Home / Mobile / Facebook Messenger’s AI ‘M’ suggests features to use based on your convos

Facebook Messenger’s AI ‘M’ suggests features to use based on your convos



Write “You owe me $20” and Messenger will suggest you use its payments feature. If someone says “goodbye,” Messenger will recommend a “bye-bye” GIF. That’s how Facebook’s new M Suggestions work.

Facebook’s artificial intelligence assistant will privately interject these recommendations above the redesigned one-line message composer in your Messenger conversations, and your friends won’t see the results unless you approve. M Suggestions roll out to all iOS and Android users in the U.S. today, with more countries coming.

They’re designed to surface all the features buried inside Messenger as it strives to become more than just an SMS replacement. Beyond stickers and payments, M Suggestions also can recommend location sharing, timed reminders, group chat polls and ridesharing options from Lyft and Uber.

The old M learns, the new M helps

You might remember M from when it began testing in 2015. Since then, a small number of users have been allowed to message M any request, which it tries to fulfill, first relying on AI to parse the question and retrieve answers it’s already learned. If it’s stumped, it falls back to a team of human staffers who try to help, and then teach M to do the job by itself next time.

The problem is “it cannot scale to millions of people” says Facebook M product manager Laurent Landowski, who co-founded the natural language processing startup Wit.AI that Facebook acquired.

So now, “The original version of M will continue to help us understand and learn more about . . . what we need to build and what people want.” Essentially, the open-ended M will continue to process requests for a closed set of beta testers so Facebook can determine what tasks its AI can reliably fulfill, and what text indicates what intent. Then it will slowly bring those capabilities over to M Suggestions.

Facebook began testing M Suggestions in December, as noted by BuzzFeed, and The Information reported that the full-fledged M personal assistant would be switching gears to become more of a training ground for Facebook’s AI.

Facebook’s first personal assistant at scale

M Suggestions will open Facebook’s AI to a much wider audience. With the new Messenger composer redesign, many of the buttons for its different features have been collapsed inside an expandable menu, but M Suggestions will help you fish them out.

And rather than apply the same experience to everyone, M will learn individual users’ habits and personalize itself for them. If you always ignore the M Suggestions, you’ll see less of them. Or if you never use its payments feature but constantly send stickers, it’ll just suggest you use the latter.

And if you really hate M, you can use the Messenger settings to mute it entirely or turn off certain kinds of recommendations. Even though it’s only an AI, not a human, some people might be a bit freaked out by Facebook scanning the content of their messages to provide these suggestions.

  1. M-Plans

  2. M-Stickers

  3. M-Poll

  4. M-Location

  5. M-Payments

  6. M-Rides

For now, M Suggestions are focused on Messenger’s internal features. But the inclusion of the Lyft and Uber ride-hailing option signal it could have bigger ambitions for recommending outside developer services. Facebook currently has a big issue with bot discovery, having launched the automated messaging agents last year without an easy way to find ones to use. M Suggestions could potentially evolve to recommend bots for you to pull into your conversations.

While there are plenty of problems M could one day help you solve, for now it’s dealing with one of Facebook’s longest-running issues: There are more features than you know what to do with.



Source link

About techguy

Check Also

NewStore raises $50 million for mobile commerce

Demandware founder Stephen Schambach has moved on to his next startup and investors are betting ...

Leave a Reply

Your email address will not be published. Required fields are marked *