Facebook wants to be more intelligent.
To help, the company has built a deep learning program called DeepText - a primitive AI which is collecting all your statuses, posts and messenger requests and learning from them.
By analysing the language we use in posts, or messages, DeepText is able to understand us better, and in turn suggest more useful features.
A perfect example is when you talk about a specific subject in messenger. If you say something like, "I need to book us a taxi" then DeepText will tell Messenger to pop up a taxi link underneath your message.
What makes DeepText really clever is that it will know the difference between that and the phrase, "I just came out of the taxi."
If you want to get really technical then the way in which it does this is through a process called "word embeddings," a mathematical concept that creates relationships between different words.
DeepText knows that "mother" and "ma" are two very different words, but it also knows that their relationship is much closer than that.
It will then recognise the context, which can often be that "mother" is used in formal conversations and "ma" is used in more relaxed situations.
The hope is that in the future DeepText will become so good at understanding what we're saying that it will be able to recommend deeply contextual information such as trying to sell your bike on eBay or filtering spam messages.