A Trained Canvas
令人屏息的絕美AI藝術
The MoMA Dabbles in a New Way of Making Art
FAREED ZAKARIA, CNN CORRESPONDENT
This piece [is] called Unsupervised. It is currently on display at the Museum of Modern Art in New York City. The artist, Refik Anadol, trained an AI model using data from more than 200 years’ worth of the museum’s art collection, which included nearly 90,000 works of art from over 26,000 artists. The machine is always learning and imagining new artworks. If you watched for a hundred years, you would not see the same screen twice.
I sat down with Anadol and the museum’s curator of paintings and sculpture, Michelle Kuo, to discuss this extraordinary work. Here is a piece of that conversation.
Tell us a little bit about how you trained it.
REFIK ANADOL, MEDIA ARTIST
There’s actually [a] supercomputer behind this wall. So, we … there’s [there are], like, literally two computers. One is literally preparing our next dream, while the current one is showing us a new dream …
FAREED ZAKARIA, CNN CORRESPONDENT
Yeah. Yeah.
REFIK ANADOL, MEDIA ARTIST
… that we are witnessing. And these two machines is [are] trained in a way that their decisions are also interacted by [interacting with] a camera, and a microphone and the weather conditions. A rainy day is a different day for us and also for [the] AI. It’s a loud day. It’s a calm day. It is a moment early morning [when] students comes [come] or maybe late afternoons [when] people are more calm, and serene, and focused and maybe meditative. These [are] all states of, like, our wondering …
FAREED ZAKARIA, CNN CORRESPONDENT
So, this data is being fed in.
REFIK ANADOL, MEDIA ARTIST
Yes.
FAREED ZAKARIA, CNN CORRESPONDENT
It’s not a closed loop.
REFIK ANADOL, MEDIA ARTIST
No.
FAREED ZAKARIA, CNN CORRESPONDENT
It’s not just the painting[s].
REFIK ANADOL, MEDIA ARTIST
Exactly. And there’s a dialogue between human and machine, too.
MICHELLE KUO, THE MARLENE HESS CURATOR OF PAINTING AND SCULPTURE, MOMA
What’s interesting [is] you’ll notice different modes, and Refik, I think, has carefully choreographed and composed them. But these different modes are special because with generative AI, first of all, Refik is choosing the dataset, which is a huge step. It’s not just using everything that’s out there …
REFIK ANADOL, MEDIA ARTIST
Yeah. Yeah.
MICHELLE KUO, THE MARLENE HESS CURATOR OF PAINTING AND SCULPTURE, MOMA
… as other kinds of machine learning models are doing right now. And the other thing that Refik is doing is basically working very, very hard in concert with the machine learning model to create the shapes that you see. You might just get noise. If you were to yourself say, “Here’s a generative AI. I’m just gonna plug all this information into it,” you might get stuff that doesn’t make any sense, or is chaotic or doesn’t visually resolve into anything. So, that, to me, is what is so, sort of, mind-blowing about what you’re doing.
FAREED ZAKARIA, CNN CORRESPONDENT
Do you think that all artists in the future are gonna have to know some degree of AI?
REFIK ANADOL, MEDIA ARTIST
I mean, that’s a very beautiful, wishful thinking, but I think … I hope everyone [will] have an understanding of what AI is because it’s beyond the art—it’s for humanity. It’s a powerful technology, and I think in the human history, we[’ve] never have [had] something [this] powerful and [that] has the [this] imagination and the thinking intelligent [thinking] capacity before.
CNN特派員 法瑞德.札卡利亞
這件作品叫做《無人督導》。它目前在位於紐約市的現代藝術博物館展出。藝術家拉菲克.阿納多爾用來自該博物館藝術品收藏相當於兩百多年的資料訓練了一個人工智慧模型,其中包括超過兩萬六千名藝術家筆下近九萬件的作品。這部機器一直在學習和想像新的藝術作品。你要是看一百年,同樣的畫面你不會看到兩次。
我和阿納多爾及該博物館的繪畫與雕塑策展人蜜雪兒.郭坐下來一起討論這件非凡的作品。以下是那段談話的片段。
稍微跟我們說說你是怎麼訓練它的。
媒體藝術家 拉菲克.阿納多爾
這面牆後面其實有一部超級電腦。所以我們……確實有兩部電腦。一部實際上正在準備我們的下一個夢境,而當前這一部則向我們展現一個新夢境……
CNN特派員 法瑞德.札卡利亞
是。是。
媒體藝術家 拉菲克.阿納多爾
……是我們正在看到的。而且這兩部機器的訓練方式是它們的決定也會與一架攝影機、一支麥克風和天氣狀況相互影響。雨天對我們而言是不一樣的日子,對人工智慧來說也是。今天很熱鬧。今天很平靜。這一刻是一大早學生們來,或可能是傍晚時候人們比較平靜、安詳、專注,也許在沉思著。這些所有狀態都是我們思索著……
CNN特派員 法瑞德.札卡利亞
所以這數據正在被餵進來。
媒體藝術家 拉菲克.阿納多爾
是的。
CNN特派員 法瑞德.札卡利亞
這不是一個封閉的迴圈。
媒體藝術家 拉菲克.阿納多爾
不是。
CNN特派員 法瑞德.札卡利亞
不是只有畫作。
媒體藝術家 拉菲克.阿納多爾
沒錯。而且這裡也有人和機器間的一場對話。
現代藝術博物館瑪琳.赫斯繪畫與雕塑策展人 蜜雪兒.郭
有趣的是,你會注意到不同的模式,而且我認為拉菲克精心編製和建構了那些模式。不過這些不同的模式很特別,因為首先,拉菲克會用生成式人工智慧挑選資料集,這是很重大的一步。這不只是使用那裡現有的所有東西……
媒體藝術家 拉菲克.阿納多爾
對。對。
現代藝術博物館瑪琳.赫斯繪畫與雕塑策展人 蜜雪兒.郭
……像目前其他種機器學習模型在做的一樣。而且拉菲克在做的另一件事,基本上就是非常、非常努力跟那個機器學習模型共同合作,來創造出你看到的這些形狀。你可能只會得到雜訊。要是你對自己說:「這裡有一個生成式人工智慧。我要直接把這所有資訊塞進去,」你可能會得到毫無意義、一團混亂,或在視覺上無法辨別為任何東西的玩意。所以說,對我而言,那就是你在做的事如此令人驚奇的地方。
CNN特派員 法瑞德.札卡利亞
你認為未來所有藝術家都會必須對人工智慧有某種程度的了解嗎?
媒體藝術家 拉菲克.阿納多爾
我是說,那是非常美好、一廂情願的想法,但我認為……我希望每個人都會對人工智慧是什麼有所了解,因為它是超越藝術的——這是為了人類。它是一項強大的技術,而我認為在人類歷史中,我們過去從來沒有具備這種想像力與智慧思考能力、這麼強大的東西。