Facebook Releases Open-Source Chatbot Blender, Claims It Beats Google’s Meena
Facebook has launched a new chatbot called Blender that it claims is the best in the world. The open-source chatbot is designed to handle conversation on any subject and show empathy when interacting with humans.
Blender is named for its ability to merge multiple conversational skills at once. The chatbot is built from what Facebook calls the neural model in the world of up to 9.4 billion parameters. The AI was trained using 1.5 billion examples of conversation, making it so large that it had to be broken up into pieces in order to handle larger sets of data. The AI then uses what Facebook calls Blended Skill Talk (BST) to merge various chatbot abilities. The goal is a chatbot with a stable personality that can converse naturally and with an understanding of the emotional context so that it can match the user’s mood. That’s crucial to avoid the chatbot sounding inappropriate or offensive.
“Blending these skills is a difficult challenge because systems must be able to switch between different tasks when appropriate, like adjusting tone if a person changes from joking to serious” the researchers explained. “Our new BST data set provides a way to build systems that blend and exhibit these behaviors. We found that fine-tuning the model with BST has a dramatic effect on human evaluations of the bot’s conversational ability.”
Meena and the Future
Facebook brought in people to evaluate Blender and its conversation using chat logs. The tests specifically compared Blender to Google’s new Meena chatbot, no doubt because of Google’s boast at the time of how human-like Meena can be during conversations. Two-thirds of the evaluators judged Blender to be more human-sounding, and three-quarters of the evaluators picked Blender over Meena as the chatbot with which they would rather have a long conversation. Arguably the most impressive response was that 49% of the judges said they’d pick a conversation with Blender over a human because of its blending skills.
That doesn’t mean Blender is flawless. The researchers made it clear that there is plenty of room for improvement for the AI. The accomplishments so far mark a beginning point. That’s partly why Blender is open-source as that means the public can experiment with ways to make the AI better.
“We’re excited about the progress we’ve made in improving open-domain chatbots. However, we are still far from achieving human-level intelligence in dialogue systems,” The researchers wrote. “Though it’s rare, our best models still make mistakes, like contradiction or repetition, and can “hallucinate” knowledge, as is seen in other generative systems. Human evaluations are also generally conducted using relatively brief conversations, and we’d most likely find that sufficiently long conversations would make these issues more apparent.”
Open-ended conversational chatbots are seeing a resurgence of late, fueled in part by the current COVID-19 pandemic. Governments and healthcare providers are very interested in using AI to communicate with people about what’s happening during the health crisis, as are companies that use call centers to respond to customers. Meena and Blender both carry forward the idea of a more human-seeming AI that can reflect a user’s mood. As an Apple study found, people are more likely to trust a voice AI that can mimic them. However the differing AI models are used, they could be part of everyday life in the not-too-distant future.