Skip to main content

Facebook wants its AI chatbots to chat more like humans

Just over one year since Facebook first unveiled chatbots, the social network giant isn't done making much-needed improvements to dialog-driven assistants.

Today, Facebook Artificial Intelligence Research (FAIR) released a new framework for developers intended to help fine-tune and improve the conversational abilities of chatbots, of which Facebook Messenger has over 30,000 as of the end of last year.

ParlAI (pronounced "parley") is an open-source framework of over 20 different Python-based data sets for training and researching dialog models that power everyday chatbots.

FAIR believes that by supplying a unified platform for researchers and developers, chatbots can use the data gathered by the ParlAI community to make chatting with a bot faster and more human-like than, well, robotic.

One example of how ParlAI could improve chatbots is its potential to bolster multitasking. By teaching a dialog model to understand more than one sentence at a time, a chatbot can handle more natural conversation instead of short or stilted single commands.

ParlAI isn't FAIR's first pass at improving artificial intelligence, either. The open-source framework joins the likes of FastText, which helps AIs classify text quickly, and CommAI, a general repository for artificial general intelligence, as other Facebook-backed tools designed to put the company on the bleeding edge of AI research.