From behind, you may not even know the Berenson robot isn't human.
With a top hat, delicately wrapped white shawl, and long black coat, it looks like any other art lover hopping between galleries in Chelsea.
But Berenson is, in fact, a robot — just with some unique human tastes.
Developed in 2011 by robotics engineer Philippe Gaussier and anthropologist Denis Vidal, the robot is capable of judging whether it likes a piece of art and shaping its artistic preferences.
That may not seem like a huge deal, but seeing a robot that can exhibit a certain amount of emotional and aesthetic preference isn't commonplace. There's Pepper, a Japanese robot capable of providing emotional support, but not many others that go beyond doing physical tasks.
Berenson works by recording art goers' reactions to pieces of art, and then using the data it collects to create its own artistic preferences.
"If we smile to the robot while presenting a neutral object, it will learn to associate it to a positive value and will try to grasp it," Gaussier told Tech Insider in an email. "At the opposite, if we are doing an angry face, the robot will learn to avoid the object."
The robot made the rounds at an exhibit titled "Persona: Oddly Human" at the Quai Branly museum in Paris Tuesday to show off its abilities.
The exhibit, which is showing through mid-November, aims to explore how inanimate objects become animate and how people establish relationships with objects.
On Tuesday, visitors showed Berenson one object they either liked or didn't like. After he was shown 20 statues, Berenson was able to relay whether he had a positive, negative, or neutral reaction to a piece of art.
The robot sees through a camera in its right eye where it records a black-and-white image of a piece of art, the International Business Times reported. In this case, that image was sent to a computer tucked behind a wall in the museum.
Berenson then zeroes in on aspects of the art that it likes, which is represented by green circles, and doesn't like, which is pointed out in red. If it likes something, the robot will even head in its direction and smile.
Some may wonder then if Berenson is really forming its own opinion or simply copying what it learns from others. The truth is, it's a mixture of both. The robot runs on a neural network, meaning it learns based on positive reinforcement.
"Berenson generalizes its learning to new objects and even to people, allowing sometimes some strange behaviors where the robot smiles to some people and gives a sad face to others," Gaussier told Tech Insider.