【你的帖子被删除,也许是AI干的】

【你的帖子被删除,也许是AI干的】在拥有全新的AI介入色情、暴力等违禁图片的识别后,Facebook的员工终于可以松口气了。这家全世界最大的社交媒体公司长期以来只能依赖员工自己去识别每天上传的350亿张照片。借助于40petaflops算力的计算机设备,AI可以分析数以万亿计的数据,并在图片公布之前完成扫描检查,据称目前该AI举报的图片数已妥妥超过人类用户。

Facebook's AI that can spot hateful, pornographic or violent pictures now flags more offensive photos than human users

  • After ad success, Facebook invested in AI and machine vision technology
  • Uses 40 petaflops of computing power to analyzes trillions of data samples
  • Technology scans images before they are released for human users to see
  • AI also ranks stories in News Feed and reads aloud  picture content 

Using 40 petaflops of computing power, Facebook analyzes trillions of data samples using billions of parameters. The AI is also used to rank stories in the News Feed and instantly creates closed captions for video ads. And it reads photo contents aloud to users that are visually impaired 

Facebook has long relied on human eyes to monitor the 350 million photos that are uploaded daily.

Now, the social media giant is using artificial intelligence to shoulder some of the weight and says the technology has flagged more offensive photos than its human users.

Photos deemed offensive include content that is hateful, pornographic or violent and the AI scans every image to determine if they violate Facebook guidelines prior to releasing them on the site.

Facebook has used its ad success to invest in artificial intelligence and machine vision, which is currently being used to sniff out offensive content.

'One thing that is interesting is that today we have more offensive photos being reported by AI algorithms than by people. Joaquin Candela, Facebook's director of engineering for applied machine learning told TechCrunch.

'The higher we push that to 100 percent, the fewer offensive photos have actually been seen by a human.'

The moment after someone uploads a photo, Facebook's AI scans it for offensive content.

This method allows the technology to pull the image before any human users can get a glimpse of it.

This is done with 40 petaflops of computing power that allows the firm to analyze trillions of data samples using billions of parameters.

The AI is also used to rank stories in the News Feed and instantly creates closed captions for video ads.

And it reads photo contents aloud to users that are visually impaired.

Prior to using AI, the firm relied on outside startups like CrowdFlower or outsourcing in the Philippines to scour the site.

This process has proved successful, but it seems with the incredible amount of photos appearing on the site some may go unnoticed by human eyes.

Also, by the time a user sees the image and flags it for Facebook to review, the offensive content may have already done the damage it was designed to do.

 

Facebook isn't just targeting hateful images, the social media tycoon is researching the meaning of text shared on the platform.

Some 400,000 new posts appear every minute and 180 comments are left, so by developing technology that can hunt down hateful speech would be a great achievement for Facebook.

Developing an AI that reports offensive content is not just a milestone for Facebook, but it also falls in line with the firm's mission to fight against abuse and the agreement the firm signed this week.

 

Among the measures agreed with the EU's executive arm, the firms have said they will establish internal procedures and staff training to guarantee that a majority of illegal content is assessed and, where necessary, removed within 24 hours.

The US firms insisted that following the EU rules would not compromise freedom of speech.

They have also agreed to strengthen their partnerships with civil society organisations who often flag content that promotes incitement to violence and hateful conduct.

'The internet is a place for free speech, not hate speech,' said Vera Jourova, the EU commissioner responsible for justice, consumers and gender equality.

She added the code of conduct, which will be regularly reviewed in terms of its scope and its impact, will ensure that public incitement to violence or hatred has 'no place online.'

http://www.dailymail.co.uk/sciencetech/article-3620220/Facebook-s-AI-spot-hateful-pornographic-violent-pictures-flags-offensive-photos-human-users.html


Comments are closed.



无觅相关文章插件