Culture is future »


Contribution: "The Next Rembrandt: When the machine surpasses the master" by Mathilde Funck Brentano

After the overwhelming victory on the Go world champion, artificial intelligence has proved that it is able to create an original painting in the style of Rembrandt. The project The Next Rembrandt project opens vertiginous perspectives on the combination of machine and art, what was discussed during the debate “What are algorithms dreaming of?” at the Forum d’Avignon @Bordeaux in the end of March. 

"We really wanted to understand what makes a face look like a Rembrandt" told to the BBC Emmanuel Flores, director of technology for The Next Rembrandt project which gathers Microsoft, ING, TU Delft (university of Technology) and two prestigious museums: the Mauritshuis and Rembrandthuis. Their roadmap was immense: looking for possible connections between technology and art, but also testing artificial intelligence’s potential in an artistic activity. The first step was to collect data about the style and the techniques used by Rembrandt, based on 346 HD 3D scanned paintings. The choice of the Flemish painter is not a coincidence: he is one of the most difficult painters to identify.

From the ‘deep learning’ to the ‘deep painting’. The algorithm “Next Rembrandt” received in a second time a simple direction: “Draw in the style of Rembrandt a portrait of a caucasian man between 30 or 40 years old looking to the right”. With an unknown factor: the model. The algorithm has produce, even “imagine”, a picture thanks to its capacity of “deep learning”: its “independent learning” is based on 160 000 details such as the gap between the eyes or the shape of the face, as many potential facial features of the “Next Rembrandt”.

Accumulating thousands of morphological details is not enough. It was also necessary to tame the specific features of Rembrandt’s style, which is easily recognizable by the chiaroscuro, by brown and mud-coloured tones and by visible brushstrokes. From the beginning of his career, Rembrandt “sculpted” the pigments on the canvas. It is this characteristic “relief” that interested the researchers. To make the painting look even more realistic, the machine can discern and reproduce the brushstrokes. Once the digital image produced, the “painting” work with real pigments has been made possible by a 3D printer which permits to split up the paint layers and the successive brushstrokes of the painter. In the same way as the artist processes by small touches, the machine dropped off the pigments in 13 layers of different colors and thicknesses, reproducing the typical depth and light effect of the young Rembrandt for a result as credible with the naked eye that an original.

« We use technology and data analysis as Rembrandt used his paintbrush to create something new » : the words of Ron Augustus developer at Apple illustrate the ambition of the successful project after 18 months of efforts by the production of an unprecedented work « in the style of » which can be compared to the work of an apprentice capable of imitating his master. The machine still can be improved: if it is able to create a « young Rembrandt » it seems more complicated to realize a masterpiece like Rembrandt did at the end of his life. Indeed, the artist multiplied stylistic breaks and innovations, without talking about the overwhelming humanity reflected by his paintings. However, the algorithm is still lacking a proper style which characterizes artists.

This new step towards a « creative » artificial intelligence confirms or opens huge ranges of application. In the edition, the algorithm is already able to analyze special features of a style, while raising theoretical squabbles on the « authors’ reality » from Molière to Shakespeare. The publishing house Short Edition announced that it wanted to produce an algorithm able to judge the quality of a book and of its author’s style and so to choose to publish it or not. The publishing houses are not the only ones to use the astounding progresses of digital. The museums also integrate them to value their heritage in a different way. For example, the National Gallery of Canada in Ottawa authorizes visitors to touch the most famous masterpieces as the Iris by Van Gogh to allow them to feel the brushstrokes on the painting… This was made possible thanks to reproductions by 3D printer.

« In this project, data is used to make life better, to touch the human soul, confided Ron Augustus, developer at Apple, while data is used by a lot of people to help them to be more efficient in their everyday life or by great companies for economic reasons.” The Forum d’Avignon in Bordeaux already wondered about “What are algorithms dreaming of?” on the role of data for culture with Philippe Torres, deputy director and Head of consulting in Digital strategy at Atelier BNP Paribas, Pierre-Louis Xech, DX Research Lead, Microsoft France, Gary Shapiro, CEO of Consumer Electronics Association, Dominique Cardon, sociologist at Orange Lab, Cynthia Fleury, philosopher, Associate Professor at the American University of Paris, ORLAN, artist and moderated by Eric Scherer. A central question remains: when will an original act of creation be performed by a robot? With The Next Rembrandt, artificial intelligence already imposes itself in a spectacular way on artistic fields and contradicts prejudices on his capacities, supposedly technical or practical only. For better (the scientific reconstitution of lost paintings) and worse (mercantile usurpations). 

About Mathilde Funck Brentano

Mathilde Funck-Brentano is a student in La Rochefoucauld high school in Paris, in Economic and social class. Recently intern at the Forum d'Avignon, she would like to study political science and economics.

Read also: everything on Data & Algorithmes by the Forum d’Avignon