Humans still beat A.I. in "reading the room"

People pick up on cues that A.I. entities miss

Excerpt: "One thing AI can’t do? 

Read your body language. 

Researchers at Johns Hopkins University pitted human beings against AI models and asked each to assess short videos and images showing two people who might—or might not—be interacting with each other.

Both human volunteers and the AI models were asked to rate features important for understanding social interaction, such as how close the individuals were to each other, on a scale of one to five. 

In one video, a man who appears to be a chef in a professional kitchen speaks confidently to a smiling woman who returns his gaze. In another, a father chats with his toddler at a craft table, at one point wrapping his arm around the child. In a still image, a tennis player stretches to hit a ball as a linesman crouching several feet behind her looks on.

Humans mostly agreed with each other on their evaluations of the scenes. However, the AI video models were unable to accurately assess what people were doing, and image models that were given still frames of the scenes couldn’t reliably predict whether people were communicating."