Facebook has released an open-source framework that allows human tutors to teach AI chatbots how to have more natural conversations.
The company's artificial-intelligence research team, known as FAIR, has released the Python-based ParlAI tool, which gives researchers the ability to train bots over many different datasets at once, including benchmarking datasets like WebQuestions, WikiMovies and bAbl tasks.
Training is split into five tasks: general question answering; sentence completion, where the bot has to fill in the next word of a dialogue; back-and-forth chit-chat; visual dialogue with conversations involving images; and goal oriented dialogue, where a conversation is intended with a specific outcome.
"For the first time, ParlAI provides a place for researchers to collect all the significant dialog tasks in one place," wrote FAIR researchers Jason Weston, Alexander Miller and Will Feng in a blog post.
"Not only does this allow for much easier iteration on any of these tasks individually, but also makes it possible to easily train a bot on all of them (which eventually should lead to better bots) and to evaluate bots across all those skills."
The framework uses Amazon Mechanical Turk, a programme which uses low-paid human operators to do repetitive grunt-work that goes into training AI models, to improve the quality of AI conversations through question-and-answer sessions.
The release comes as part of the company's push into the AI space. The goal is to provide advertisers and brands with the ability to roll out natural-language chatbots on platforms like Facebook Messenger and WhatsApp, allowing customers to make enquiries, bookings and purchases just by talking to an AI helper.
"ParlAI is a platform we hope will bring together the community of researchers working on AI agents that perform dialog," Facebook's researchers said, "and continue pushing the state of the art in dialog research."
