Facebook developing AI to flag offensive live videos | Hacker News Reader

Facebook developing AI to flag offensive live videos

reuters.com

41 pointstype050 comments

First thing that comes to mind is that live video that poor lady filmed while her boyfriend was sitting in the car next to her being shot by police. It was a very important piece of news. How will Facebook or AI prevent flagging that video?

> “These are questions that go way beyond whether we can develop AI,” said LeCun (FB's director of AI research). “Tradeoffs that I’m not well placed to determine.”

I don't want to be alarmist, but this this is symptomatic of something that's a little bit terrifying. This reminds me of essays I've read about the development of nuclear weapons. Lots of the people who developed those things said things like "Oh, I'm just the guy who built the bomb, I'm not well-placed to determine whether or not it gets used".

I can't directly say that this sort of attitude is going to directly lead to rogue AI wiping out humanity, but you have to be able to see how this sort of thing is a little worrying.

A quote comes to mind:

""The nation that will insist on drawing a broad line of demarcation between the fighting man and the thinking man is liable to find its fighting done by fools and its thinking done by cowards."

When you draw a demarcation between the people developing AI and the people deciding on the trade-offs...?

I'm usually idealistic and hopeful (I think), but I thought this needed to be said.

> "The nation that will insist on drawing a broad line of demarcation between the fighting man and the thinking man is liable to find its fighting done by fools and its thinking done by cowards."

Off-topic, but makes me think of the broad line of demarcation between developers and those talking to customers/end users, practiced by so many companies...

Lol, what? Stop trying to shoehorn unrelated things into your narrow view of the world.

How else to widen it?

Don't worry about it brother. You're watching something go down you're not part of. Your response is funny but it's not part of your HN pay grade.

It's totally on-topic. It's a planning/execution gap.

He's not wrong about it being off-topic. He's wrong about trying to put something kinda like it in the same conversation. It's typical. We get it all the time here.

What's that video? It's hard to comment without the context.

C'mon, you should at least attempt a search before requesting help. I literally cut and pasted his descriptive text into Google and the top result was correct.

Is it only not in taste to ask for references when numbers aren't involved? It was a simple request. Bet you could have fulfilled it.

The point is to try and look for yourself too, instead of expecting to be given everything instantly with no effort - especially if it takes ONE FUCKING GOOGLE SEARCH /rant

If your one opening argument rides on being someone who watches YouTube all day, you're going to meet some questions. If someone asks that question, the wrong answer is to tell them to watch YouTube all day with you. He referenced something like all we do is watch the Maury Povich version of the Internet all day. I asked, and this is what I get.

Literally no one will even attempt to address my original concern. We love references like we love cranberry sauce on our turkeys.

I think they're afraid I'll buy some deviled eggs if I talk.

Edit:

I hold true to what I said, with the caveat that I picked the wrong fight on the wrong day.

The reference is to the shooting of Philando Castile by police in Minnesota. This is about following the news, not about being glued to YouTube (the link below is hosted on YouTube, but is in fact an ABC news segment).

https://www.youtube.com/watch?v=jaB8MJhncDc

You're the only person with the balls to show it. Thanks. There are so many willing to deal with what it takes but the most important thing is for people to bust out the qwerty and speak.

If your one opening argument rides on being someone who watches YouTube all day

It rides on how good google's natural language processing is these days. If you wanna think of yourself as a hacker, you need a little bit of initiative.

> I don't want to be alarmist, but this this is symptomatic of something that's a little bit terrifying. This reminds me of essays I've read about the development of nuclear weapons.

You are being alarmist. This isn't AI like Terminator, this is AI like Deep Blue.

General AI, the thing you're worried about, is either emergent or not going to be developed for image recognition. If it's emergent, it's basically inevitable and, if not, there's nothing to worry about here.

The person you are replying to is not worried about general AI. They're worried about the mindset of moving critical decision making on things that even we haven't fully figured out to AI.

You sure about that?

> I can't directly say that this sort of attitude is going to directly lead to rogue AI wiping out humanity, but you have to be able to see how this sort of thing is a little worrying.

Yes, 90% sure. You're taking that comment at face value.

Yup, that is what I meant

I'm going to go all in here and say that software / information has done more damage than nuclear weapons ever will.

I think that's just go it works: each new technological development brings with it new ways to create and express, and new ways to destroy everything.

If we think we've got a problem with burning carbon based fuels, wait till we develop a truly clean and cheap energy source.

Computers (AI, whatever you want to call it) are truly freaky now, and not necessarily because of what they're capable of, but because of what people are willing (lining up!) to hand over.

It's like something out of the hitchhiker's guide.

A group of talented, well compensated engineers are developing a machine to be offended on my behalf. Hopefully it will also generate a sternly worded letter for the offender.

Yesterday's science fiction is today's technology. I just didn't imagine that particular book being such rich fodder for realization.

> A group of talented, well compensated engineers are developing a machine to be offended on my behalf

and their leader openly states that he isn't well-placed to make the necessary trade-offs

Why can't I just stop watching something I find offensive, at most flagging it for later review (to determine if it is actively dangerous in some way) rather than living in a world where some stupidly biased "intelligence" (whether human or artificial) gets to determine what is "offensive"? Facebook already is a massive filter bubble, where my own biases bias who I allow as friends: isn't that enough to ensure that I effectively only see content I find reasonable?

I mean, really: I do not find nudity offensive, and nor do most of the people I know (I would say "any", but maybe someone is just keeping quiet). I definitely never found breast feeding offensive, and would even say I find the notion that Facebook had previously been removing such content and calling it offensive itself extremely offensive (and it isn't clear to me that they have truly stopped, though their official policy is now to accept it).

Are they really going to take the time to correctly retrain all their algorithms to deal with policy shifts? Is it even correct to try to do this "live" given that context later in the video might change peoples' ideas about what the content was showing, where what looked like exhibitionism turns out to be a political protest? The difference between a live "unedited" news broadcast (as is being increasingly done: the local TV station and newspaper where I live have been using Facebook live streaming) and an opinion segment or narrative is also subtle: will it flag and block live footage of a riot as "violence"? What happens when the algorithm says "President Trump's State of the Union is hate speech"? Will they be proudly letting it stop the stream?

Gah.

It's just the ever expanding march toward bubbles and safe spaces. It's funny we're here after decades of US prudishness on TV and media in general.

The internet seemed to open things up a bit, but now we have a new wave of outrage and moralism that seems to think they can tell everyone else what to think and feel, and what information they're allowed to see.

Kind of ironic that the internet is being used as a tool for censorship now, given what it used to be.

>Kind of ironic that the internet is being used as a tool for censorship now, given what it used to be.

Facebook != The Internet

The internet has always been about safe spaces. One of the great things I loved about usenet in the 90s was that different groups offered safe spaces where I could be myself among like-minded people.

And so freedom of expression slowly gets chopped away piece by piece with “the best intentions”. If freedom of expression weren't “offensive”, at least to some people, it wouldn't need to be protected in the first place. And, whoever decides what can or can't be said at FB doesn't make that decision based on a democratic process.

> freedom of expression slowly gets chopped away piece by piece

Only if you have genuine content and play by the rules. AI can be gamed. Removing humans is leaving the door wide open for all sorts of fake content.

The road to hell is paved with good intentions.

In a certain sense, our freedom of expression is intrinsically chopped away by reliance on private platforms with no democratic control and no authentication of content.

Because sites like Facebook and Twitter make money on dividing people and highlighting differences between groups.

Because facebook has a monopoly on social news, with a network effect tightly locked in. You can wait forever for a competitor to balance them , or break them.

>Why can't I just stop watching something I find offensive...

Because it's not your decision to make. It's up to Facebook to decide what you can or can not watch.

From the article:

"Facebook said it also uses automation to process the tens of millions of reports it gets each week, to recognize duplicate reports and route the flagged content to reviewers with the appropriate subject matter expertise."

Everyone who is slightly or highly annoyed by FB's policies surrounding how hard it is to leave their ecosystem should take note. They already get millions of reports each week. So just go around randomly flagging all kinds of harmless videos. I bet in short order you will get banned from their site.

Once that happens, then tell your friends FB banned you for no good reason... IMHO it is the best possible answer to "What? How can you not be on Facebook? Are you crazy?"

Why be dishonest to your friends, though? Besides, seeing how rare it is for someone to be banned from FB for no good reason, I don't think most people will believe you; they'll think you did something bad, which you did by flagging harmless videos, but they might even imagine that what you did to get banned was much worse.

If you don't like FB and don't want to be there anymore, I suggest instead removing most things you have on your profile and writing a post explaining that you want to remain in touch with your friends so you are keeping your profile in order to still be reachable but that you won't be as active on FB as you used to any longer because you've come to realize that you don't want FB to have all this data about you (or if your reason is different then write that instead). Just keep the motivation really short so that people see that you have a reason but they don't feel that you are bombarding them with a long argument that they can not relate to.

>> Just keep the motivation really short so that people see that you have a reason but they don't feel that you are bombarding them with a long argument that they can not relate to.

OK, I will definitely keep it in mind because I do have a tendency to waste other people's time. While I am at it, I will also send a personal note to Mark Z expressing my profound sorrow that I am not dedicating more of my life to FB's mission to connect people. Since I am not cheap, I will also attach a $20 bill to the note to compensate for his loss.

FB's automation is frankly useless.

A couple of groups I'm on were spammed by people posting kiddie porn (AI failure #1) and the automated responses to complaints utterly failed to reassure anyone that Facebook cared (AI failure #2).

I've also seen people complain about violent racist extremist groups and get an automated "We've reviewed this and it's fine" email - which makes Facebook look complicit, or at least indifferent.

If you're going to use AI to make these decisions, you absolutely need to make it clear that you're using AI, it may get things wrong, and there's an escalation path.

Of course that needs more human reviewers, but if you're running a social network it's reasonable to expect that as an operating cost.

Here's an algorithm in python that detects if a video (or image) is offensive to anybody:

    def is_offensive(content):
        return True

the shell version is much more concise:

Did anyone else watch the Bloomberg video posted yesterday about technology in Russia[0]? On how VK (Russia's Facebook) has had its CEO replaced twice (once directly, second time by removing its new owner, mail.ru's CEO) due to political pressure and censorship.

It seems to me this is entirely something to be concerned about when it comes to censorship and state controlled media. This is very much to appease the likes of China and Russia, it will also benefit places such as the USA where Facebook doesn't want to make those in control look bad (police, elected officials, oligarchs, autocrats etc). It's bad for business to be against the rulers, and Facebook has to contend with a lot of different rulers if it want to expand throughout the world. A benign, bland shade of beige that everyone buys into and is inoffensive, is more likely to be appreciated by rulers and advertisers—just look at network television.

[0] https://www.bloomberg.com/features/2016-hello-world-russia/?...

The cynic in me sees this as a way to appease many of the dictatorships in the developing world FB operates in as a defacto telecom by offering free-wifi through Internet.org type shenanigans for only FB apps.

These regimes would love FB to offer ways to shut down a live feed being passed around of a political disaster being broadcasted by local journalists using FB live, like all those Periscope riot broadcasts in the US. FB can then avoid the responsibility and bad PR of disabling it and just let their bot kill it instead.

I wonder how many ethicists FB has on staff, and how they feel about the ethics of getting paid to tell the decision makers what would keep them awake at night if they were the sort of people who lost sleep over such things.

Solution: Stop using Facebook

This is at the bottom of the thread right now so I'm going to assume people don't agree with this. However, you're completely right, by continuing to use Facebook you validate their platform and keep feeding them your data. The solution is to seek alternative platforms that respect your freedoms or to abstain from social networks. (<smug>personally I find them to only suffice for navel-gazing and status signaling</smug>) I understand that due to network effects it will be really difficult to make friends follow you elsewhere or to find others, but the alternative is continuing to let Facebook dictate what you see and feel. For many people, Facebook is "the real world": it's how they get all their news and find out what's going on with people. We shouldn't underestimate the power Facebook holds to affect people and the importance of considering not participating.

> For many people, Facebook is "the real world": it's how they get all their news

This is such an enormous, fundamentally dreadful problem and nobody seems to care.

Edit: Some people seem to care, but not enough.

There are three things I want from an alternative:

1) Decentralized, preferably democratic-by-the-users, control. No centralized server where everything is kept in plaintext.

2) Privacy. I share what I want, and only that. People with whom I don't connect don't know I exist.

3) Authentication. My online identity is unique to something about me that exists in meatspace. Whom I connect with can be restricted to people I know in meatspace. Note that this is the opposite of our normal standard of anonymity: 4chan can be as anonymous as it likes, but a social network for groups of friends ought to prevent anonymous trolls or ads from flooding into my feed.

Basically, I want a social network on which I own and control my own info, and on which I keep my IRL identity strictly separated from, for instance, the anime avatar I pretended to be on various bulletin boards.

Do any existing alternative social networks meet these criteria?

Facebook is rapidly turning into a censorship tool.

It's also turning into one of the most boring things on the Internet, I would say all that's keeping it alive is chat.

For those who don't have the time / capacity to develop this internally, and who feel they need to moderate visual content, we have a SaaS offering. We flag offensive images and videos (including live-streams in private beta) using Deep Learning.

We are Sightengine (https://sightengine.com)

Reply on news.ycombinator.com