Facebook on Monday is releasing new software that artificial intelligence (AI) researchers can use to test their systems for having text conversations with people.
The end goal: to help speed the development of chat bots that seem more human -- without painstakingly training them on millions of different scenarios.
"Ultimately one of the objectives of this is to have your own digital friend, your virtual assistant that is basically customized for you and under your control," Yann LeCun, head of Facebook's Artificial Intelligence Research (FAIR) group, said in a meeting with reporters.
The software, called ParlAI (pronounced like "parlay"), comes with data sets in several research areas -- like recommending restaurants and discussing movies -- that researchers can train their systems on. Having so many domains available means researchers will be able to teach their models how to a handle a wide variety of questions and statements. Facebook will show the highest-performing systems on a web-based leaderboard for the wider community to explore.
Facebook is releasing the ParlAI software under an open-source license. Google's DeepMind released a similar open source AI training environment a few months ago, but the Facebook tool is purpose-built for evaluating dialogue, said LeCun.
The FAIR researchers work closely with the people who develop Facebook M, a virtual assistant that can now do provide location information and other data for you in the Messenger messaging app. Messenger also offers chat bots from companies like Nike, Western Union and Whole Foods. Microsoft's Skype app and privately held Slack's team communication app also let their users talk with third-party bots.
Over time, Facebook hopes that this public testing ground will help make all of these chat bots more useful, and eventually turn them into vital parts of our lives.
"It mediates your interactions with your world, your friends and other people. It helps you go through an incredible amount of information we're bombarded with every day. That will take a while before those things are general enough that they can take care of all the things of a human assistant. We're talking decades."