Facebook is teaching robots to behave like humans

The world's largest social media company Facebook's artificial intelligence (AI) lab has developed a bot that can analyze the facial expressions of the people it interacts with and adapt itself to the human face.

It is powered by a deep neural network that was fed image data from 250 Skype video clips between two people as part of its training process.

Researchers identified 68 “facial landmarks” to detect subtle reactions and micro-expressions on human faces.

Bots can analyze people's facial expressions

“Although the human faces in our dataset are different, their expressions have similarities that can be extracted from the structure of facial landmarks,” Facebook scientists explain in an article published in New Scientist.

"For example, when people fumble with something, the configuration of their eyebrows and mouths reveals the most about their emotional state."

The scientists then tested the bot's reactions on a group of people.

The bot was asked to "look for how its facial expressions react to the user, especially facial images that appear natural, appropriate, and social" and decide whether or not to engage in conversation.

The results showed that the bot passed the test with its decision to choose facial expressions that were "natural and confident" as well as "realistic".

Human subjects interacting with bots are becoming increasingly common in many application areas such as education, healthcare and personal assistance, researchers say.

The next step in training the bot is to get it to interact with real humans in the real world, the scientists say.

According to Vietnam+