The Story Behind Berenson the Robot - Interview with Denis Vidal and Philippe Gaussier

New Media, Interviews

March 16, 2016

Two weeks ago, a vast number of important and relevant media outlets published the news about a new robot – a robot that is named Berenson and that could be an art critic. Berenson appeared at Paris's Musée du Quai Branly during the exhibition Persona: Oddly Human. So, what is Berenson the robot actually capable of? He is able to analyze the facial expressions of fellow museumgoers, sort them into "negative" and "positive" categories, and feed the information into a neural network simulator called Prométhé. The program allows Berenson to use this data to develop an overall impression about the art object. In case it is positive, he smiles and moves towards it. If it is negative, he frowns and moves away.

This news left many art enthusiasts and art critics with a lot of questions. Can a robot really be an art critic? Was the intention of Berenson’s inventors to create a robot that would be an art critic, or the intention was something else? Is it possible that non-humans become one of the major actors of art criticism? In order to find out answers to these (but other) questions, we spoke with Berenson’s creators – Mr. Denis Vidal and Mr. Phillipe Gaussier. Denis Vidal is an anthropologist. He is a Research Senior Fellow at the Institute of Research for Development (IRD-URMIS/Paris Diderot University). Philippe Gaussier is Professor of Robotics and Artificial Intelligence at the University of Cergy-Pontoise. We had an honor to speak with both of academics. They were so kind to jointly participate in answering our questions. Since this subject is quite complex, it covers a variety of different areas (it’s not only about robots and art, but also about the issue of perception, about aesthetic taste, about robotics…).

Scroll down and read more about the robot “art critic”!

Robots and Art Perception

Widewalls: Scientists have been creating different types of robots; these robots have different purposes, but we haven’t seen any robot, until now, that would be an art critic. What is so special about contemporary art (and art in general) that motivated you to create a robot that would be an art critic (that is an art critic)?


Denis Vidal and Phillipe Gaussier:  Berenson is more an art visitor than an art critic. Let us be clear about it. We never intended to make a robotic art critic. What we intended to do, from the beginning, was to explore the potential of aesthetic judgements at a cognitive level and from an interactional point of view. It is what we are trying to do in the context of a museum with this humanoid robot.  Berenson is learning through its interactions with other visitors.

Think about a portrait. It is not because a painting or a picture may be inspired by a person that people will necessarily confuse it with this person. Remembering the fact that we never intended to make a robotic ‘art critic’, we called it Berenson and gave it the loose appearance of this famous nineteenth century art critic because we wanted - not always successfully - to dissociate our project from all sort of mythologies and of speculative fantasies which are surrounding the future of robotics. But there is still a lot to be done in this regard.

Widewalls: In a short video, we saw how the robot functions. Still, many questions could be posed. One of them is: What authority does the robot have to characterize one piece of art as “good”, and the one as “bad”?

D.V. and P.G: Let us also be clear on this point: our robot has absolutely no authority of any sort to characterise any piece of art as ‘good’ or ‘bad’ and, certainly not, either, to impose its ‘ artificial taste’ on others? What we are trying to do progressively is to make a robot which may acquire, develop, express and share some sort of ‘artificial taste’ by  interacting with art objects and with visitors in a Museum. It is also to explore how the acquisition of an artificial taste may have an  impact on  its general behaviour   outside of the museum and its ability to accomplish various  tasks; this is a very different from  building  a robotic  ‘art critic’.

Widewalls: Do you believe there is any authority at all (including humans) that would judge the artistic value of certain piece of art?

D.V. and P.G: No, we don’t believe so; but we believe in the fundamental importance of one’s ability to form one’s own aesthetic taste and judgment and to share it with others.  By the way, is this not precisely of  job of art critics ?

Widewalls:  Should human art critics be afraid for their jobs?

D.V. and P.G: Definitively not. As we have said before, we are exploring the rudimentary stages of the development of an aesthetic taste and its potential impact on cognition and social interaction. If something, the behaviour of Berenson should be rather compared   to the one of a toddler visiting a museum. It has nothing to do with specialised art criticism which involves social, cultural  and existential dimensions, which are well beyond any foreseeable progress of A.I. .If anything, we have a lot to  learn from the best  art specialists and critics ; and  we do so often in dialogue  with them.

Widewalls: Do you believe that, in the near future, there would be even more robots art critics, who perhaps, would judge art better than humans? As robotic art is growing, do you expect that we see even more robot art critics?

D.V. and P.G: Definitively not. What may eventually happen is the ability to build robots which may explore their environment on the basis of their ‘artificial tastes’, which may be useful ….or not, depending on their use.

About Berenson of Today and Its Future

Widewalls: It seems that Berenson judges a piece of art only based on its aesthetics, or, am I wrong? Is Berenson capable of judging the concepts as well (maybe the pieces of conceptual art)?

D.V. and P.G: Berenson is simply learning to respond to certain features of its environment with  ‘positive’  forms of behaviour ( smiling while looking at it going in the  direction of what it likes, etc.) ; and other features of its environment with ‘negative’ ones ( going away from it, frowning at it, etc.) . Let us also precise that it is not necessarily the same features that it is using to interpret the reactions of the visitors. To give a simple example, we do not always express our feelings about works of art by our facial expressions. So it is not one of the clues that Berenson will use while interacting with the visitors.

Berenson obviously can’t understand any concepts at all.  But the people who are observing its reaction seem often tempted to conceptualise his behaviour at all sorts of levels (anthropological, sociological, technological, etc.) it. There is also a significant gap between what Berenson can effectively do and how people interact with it. What is equally significant from an anthropological point of view is that it remains true even when people are  fully aware of its mechanical limitations.

Widewalls: I suppose that there is always a space for certain improvements. Do you plan to work on Berenson in the future, or your job is done? Can we expect to see Berenson in other art spaces, besides Paris's Musée du Quai Branly?

D.V. and P.G: As you may have guessed from our previous answers, we are still at the very first stages of this project. There is still a lot of work to do before we can fully judge its theoretical and practical implications. And we are certainly interested to present our robot in other art spaces if other museums are interested, too.

Widewalls:  Finally, what the creation of an art critic means for academia? Mr. Vidal, you are an anthropologist, while you Mr. Gaussier are a robotic engineer? How much important the creation of Berenson is for your professions

D.V. and P.G:  Let us remind once more you that we never intended or claimed to make a robotic ‘art critic’.  Nor did we intend, by the way, to survey the behaviour of Museum visitors in any sense. Berenson was built with the help of the Patrima foundation and of the Musée du Quay Branly. It is also involving the work of various students and researchers ( Joffrey Becker, Ali Karouazene, Aliaa Moualla, Sofiane Boucenna, Mathias Quoy) . Our first aim, as we explained before, is to explore the potential of aesthetics through a robot,  and to research how  it may evolve some sort of artificial ‘taste’ by interacting with art objects and with visitors in a museum; both as a mediation tool  and as a robotic tool for understanding the first steps of aesthetic preferences. It is why we have the help of various museal institutions.

Now, from the robotics side, we had to address several issues starting from having a safe robot! Berenson has to be able to learn online and to use one shot learning to associate some local views of an object to a given value provided by a visitor. Next, Berenson has to be able to recognize and come back to one of the learned objects from different starting positions (pattern recognition issues). One of the opportunities for us is to study long-term learning in interaction with the visitor of the museum (not some experts in AI) in order to see if the robot can really learn on its own in a developmental way. We can test how robots starting with exactly the same neural network will become robots with quite different preferences according to the history of their interactions with the visitors. Another very important aspect of this work is related to the study of Human-Robot interactions. What are the important features allowing a visitor to easily guess what the robot is intended to do without any explicit communication? The control of the robot gaze direction is clearly one important component as well as the expressiveness of its head. Another very important element looks like to be related to the robot capability to react synchronously to the human partners (desynchronized activities of the robot looks like to prevent interactions with the visitors).  All of this will allows us to have a better understanding about how we interact with robots.

Finally, an important aim of the project is to present robotics in a different light, with the hope of opening the debate about its potential uses and implications, not only to academics and roboticists but also to the general public. This is also one of the main aims of the exhibition Persona, étrangement humain  which is going on at  the Musée of the Quai Branly in Paris. It is no coincidence that we have included contemporary art in this exhibition, with the help of Thierry Dufresne, a well-known art specialist. From this perspective, I don’t believe our aim to be fundamentally different from that of  some of the most creative contemporary artists.

If you are more interested in these subjects and if you want to find out more about the project ‘Berenson’, you may consult various writings where Vidal and Gaussier analyze in more detail some of the first implications of this project:

  • Denis Vidal Aux frontières de l’humain, Paris, Alma Editions, 2016-03-05
  • Karaouzene, A., Gaussier, P., & Vidal, D. (2013, August). A robot to study the development of artwork appreciation through social interactions. In Development and Learning and Epigenetic Robotics (ICDL), 2013 IEEE Third Joint International Conference on (pp. 1-7). IEEE.
  • Boucenna, S., Gaussier, P., Andry, P., & Hafemeister, L. (2014). A robot learns the facial expressions recognition and face/non-face discrimination through an imitation game. International Journal of Social Robotics, 6(4), 633-652.
  • Boucenna, S., Gaussier, P., & Hafemeister, L. (2014). Development of first social referencing skills: Emotional interaction as a way to regulate robot behavior. Autonomous Mental Development, IEEE Transactions on, 6(1), 42-55.

Featured Images: Berenson the Robot (courtesy of; Berenson in Paris' Museum (courtesy of; Berenson (courtesty of All Images used for illustrative purposes only.