Facebook's Deeptext AI can understand text posts @ 7K posts a second

[Independent voter]

For example a post like "Wanted to sell a bicycle for $200", Deeptext can understand that and tell the poster it might want to use Facebook's want ads.

But think of all the jobs that require reading text and acting on it, lawyers are one, moderators on MMO-C, insurance estimators. Troll prevention?






Facebook is using its latest AI project to get a lot smarter at understanding text. In fact, the social network says DeepText, its new "text understanding engine," is so good, it can interpret "several thousands posts a second" with "near-human accuracy."

Introduced Wednesday, DeepText offers an intriguing look into how Facebook is using artificial intelligence to make its platform better at parsing the billions of lines of text that pass through it each day.

SEE ALSO: Facebook's AI chief: 'Facebook today could not exist without AI'

This may sound a bit dystopian but the goal of DeepText is to more precisely understand the content and context of text on Facebook, in order to improve its overall user experience. And Facebook is already experimenting with this technology in a couple different ways.

On Messenger, for example, it's being used for something Facebook calls "intent extraction" — figuring out the difference between messages that may sound similar but have very different meanings. Writing "I need a ride," for instance, may trigger a prompt for you to call an Uber but writing "I found a ride," shouldn't.

DeepText could also be used to proactively steer users toward Facebook tools based on the content of their updates. As Facebook's researchers explain:

For example, someone could write a post that says, “I would like to sell my old bike for $200, anyone interested?” DeepText would be able to detect that the post is about selling something, extract the meaningful information such as the object being sold and its price, and prompt the seller to use existing tools that make these transactions easier through Facebook.

It can also help with the seemingly simple task of filtering comments on crowded public posts so you're less likely to see spam.

Some of these improvements may sound like minor changes, projects like DeepText are going to be increasingly important to the company as it pushes further into AI. That's because the better Facebook is at understanding what you're talking about, the better it will be at curating your News Feed, or keeping your conversations within Messenger and hiding spam — all of which stand to keep more people on Facebook and Facebook products for longer.

Though still in a relatively early stage, Facebook says it's continuing to make DeepText better and find new ways of using it within Facebook, including using it to understand the contents of photos and videos.

Have something to add to this story? Share it in the comments.

[RaoBurning]

I wonder if/how well it can/will be able to discern sarcasm. It frequently doesn't come across well in text so that should make for an interesting challenge to the AI's learning.

[Deleted]

yeah im sure this is gonna go just aswell as microsofts AI did

[Deleted]

But can it take my McOrder?

[Independent voter]

I wonder if/how well it can/will be able to discern sarcasm. It frequently doesn't come across well in text so that should make for an interesting challenge to the AI's learning.

Probably it can't. Adam is a poster on the forum, he was in a convo with someone else and the subject of culture came up. He said "America has a culture?"

To this program I'm sure it would take the question literally but if he knew Adam you'd know that he was being sarcastic. Maybe the art of sarcasm will spread across the world?

[RaoBurning]

Probably it can't. Adam is a poster on the forum, he was in a convo with someone else and the subject of culture came up. He said "America has a culture?"

To this program I'm sure it would take the question literally but if he knew Adam you'd know that he was being sarcastic. Maybe the art of sarcasm will spread across the world?

So it would have to have per-user memory and be able to detect opinions which differ from their norm and/or come off as insulting. It would be one hell of a feat if they could actually build a sarcasm detecting textbot. Next step: Portable sarcasm meter.

[ComputerNerd]

yeah im sure this is gonna go just aswell as microsofts AI did

The difference there is Microsoft was trying to create something that behaves like us, and in some ways it did replicate what is subjectively the worst of us.