Facebook has created a chatbot to rival that of Google’s Meena, that it brought to life using 1.5 billion Reddit posts as examples of human interactions, and now the AI can show off a “diverse” set of conversational skills, including “empathy, knowledge, and personality” which already puts it miles ahead of my ex-boyfriend.
Blender – as it’s called – was cobbled together by the Facebook AI team, and is apparently a state-of-the-art open source chatbot.
When conversing with humans, the flesh bags said that Blender felt more human than other chatbots. With Google’s Meena being used as a benchmark, 67 per cent of people said Blender sounded more human than Meena, while 75 per cent would prefer to have an extended chat session with Blender over Meena.
Interestingly, when comparing human-to-human interactions, and human-to-Blender conversations, 49 per cent of people preferred chatting to Blender over its meaty counterparts. Of course, using Reddit as a model of conversations has some not-very-ideal effects as well, with Blender replying to his human pals with offensive language, and making up facts – bringing him back down to the level of my ex-boyfriend. Oh Blender, you suck.
The team isn’t discouraged though, saying it’s building stronger classifiers that will filter out “harmful language”. [BBC News]
This post originally appeared on Gizmodo UK, which is gobbling up the news in a different timezone.