這幅畫賣出300萬天價(jià) 但它的創(chuàng)作者卻“不是人”

佳士得最近舉行了一場為期三天的拍賣會(huì)，拍品中有20多幅畢加索的畫作，但有一幅畫作拍出的價(jià)格比畢加索的這些畫作都要高。這幅畫作的創(chuàng)造者是“算法”，其43.25萬美元的拍賣價(jià)在全場300多件拍品中名列第二。

本周，一幅具有顛覆意義的肖像畫出現(xiàn)在紐約佳士得拍賣行的拍賣臺(tái)上，此舉頗有歷史意義。這幅畫作由人工智能創(chuàng)作，是藝術(shù)界首次拍賣由算法創(chuàng)作的藝術(shù)品。

The creepy-looking painting of a fictitious man in a dark frockcoat left the auction block at Christie's for a whopping $432,500 on Oct. 25 in New York City.

這幅畫畫的是一位身穿黑色雙排扣長禮服的虛構(gòu)人物，看上去有點(diǎn)詭異。10月25日，在紐約佳士得拍賣行拍出了43.25萬美元(約合300萬元人民幣)的價(jià)格。

很多人可能覺得藝術(shù)品拍出這個(gè)價(jià)格很正常，但是看到畫作的簽名時(shí)估計(jì)你就不這樣想了。這幅畫的作者簽名是一個(gè)數(shù)學(xué)方程式：min max Ex[log(D(x))] + Ez[log(1-D(G(z)))]。

The portrait — designed in the "Old Master" style reminiscent of European fine artists from centuries ago — only partially fills the canvas, leaving empty space around the central figure. It appears to represent a man with a blurred face, dressed in clothing similar to that worn by subjects painted by the Dutch artist Rembrandt van Rijn in the 17th century.

這幅畫的顏料僅僅覆蓋了部分畫布，主體人像周圍留出了大片空白。畫作遵循數(shù)百年前的歐洲古典美術(shù)風(fēng)格，似乎描繪了一名五官模糊的男子，其穿著的外衣類似于17世紀(jì)荷蘭畫家倫勃朗·梵·萊茵筆下人物的穿著。

當(dāng)佳士得拍賣行決定拍賣這幅畫時(shí)，團(tuán)隊(duì)成員都深感意外，但佳士得認(rèn)為這次拍賣將標(biāo)志著世界拍賣舞臺(tái)上人工智能藝術(shù)的到來。

在拍賣之前，拍賣行給出的估價(jià)是1萬美元。但是沒想到，這幅畫作遭到了瘋搶。一位匿名的競拍者進(jìn)行了5次競價(jià)，最終拿下了該畫作。

Of course, a computer didn't spontaneously pick up a brush and become an artist. The AI that generated the image had human programmers — a Parisian art collective called Obvious, Christie's reported. Their collaboration, titled "Portrait of Edmond De Belamy," is part of a series of paintings of the fictional Belamy family and was expected to fetch $7,000 to $10,000, according to Christie's.

當(dāng)然，計(jì)算機(jī)本身并不會(huì)自發(fā)地拿起畫筆，變身藝術(shù)家。據(jù)佳士得介紹，創(chuàng)作這幅肖像畫的人工智能依靠了人類程序員——一個(gè)名為“顯而易見”的巴黎藝術(shù)團(tuán)體。按照佳士得的說法，這幅畫題為《埃德蒙·德-貝拉米像》，是虛構(gòu)的貝拉米家族系列畫作中的一件，原本估價(jià)7000至10000美元。

To generate the portrait, the Obvious team first fed the neural network a diet of 15,000 images painted between the 14th and 20th centuries, to train it to recognize visual elements in fine art, Obvious artist Hugo Caselles-Dupré told Christie's. The algorithm that eventually created an original image had two parts that worked against each other, called the Generator and the Discriminator; they dubbed this combative AI "generative adversarial network" (GAN), Caselles-Dupré explained.

“顯而易見”團(tuán)隊(duì)的藝術(shù)家雨果·卡塞列斯-迪普雷告訴佳士得拍賣行，為生成這幅肖像畫，他們首先將15000幅14至20世紀(jì)作品的圖像輸入一個(gè)神經(jīng)網(wǎng)絡(luò)，訓(xùn)練它辨認(rèn)藝術(shù)品中的視覺元素。據(jù)介紹，最終生成原創(chuàng)畫作的算法由兩個(gè)相互矛盾的部分組成，分別被稱作“‘生成者與“辨別者”。“顯而易見”團(tuán)隊(duì)把這種斗爭性人工智能稱為“生成式對(duì)抗網(wǎng)絡(luò)”。

"The Generator makes a new image based on the set, then the Discriminator tries to spot the difference between a human-made image and one created by the Generator. The aim is to fool the Discriminator into thinking that the new images are real-life portraits. Then, we have a result," he told Christie's.

卡塞列斯-迪普雷說：“‘生成者’依據(jù)作品庫生成新圖像，然后‘辨別者’會(huì)努力分辨人類畫作與‘生成者’畫作的區(qū)別。我們的目的是把‘辨別者’搞糊涂，讓它以為新圖像就是真人創(chuàng)作的肖像畫。這時(shí)，我們的成果就產(chǎn)生了。”

GAN's final image was then inkjet-printed on canvas and framed, according to Obvious. At the bottom of the portrait is a mathematical formula representing the algorithm that created it, a nod to the relationship between the Generator and the Discriminator, Obvious artists wrote on the collective's website.

然后，“生成式對(duì)抗網(wǎng)絡(luò)”創(chuàng)作出的成品會(huì)被噴墨打印機(jī)繪制在畫布上，并裝框成畫。“顯而易見”團(tuán)隊(duì)在其官方網(wǎng)站上介紹說，《埃德蒙·德-貝拉米像》的畫面底端有一行數(shù)學(xué)公式，代表了生成這幅畫的算法，肯定了“生成者”與“辨別者”之間的關(guān)系。

The goal of the painting and of Obvious, also co-founded by Hugo Caselles-Dupré and Gauthier Vernier, was to prove "artificial intelligence can do more than operate driverless cars or transform manufacturing — it can be creative," CNBC reported.

據(jù)美國全國廣播公司財(cái)經(jīng)頻道報(bào)道，“顯而易見”團(tuán)體創(chuàng)作這幅畫是為了證明“人工智能不僅能駕駛無人汽車，或者改變制造業(yè)，它還很有創(chuàng)造性。”這一團(tuán)體由雨果·卡塞列斯-迪普雷和戈捷·維爾涅共同創(chuàng)立。

Portraiture is a tough genre for AI to take on, according to Christie's, "since humans are highly attuned to the curves and complexities of a face in a way that a machine cannot be."

佳士得拍賣行表示，對(duì)于人工智能來說，肖像畫是非常難呈現(xiàn)的一個(gè)繪畫體裁，“因?yàn)槿藗兒苌瞄L畫臉部的曲線和復(fù)雜部位，而人工智能往往做不到。”

This difficulty was part of Obvious' thinking when they created the portrait.

“顯而易見”團(tuán)體創(chuàng)作這幅畫時(shí)也考慮到了這個(gè)難點(diǎn)。

"We did some work with nudes and landscapes, and we also tried feeding the algorithm sets of works by famous painters. But we found that portraits provided the best way to illustrate our point, which is that algorithms are able to emulate creativity," Caselles-Dupré said.

卡塞列斯-迪普雷說：“我們還用人工智能創(chuàng)作了一些裸體畫和風(fēng)景畫，還曾嘗試把名畫家的畫作輸入進(jìn)去。但我們發(fā)現(xiàn)肖像畫能最好地闡明人工智能畫作的特點(diǎn)，也就是算法也可以模擬出人類的創(chuàng)造性。”

"Edmond de Belamy" is one of 11 AI paintings made by Obvious.

“顯而易見”團(tuán)體共創(chuàng)作了11幅人工智能畫作，這是其中一幅。