Does artificial intelligence help Facebook solve the problems it faces?

Does artificial intelligence help Facebook solve the problems it faces?

It is known that Facebook faces many problems, the most important of which are the false news circulating on it, the publications calling for violence and terrorism as well as the problem of influencing the American voters in the recent elections. Therefore, the importance of using artificial intelligence techniques to find solutions to these problems has emerged. In a challenging area.

It was clear that Facebook was going to rely more on artificial intelligence techniques, where Facebook CEO Mark Zuckerberg mentioned the word "artificial intelligence" more than 30 times during the US Congressional hearings last month when he was explaining how Facebook deals with activities that God be upon him.

The implementation of the strategy for relying on artificial intelligence seems to be the responsibility of Mike Shkruiber, Facebook's chief technology officer, who spoke about this at the Facebook Developers Forum last week. He told journalists and developers at the conference: "Artificial intelligence Is the best way to make Facebook safer. "

After the two congressional hearings, some accused Mark Zuckerberg of making misleading statements to show listeners that all the challenges facing Facebook were technical challenges. Shkruiber told Wired that Facebook had already made mistakes, More than 2 billion site users Artificial intelligence is the only way to efficiently handle their data.

"Even if the company can afford to hire people who check every publication, you will not want to. If I tell you that someone reads each of your posts before publishing, it will change what you want to publish."

Facebook has already used automation to control its platform with some success. Since 2011, Facebook has used a tool called PhotoDNA, developed by Microsoft to detect child pornography, for example. "The company's algorithms have improved enough to identify other images that we would like to prevent from being deployed across our platform," Shroeber says.

But there are still tough problems. In recent months, Facebook has been investing a lot more in teams working on problems such as election integrity, bad advertising, and fake news. "It is fair to say that we have channeled a large share of the company's energy in recent months to all these issues," he says. Zuckerberg said earlier this week that he expected to spend three years building better systems to find spam.

Facebook's plan to create a security network based on artificial intelligence is facing major challenges, especially software problems. It is required to read the publications' words and determine whether it is in the spam content. This will not only determine the unwanted content through analysis Only the content of the images and will extend to full text analysis. This will be useful for programs that help combat counterfeit news, online harassment and propaganda campaigns such as those launched by Russia during the 2016 elections because they need to understand what people are saying, not just image analysis.

Although web search and translation have been successful, text-based software is still not very good at fully understanding the language context. The director of artificial intelligence and computer learning at Facebook Srinivas Narayanan explained the challenge during his speech at the developers' conference last week.

Facebook is making some progress with the algorithms it is reading. On Wednesday, the company said the system, which looks for signs that a person could harm himself, has received more than 1,000 calls to early adopters since its release late last year. Facebook algorithms also helped remove nearly 2 million pieces of terrorism-related content in the first quarter of this year.

"Facebook has improved its Bullying detection systems by training them on false data from programs that are taught to generate insults," says Shroeber. "This puts Facebook in an increasing number of companies that use synthetic data or counterfeit to train their own learning systems.

Another obstacle Facebook faces is the ability to understand other languages. The language technology in Facebook works better with English, not only because the company is American but these techniques are trained using texts taken from the Internet that prevails in English. According to Facebook figures, more than half of its employees do not speak English. "This is a big problem," says Shkruiber.

Currently, Facebook is working on a project called MUSE, which will facilitate the application of new developments for a single language in different languages ​​without the need for additional training efforts to train computer learning.