Facebook's AI could be reading your messages to teach itself how humans communicate

Facebook envisions a world where humans and bots casually communicate with each other.

And now the social media giant is one-step closer with a natural engine that analyzes 10,000 posts every second in 20 languages.

Called DeepText, this artificial intelligence uses deep neural networks to understand the meaning of text being shared on the site 'with near-human accuracy' and gives relevant suggestions.

Scroll down for videos 

Facebook is one-step closer to perfecting chat bots with a natural engine that analyzes 10,000 posts every second in 20 languages. Called DeepText, this artificial intelligence uses deep neural network architecture to understand the meaning of text being shared on the site 'with near-human accuracy' 

'Understanding the various ways text is used on Facebook can help us improve people's experiences with our products, whether we're surfacing more of the content that people want to see or filtering out undesirable content like spam,' Facebook shared in a recent blog post.

'With this goal in mind, we built DeepText, a deep learning-based text understanding engine that can understand with near-human accuracy the textual content of several thousands posts per second, spanning more than 20 languages.'

Facebook has already started testing the AI in Messenger, where it sits behind the scenes distinguishing text in conversations.

For example, DeepText knows that when a user says 'I need a ride' they are talking about requesting a taxi.

And when they say 'I like to ride donkeys', the AI understands this is a casual conversation that requires no action.

As DeepText digests Messenger conversations, it will then be able to move on to statuses.

The firm says that the AI will recognize when you are selling an item in your status.

And it will automatically offer to cross-list it in your regional network to help if sell faster.

The ultimately goal of DeepText is to give people information they need and want, rather than random advertisements that pop-up on their News Feed.

Deep neural networks is not a technological breakthrough, as both Google and Facebook are currently using them.

Just last year, Google gathered articles from CNN and Daily Mail to teach its AI programs to read.

Called DeepMind, the technology scanned articles as it picked up on certain linguistic relationship, prior to being asked specific queries -- the same technology that beat a human in the game Go.

Facebook has already started testing the AI in Messenger. It knows that when a user says 'I need a ride' they are talking about requesting a taxi. And when they say 'I like to ride donkeys', the AI understands this is a casual conversation that requires no action

And the search giant's most recent open source system, called Syntax, uses neural nets to understand grammatical logic of sentences.

However, unlike Google, Facebook isn't open sourcing its technology yet, as the firm has just started using DeepText for its own services, reports Wired.

And it has not been revealed when or if the firm will do so.

DailyMail.com has asked Facebook for comment, and is yet to hear a response. 

DeepText leverages several deep neural network architectures, including convolution and recurrent neural nets, similar to Google's Syntax, and has the ability to perform world-level and character-level based learning.

What experts say is the key to teaching a system to understand causal conversation is natural language in digital form and Facebook seems to have this part covered

Every minute 510 comments are posted, 293,000 statues updated and 136,000 photos are uploaded.

The social media site is a massive robotic library that they are using to expand its technology vocabulary.

DeepText leverages several deep neural network architectures, including convolution and recurrent neural nets, similar to Google's Syntax, and has the ability to perform world-level and character-level based learning. Facebook has not revealed when or if they will open source DeepText

'Text is a prevalent form of communication on Facebook,' states the firm.

'Text understanding on Facebook requires solving tricky scaling and language challenges where traditional NLP techniques are not effective. '

GOOGLE'S NEURAL NETWORKS AT WORK: HOW ALPHAGO WORKS 

Traditional AI methods, which construct a search tree over all possible positions, don't have a chance when it comes to winning at Go.

So DeepMind took a different approach by building a system, AlphaGo, that combines an advanced tree search with deep neural networks.

These neural networks take a description of the Go board as an input and process it through 12 different network layers containing millions of neuron-like connections.

One neural network called the 'policy network,' selects the next move to play, while the other neural network - the 'value network' - predicts the winner of the game.

'We trained the neural networks on 30 million moves from games played by human experts, until it could predict the human move 57 per cent of the time,' Google said.

The previous record before AlphaGo was 44 per cent.

Traditional AI methods, which construct a search tree over all possible positions, don't have a chance when it comes to winning at Go (pictured)

However, Google DeepMind's goal is to beat the best human players, not just mimic them.

To do this, AlphaGo learned to discover new strategies for itself, by playing thousands of games between its neural networks and adjusting the connections using a trial-and-error process known as reinforcement learning.

Of course, all of this requires a huge amount of computing power and Google used its Cloud Platform.

To put AlphaGo to the test, the firm held a tournament between AlphaGo and the strongest other Go programs, including Crazy Stone and Zen.

AlphaGo won every game against these programs.

The program then took on reigning three-time European Go champion Fan Hui at Google's London office.

In a closed-doors match last October, AlphaGo won by five games to zero.

It was the first time a computer program has ever beaten a professional Go player.

'Using deep learning, we are able to understand text better across multiple languages and use labeled data much more efficiently than traditional NLP techniques.'

With deep learning, the firm says it can use 'word embedding's, a mathematical concept that preserves the semantic relationship among words, instead of traditional NLP approaches that assign an integer ID – allowing you to see the exact word and not a numerical ID.

And using this method, DeepText can learn French or Spanish the same way it would English.

'Using word embeddings, we can also understand the same semantics across multiple languages, despite differences in the surface form,' explains Facebook.

'As an example, for English and Spanish, 'happy birthday' and 'feliz cumpleaños' should be very close to each other in the common embedding space.'

'By mapping words and phrases into a common embedding space, DeepText is capable of building models that are language-agnostic.'