As Twitchy reported, Facebook CEO Mark Zuckerberg told Congress Tuesday that he thinks the company is 5 to 10 years away from developing artificial intelligence that will allow Facebook to catch “hate speech” before it’s posted — an idea that had many asking who exactly would define what constitutes hate speech.
Video game designer Mark Kern, who has worked on games such as World of Warcraft, was among those struck by the idea of AI policing speech before it was even posted to the service.
So today Zuckerberg revealed his long term plan: To have AI watch what you type and stop you if you type "hate speech," or what we like to call "wrong think" because he can't even define what hate speech is.
— Mark Kern (@Grummz) April 11, 2018
A lot of people think AI's and Programs and Algorithms are inherently neutral, because they are machines. But this could not be farther from the truth.
— Mark Kern (@Grummz) April 11, 2018
AI's had a breakthrough just a few years ago that made a lot of things, previously hard, possible. Neural nets had always been around, but the ability to teach an AI with deep learning didn't become possible until recently.
— Mark Kern (@Grummz) April 11, 2018
Now it was possible, using a lot of computation power, to learn things from huge datasets of information. The problem was, of course, where to get this information?
— Mark Kern (@Grummz) April 11, 2018
Turns out that the huge dataset collected by social media companies and big data firms like Google are PERFECT for training AI. You see, AI needs TONS of information to train, unlike humans. They need thousands or millions of data points.
— Mark Kern (@Grummz) April 11, 2018
But even if you have this data, the AI doesn't know ANYTHING about what to do with it. You have to "train" the AI first. And just like training a pet's behavior, the actions of the trainer have a lot to do with the outcome.
— Mark Kern (@Grummz) April 11, 2018
When you train an AI, you are telling it what information is important, what to look for. You are also teaching it "desired" results from "undesired" results in many cases. Humans teach the AI. Humans with …well, bias.
— Mark Kern (@Grummz) April 11, 2018
As a result, the AI's "values" and "beliefs" are those shared by their trainers, who tell the AI what to look for, what is important and what is not, and what future prediction results are "right" and "wrong."
— Mark Kern (@Grummz) April 11, 2018
So when Zuckerberg says that Facebook is going to use AI to police what you say, before you say it, he is saying that Facebook employees will be training that AI what to look for and what is "right" and "wrong."
— Mark Kern (@Grummz) April 11, 2018
Make no mistake, there is nothing neutral about a Facebook AI. The AI will reflect the values, goals and biases of the company that made it. So what values does Facebook have?
— Mark Kern (@Grummz) April 11, 2018
Well, according to Zuckerberg, FB is in a part of the country with overwhelmingly liberal views. So we know that FB's AI will be inherently liberal. We also know that it won't respect privacy much, given FB's past history of breaches in this area. The AI will have these values.
— Mark Kern (@Grummz) April 11, 2018
We also know the AI wont' be trained by a balanced staff of diverse viewpoints. He admitted the tens of thousands of human censors he is adding to review content are not screened for political views. So it won't be balanced at all in political views.
— Mark Kern (@Grummz) April 11, 2018
So you can see, the Facebook AI Zuckerberg is dreaming of making real in 5-10 years to stop you from typing things, even before you post it, is not going to be anywhere NEAR neutral or fair, or balanced.
— Mark Kern (@Grummz) April 11, 2018
So remember, when Zuckerberg says he wants to police speech on his platform of 2 billion users with an AI, he is trying to make you think it will be fair because it is machine, a program. But it won't be, at all. It will be used to propagate one view and one view only: Facebook's
— Mark Kern (@Grummz) April 11, 2018
Now, given that Facebook controls the vast majority of public, shared speech on the Internet in most of the world, and is completely unaccountable and unelected, you can see this is a very dangerous goal. Not for FB, but for our freedom.
— Mark Kern (@Grummz) April 11, 2018
Nice thread on AI in general, And FB's intentions in particular. https://t.co/uGC9Yq3Oca
— Mithun (@_nullabletype) April 11, 2018
This thread is critical if you've been following the #Zuckerberg circus today, and are wondering what he means about using AI. https://t.co/v0h9MoWGUF
— Mr. Misanthropes United™ (@RW_Crank) April 11, 2018
Thread. Read it all, then think if you would want "the other guys" being the trainers. https://t.co/fBDG9g5fqq
— Matt (@lordjim_2001) April 11, 2018
And now a quick word from Live Action founder and president Lila Rose:
.@tedcruz doing a phenomenal job cross examining Zuckerberg, who admitted conservative pages have been shut down. He claimed he was unaware if any major leftist groups have ever been suppressed (…they haven’t).
Facebook has a bias problem. #Zuckerbergtestimony
— Lila Rose (@LilaGraceRose) April 10, 2018
Great questions from @BenSasse on how #Facebook polices speech, including pro-life content. #Zuckerberg claims pro-life posts wouldn't typically violate policies. We've seen otherwise – we're prohibited from promoting our Abortion Procedure educational videos.
— Lila Rose (@LilaGraceRose) April 10, 2018
Related:
MASTER CLASS: Ben Sasse EXPERTLY grills Mark Zuckerberg on 'hate speech' https://t.co/KbEklcMAhF
— Twitchy Team (@TwitchyTeam) April 10, 2018
