Hacker News Reader

Unfortunately, I am unable to trust your words at this moment in time. I'm sure many others will agree.

While I desperately want to trust you on a human to human interaction, and I genuinely do want to trust you, we’ve been misled with disingenuous language by companies at FB’s scale so many times for years, I’m finding it difficult to trust FB at all—it isn’t you, at all, it’s Facebook’s and similar companies intentional fogging of language which leads do this distrust.

And while I do want to trust you, I can’t trust that the company/industry itself wouldn’t figure out a way to exploit it at a later date :(

It really does make me sad that I can’t trust what you’re saying, i think a fundamental part of humanity is wanting that level of trust, and it’s been intentionally exploited.

I totally agree with you. I will never trust what Facebook says. And not only that. When I look at pesenti's clarifications, the only think I can think of is: "They will probably start safe and really not create any face templates for now (during people's attention is on this topic), but after a few weeks/months (when the topic is no longer discussed) they will start doing it anyway. That is the amount of trust I have for Facebook (and Facebook is not alone in that category). Nothing good can ever come from Facebook.

> I work at FB and this feature is managed by my team.

How do you feel working at Facebook? Do you feel pressure from your peers on the outside about the moral compass of the company?

Maybe you can't answer these things if you aren't using a throwaway account, but I'm very interested in hearing how Facebook employees perceive the company themselves, as well as how they're treated by company outsiders.

My impression is that a lot of engineers are opposed to how Facebook operates and that this reputation must make recruiting and retention difficult.

> How do you feel working at Facebook?

See https://news.ycombinator.com/item?id=19036507 or https://news.ycombinator.com/item?id=20768607

> Do you feel pressure from your peers on the outside about the moral compass of the company?

Pressure is too strong a word. I do get into discussions with friends and family on the topic.

> My impression is that a lot of engineers are opposed to how Facebook operates and that this reputation must make recruiting and retention difficult.

That's mostly not accurate. Remember that these engineers can freely express their opinion internally or question Zuck live every week.

> My impression is that a lot of engineers are opposed to how Facebook operates and that this reputation must make recruiting and retention difficult.

In machine learning and computer vision they are a very desired company to work at, at least on the research side of things (FAIR). They have very good people and world famous ones, lots of powerful hardware to work with etc. I'm sure they have no problem with recruiting brilliant people from all over the world.

Let me guess: it defaults to on?

No, it now defaults to off. New users and users who only had the “tags suggestions” setting have that setting off by default.

This doesn't sound like the use case that the parent comment was describing. Parent comment was about what is inferred if you are a user and no longer have FaceRec setting on, but used to have it on at one point, and opted out. That's why the speculation was that "facebook will never "delete its internal record of your face" as the title states."

Edit: not to be snarky, because I'm pretty sure that wasn't your intention, but it sounded like you were doing exactly this:

> This cleverly gives facebook some cover. Bob might ask facebook, "don't use my face to surveil my friends", but facebook can say "how could we possibly do that? We don't associate your face with your account, so we can't associate your face with your preference, even if we wanted to."

Everyone understands how opting out works and what it means (I mean, some of your bosses pretend they don't, but even they do). Of course you can't create a template when the setting that disables template creation is on. The question is what do you do with data that you inferred in the past, and what kind of privacy can be expected if you opt out of a feature after it's been used. That's particularly important given this industry's history of opting in for you.

The problem is that language is ambiguous and it’s important to be accurate. I do believe that intent wise, it answers the question the way you would prefer. Let me explain.

In the case you describe, Facebook no longer creates a Face Template, that is some kind of ML model of your face that could be used to recognize it or identify that you are the same person appearing across new pictures.

But I can’t say Facebook no longer has a record of your face. It still has some pictures, uploaded by your friends, where you appear. It won’t know that it's you OR and if your friends upload two pictures where you appear, it won't know that the same person is in both unless they tag you manually (which they can only do if you are user and authorize them - but that's a different setting).

Thanks for taking time to reply to comments here.

If I understand correctly, when facebook detects faces, it does something like...

    compare face to all Face Templates
    if match
        associate picture and face with matched user
        (maybe also update Face Template of matched user)
    else
        discard data structures created during compare
        (maybe mark face as "already compared")

If the data is truly discarded, great! I'm glad if my earlier speculation is not what happens.

question: how can I ask facebook to disable facerec for me, as an ex user who disabled his facebook account ?

https://news.ycombinator.com/item?id=20901808

You have deeply questionable morals

What happens to unknown faces in photographs, you just discard them? What if I have 1000s of photo of aunt Emma, but aunt Emma is not on Facebook. When she eventually joins, and turns Face recognition on, will you rescan all my old photos to identify her? Or will Face recognition only work for new photos?

We don't discard anything because we don't specifically identify or recognize these faces in the first place.

I don't think we go back to old photos, I suspect it would not be practical. But I am not positive on this one.

What do you mean by not specifically identifying or recognizing these faces?

When a face -- which is unknown to Facebook -- is found in a photo, is that face used for graph building? Is a anonymous representation of it kept?

No: https://news.ycombinator.com/item?id=20903110

Don't you have to try to identify a face first, before you can decide you don't want to identify it? I don't see how it can work if you don't try to identify every face.

You identify that there is a face yes, that's face detection, https://en.wikipedia.org/wiki/Face_detection, which is not the same as face recognition and doesn't need a model per person.

No I mean you have to try to recognize the face. Otherwise you can not decide whether the face is from a person who wants to be identified or not.

I don't see how it could work any other way.

There are gaps in your use of language here which is exactly what the parent is talking about. Facebook could turn around later and say "we never said anything about a Face Stencil".

Can you elaborate on the "gaps"? See my other answers. I am trying to be as precise as one can be.

But on each picture you automatically recognize when there is some face. And you need to be able to learn new faces. So doesn't it make necessary for somewhere in the system to be your face model in order for system to decide to ignore it? Not my field so I'm clueless but I would assume that without such hidden model it would be more likely to match the next most similar person? (I'm guessing social network is also taken into account to make such mistakes less likely?)

The point is your algorithm will still need to detect "a" person in order to confirm it's not a person who has opted in and should be scanned

Exactly. You still need to detect which person is in the photograph in order to determine that it is a person who has opted-out. This feature is an incredibly half-assed PR attempt.