How Do You Cre­ate a New Rem­brandt?

A com­puter be­comes the Dutch Mas­ter’s best pupil

iD magazine - - Contents -

With the help of a corpse, in 1632 Rem­brandt van Rijn trig­gers a rev­o­lu­tion. The 25-year-old draws a corpse be­ing dis­sected by a doc­tor who is sur­rounded by stu­dents—and in so do­ing, breaks with cen­turies of con­ven­tion in paint­ing. Be­cause the sub­jects depicted in his work are not shown stand­ing next to one an­other in neu­tral poses as was cus­tom­ary at the time; in­stead, they are stand­ing be­hind one an­other in lively poses. The most notable in­no­va­tion: Faces re­flect ex­cite­ment and fas­ci­na­tion. “Against the dark back­ground their emo­tions look par­tic­u­larly strong, as if the painter had dipped his brush not in pig­ment, but rather in sun­light,” re­marks psy­chol­o­gist Bar­bara Diggs. This play of col­ors is what makes the 346 known “au­then­tic Rem­brandts” un­mis­tak­able. This num­ber, 346, was con­sid­ered valid for the 346 years fol­low­ing the mas­ter’s death in 1669. Then, on April 5, 2016, the lu­mi­nary’s 347th paint­ing was un­veiled in Am­s­ter­dam. It was cre­ated by a com­puter ca­pa­ble of sim­u­lat­ing the artist. But what is “The Next Rem­brandt” in ac­tu­al­ity? And could an ar­ti­fi­cial in­tel­li­gence re­ally be cre­ative?

To this day, as­pir­ing painters learn to hone their craft by try­ing to copy the im­ages made by the great masters. A com­puter can do the same—just much more dili­gently, with un­tir­ing per­sis­tence: Be­gin­ning with the knowl­edge ba­sis of a young child, a com­puter net­work took sev­eral months to an­a­lyze how Rem­brandt painted the de­tails of a face—which pro­por­tions he chose, how he used col­ors. This process is called “deep learn­ing.” Us­ing a to­tal of 168,263 im­age de­tail ex­cerpts that were taken from all of Rem­brandt’s paint­ings, a 20-per­son pro­gram­ming team even­tu­ally pro­duced al­go­rithms that de­liver the same com­po­si­tions as Rem­brandt cre­ated at his easel. “We’ve dis­tilled the painter’s artis­tic DNA out of his body of work,” says project direc­tor Bas Korsten. And build­ing on these ca­pa­bil­i­ties, the soft­ware can now even cre­ate its own mo­tifs. The re­sult: a fic­ti­tious 35-year- old with a beard, hat, and col­lar who looks as if Rem­brandt drew him. The per­fect forgery? This method does not per­mit that pos­si­bil­ity: A chem­i­cal anal­y­sis of the color pig­ments would re­veal the truth. The gen­er­ated im­age cor­re­sponds with the abil­i­ties of Rem­brandt around the year 1630, as his later works are still too dif­fi­cult to at­tempt. But just as Rem­brandt did not stop paint­ing af­ter that rev­o­lu­tion­ary anatomy lec­ture, so too will Mi­crosoft con­tinue this project. “Deep learn­ing” has only just be­gun…

“WE HAVE USED TECH­NOL­OGY AND DATA AS REM­BRANDT USED BRUSHES AND COL­ORS.” RON AU­GUS­TUS, MI­CROSOFT

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.