-demand. Or it would exist.

Sure companies like OKCupid, Facebook, POF, etc, all want to detect porn in their user-uploaded images, but they all handle it via user-reporting. That costs them nothing, and is "fast".

I suspect not many people would be prepared to wait five minutes for a real result.

What about combining a machine learning algorithm with people work for images with low confidence. You could even train the network with feedback from the people workers.

Five minutes is a long time for said web services to have to wait for an image to be validated.

Also, you might (i don't know, ianal) have some legal issues when inevitably the child porn shows up, because then your employees are being paid to look at child porn. Which if they're not cops might be a problem.

Although, if i'm running a web service, better you than my own mods, I guess.

I do not say that you have to wait for an image to be validated. Image could be validated after you have uploaded it and for example someone has flagged it.

So long as you have clear definitions for what is or isn't porn, and you have some protections and easy reporting mechanisms for images of child sexual abuse (as smartwater says, that's going to be a tricky legal area.)

But how much are you going to pay the people looking at the images?

For how many images they have processed

The other comments seem to suggest that this is unworkable, but services like this already exist:

http://www.nytimes.com/2010/07/19/technology/19screen.html

In this article I do not see company which offer such API

The article makes clear that large technology companies currently outsource moderation to companies that use human employees to detect porn and other inappropriate content. This is basically similar to your idea, regardless of the fact that the article does not go into detail about how the outsourced employees get access to content that needs to be moderated.

When I wrote my comment, the other comments suggested that having humans review content would be unworkable. They did not suggest that creating or providing an API for moderation would be the hard part of the business.

Bad idea, humans are obsolete for this task.

A program can easily scan a picture and detect nudity with a high degree of accuracy.

Reply on news.ycombinator.com