Building Inclusive AI at Facebook | Hacker News Reader

Building Inclusive AI at Facebook

tech.fb.com

58 pointsinfodocket23 comments

I am at a point of disenchantment with Facebook, that I honestly think that even if you’re doing the best, most meaningful research of your life, you’re still doing it for Facebook. And that’s inherently invalidating whatever you’re doing.

I honestly don't agree with that point of view, in fact, I find it quite disappointing to see this kind of point of view on HackerNews out of all places.

Debate about Facebook being good or bad aside, a large portion (if not majority) of human civilization's technical progress were made due to less than idealistic motivations. In fact, I'd categorize a large portion of engineering and scientific achievement under the label of "less-than-altruistic intention, but positive externality".

On the topic of Facebook, I actually think it's fantastic that the biggest social network is getting so much public scrutiny left and right that everything they do is put under a microscope.

Let's be honest, social networks will not go away, if Facebook disappears tomorrow another one will take its place before the end of the year, and they will just be as "evil" when it comes to privacy invasion and data gathering (why else would you start a social network in the first place?). However they are likely to be less incompetent than Facebook at PR and likely to have a more charismatic and likable CEO than Zuck, and before you know they are doing the exact same shady things without the public freaking out all the time.

> if Facebook disappears tomorrow another one will take its place before the end of the year, and they will just be as "evil" when it comes to privacy invasion and data gathering

This certainly wasn't true of MySpace.

Also I just disagree with this sentiment in general.

Im in a lab with people who go on to work at Google Research, deepmind, MSR, etc. I asked them about fb and they do not remotely care what fb does. Most Ph.D. graduates want to collaborate with specific individuals who are at the top of their field, and will go where ever they are..

It's sort of sad but I agree. For outstanding engineers and researches there are now better companies available to work for on cutting edge research. Facebook still pays well, and I'd assume they bump up compensation again to mitigate the ongoing exodus.

The problem with that statement is you could replace Facebook with N other companies and still say nothing of value.

I think you mean immense value—your best work is still to make others rich.

Honestly, what is so evil about your work making someone else rich? What is your point here?

Well, what are your values?

I can see where you are heading with this, and we probably do not agree, but I know you have spent more time thinking about this than I have so I will defer on core values. Still, it is an important topic that I should probably spend more time on. But I really don't see how in the modern USA (Where most of FB/HN is) you could make it through life without spending much of it making others more money than you are getting paid yourself. Further, I really don't see how getting paid 250k and producing a value higher than that is such a bad thing in practice. And, another important thing I think to add to the conversation is that it is rather difficult to ascertain what "value" an individual SWE has produced. There are also a non-zero ammount of SWE's that produce less value than they are paid for. Are they immoral for showing up to the job?

This is one-bit thinking. Things are not either great or terrible. They can always be better or worse.

I don't think we should be focusing on building inclusive AI, at all.

See this slide deck for details: https://www.chrisstucchio.com/pubs/slides/crunchconf_2018/sl...

This slide deck highlights some important tradeoffs. Thanks for sharing it!

However, it's not an argument for ignoring the problem and hoping it all works out. Knowing the consequences of what you do is important and better than not knowing, especially in large-scale endeavors. (Too many mistakes happen due to obliviousness, as if not taking time to look means you can't be blamed.)

Also, supposedly "inherent" tradeoffs are only relative to input data. If you can find a better source of data, you can do a better job and don't have to make tough decisions. (The math in this slide deck is assuming you're making do with what you have.)

In particular, it doesn't apply to the original article. As far as we know, there's no reason why Facebook can't make a camera work well for everyone.

That's a quite interesting slide deck! For others interested, I've not watched it yet, but this appears to be the talk https://www.youtube.com/watch?v=Zn7oWIhFffs

When I worked at Facebook (now more than a year ago) I remember the AI team had set up a series of demo units for anyone to try out. Small devices running "AI demoes" which were meant to show progress on the technology in some quirky and fun ways. Every single demo they had on display just would not work.

One demo, for example, had a prompt to ask the machine something about a photograph. "What color is the table?" or "What animal is sitting on the floor?"

No matter how you asked the question or what types of questions you asked, the demo would comically never get it right. "What color is the cat on the floor?" would return "A mug!" in response.

Weird. In Firefox you can't select the text in the second and third paragraph with the mouse. In Chromium, it works.

just delete facebook - you’ll will see less toxicity and will be spied on much less

It's remarkable that they would put together this piece, state that they are "test[ing] new systems through the lens of inclusivity" and then immediately, proudly show a chart of their new system that specifically leaves no room for non-binary or transgender people. It's like they are going out of their way to say they will be inclusive with AI in regards to race but gender? No they will exclude people from this, anyone who identifies or presents as neither the binary male or female will be specifically excluded from being able to correctly this system.

Yes it's nice that they write "gender presentation" rather than just "gender", as they know that is what they are trying to measure. But then why not let it train for "nonbinary" as well?

Facebook can try as hard as they want, but their inherent bias against minorities will still show through.

You want the chart to show 50+ genders? I don’t think it’d fit. That’ll probably still discriminate against those who don’t feel included. Someone is always going to feel alienated.

"Facebook can try as hard as they want, but their inherent bias against minorities will still show through."

Can you further explain how to create a system that shows no bias (positive or negative) towards any minority group? Is it as simple as don't ever categorize?

Edit: I think this poster is a troll.

It's ironic that these people who consider or signal(if you're cynical) virtue through their jobs completely ignore the larger picture.

Facebook is a cesspool which wants to lock people in the web into a single monopoly. And worse it is to prey on the users attention by dealing cognitive equivalent of drug hits.

Where were these woke engineers when Zuck tried to push his drug into developing nations in ruse of "free internet", or inumerable other breach of user trust and privacy dor sake of more profit?

Its inexcusable that when you're working in a fundamentally immoral/ammoral organization, you try to signal virtue hard. Its like a foot soldier in a mob teaching ethics lessons when cleaning blood off his shoes. Better to be honest about where you get your pay than be "holier than thou".

How far have we gone from the light of visionaries like Alan Kay and others who paved foundation to the web and computing, who wanted technology to magnify the intellect ...

Facebook building an AI is terrifying.

Maybe so, but please don't post unsubstantive comments to HN.

Reply on news.ycombinator.com