AI-Generated Creatures That Stretch the Boundaries of Imagination | Sofia Crespo | TED

譯者: Chun Ju Wang 審譯者: Shelley Tsang 曾雯海

I'd like to start by asking you to imagine a color

首先我想請你們想像一個顏色

that you've never seen before.

一個你從未見過的顏色

Just for a second give this a try.

稍微花點時間試試看

Can you actually visualize a color that you've never been able to perceive?

你真的能在腦海裡 看到從未見過的顏色嗎？

I never seem to get tired of trying this

我對此總是樂此不疲

although I know it's not an easy challenge.

雖然我知道這並不是個簡單的挑戰

And the thing is,

而且事實是

we can't imagine something without drawing upon our experiences.

我們的想像空間完全得靠經歷塑造

A color we haven't yet seen

某個我們沒有見過的顏色

outside the spectrum we can perceive

不但超越了我們能接收到的光譜

is outside our ability to conjure up.

也超越了我們憑空造物的能力

It's almost like there's a boundary to our imagination

我們的想像力似乎有個界限

where all the colors we can imagine

在這之中，我們能想像出的所有顏色

can only be various shades of other colors we have previously seen.

只會是已知顏色的各種深淺

Yet we know for a fact

但是我們明確知道

that those color frequencies outside our visible spectrum are there.

那些在人類可見光譜之外的 顏色頻率的確存在

And scientists believe that there are species

且科學家相信世上存在某些物種

that have many more photo receptors

牠們有更多的光感受器

than just the three color ones we humans have.

超越人類僅能接收到的三原色

Which, by the way,

順帶一提

not all humans see the world in the same way.

人類眼中並非只有一種世界

Some of us are colorblind to various degrees,

我們有些人是色盲，程度各異

and very often we don't even agree on small things,

且我們經常連小事都無法達成一致

like if a dress on the internet is blue and black or white and gold.

好比某件網路上的洋裝到底是 黑藍相間還是白金相間

But my favorite creature, one of my favorite creatures,

但我最喜歡的生物 應該說我最喜歡的生物之一

is the peacock mantis shrimp,

是雀尾螳螂蝦

which is estimated to have 12 to 16 photo receptors.

據估計有 12 到 16 個光感受器

And that indicates the world to them might look so much more colorful.

這也代表牠們眼中的世界 可能比我們的更五彩斑斕

So what about artificial intelligence?

那人工智慧呢？

Can AI help us see beyond our human capabilities?

它能幫我們獲取 超越人類極限的視野嗎？

Well, I've been working with AI for the past five years,

這個嘛，我已經與人工智慧共事五年了

and in my experience, it can see within the data it gets fed.

根據經驗，人工智慧的視野 侷限在我們提供給它的資料中

But then you might be wondering, OK,

那麼你可能會想

if AI can't help imagine anything new,

如果人工智慧想不出任何新東西

why would an artist see any point in using it?

藝術家用它到底是為了什麼？

And my answer to that is because I think that it can help augment our creativity

對我來說，答案是 這能幫助我們增強創造力

as there's value in creating combinations of known elements to form new ones.

因為利用已知元素創造不同組合 來形成未知組合確有其價值

And this boundary of what we can imagine based on what we have experienced

而我們根據自身經驗架構出的想像空間

is the place that I have been exploring.

其界限正是我在探索的邊界

For me, it started with jellyfish on a screen at an aquarium

這一切要從我在水族館的 影廳銀幕上看到的水母開始

and wearing those old 3D glasses, which I hope you remember,

我當時戴著舊式立體眼鏡—— 希望你們對那些

the ones with the blue and red lens.

紅色和藍色的鏡片還有印象

And this experience made me want to recreate their textures.

這個經驗讓我想要重現牠們的質地

But not just that,

不僅如此

I also wanted to create new jellyfish

我也想要創造新的水母

that I hadn't seen before, like these.

那些我從未見過的品種 好比畫面上這些

And what started with jellyfish,

而這些源於水母的想法

very quickly escalated to other sea creatures

很快便延伸到其牠海洋生物上

like sea anemone, coral and fish.

好比海葵、珊瑚和魚類

And then from there came amphibians, birds and insects.

接著蔓延到兩棲動物、鳥類和昆蟲

And this became a series called “Neural Zoo”.

然後演變成一個系列 我們稱它為「類神經動物園」

But when you look closely, what do you see?

但湊近仔細瞧瞧 你看到了什麼？

There's no single creature in these images.

這些圖片都包含不只一種生物

And AI augments my creative process

人工智慧強化了我的創作過程

by allowing me to distill and recombine textures.

讓我能將不同特色精煉再重組

And that's something that would otherwise take me months to draw by hand.

這個過程若純靠手繪將耗時數月

Plus I'm actually terrible at drawing.

尤其是我的繪畫功力實在很差

So you could say, in a way, what I'm doing

所以在某種程度上可以說

is a contemporary version of something

我只是用現代版的方式

that humans have already been doing for a long time,

來重現人們即使在相機出現以前

even before cameras existed.

便一直在做的事

In medieval times,

在中世紀

people went on expeditions,

許多人踏上征途

and when they came back they would share about what they saw

並在回來之後將所見所聞

to an illustrator.

告訴某位插畫家

And the illustrator, having never seen what was being described,

然後這位插畫家 在完全沒見過所述之物的情況下

would end up drawing

會根據他們

based on the creatures that they had previously seen

之前見過的生物來繪圖

and in the process creating hybrid animals of some sort.

並在這個過程中 或多或少地創造出混合物種

So an explorer might describe a beaver, but having never seen one,

所以，探險家形容的可能是一隻河狸 而插畫家在沒見過這種生物的情況下

the illustrator might give it the head of a rodent,

可能會畫出嚙齒動物的頭

the body of a dog and a fish-like tail.

狗的身體和類似魚的尾巴

In the series “Artificial Natural History”,

在「人造自然歷史」這一系列中

I took thousands of illustrations from a natural history archives,

我從一個自然歷史檔案庫中 擷取了上千張插圖

and I fed them to a neural network to generate new versions of them.

將它們餵給類神經網路 以產生新的版本

But up until now, all my work was done in 2D.

但截至目前為止 我所有的作品都是平面的

And with the help of my studio partner, Feileacan McCormick,

憑藉我工作室的夥伴 費利根．麥柯米克的幫助

we decided to train a neural network on a data set of 3D scanned beetles.

我們決定用一組經過三維掃描的 甲蟲資料集來訓練某個類神經網路

But I must warn you that our first results were extremely blurry,

醜話先說在前面 我們得出的第一個結果非常模糊

and they looked like the blobs you see here.

它們看起來就像這些團狀物

And this could be due to many reasons,

造成這個結果的可能有很多種

but one of them being that there aren't really a lot

但其中一個原因 是現存的資料集中

of openly available data sets of 3D insects.

存有三維昆蟲且開放使用的並不多

And also we were repurposing

我們同時也在重新定向

a neural network that normally gets used to generate images to generate 3D.

試圖讓已經習慣產生平面圖的 類神經網路開始產生三維影像

So believe it or not, these are very exciting blobs to us.

所以無論你相信與否 這些團狀物在我們眼中可是寶貝

But with time and some very hacky solutions

歷經時間和一些鬆散的解決方案

like data augmentation,

好比資料增強

where we threw in ants and other beetle-like insects

也就是加進螞蟻和其他類似甲蟲的昆蟲

to enhance the data set,

來增加資料集的內容

we ended up getting this,

我們最終得到了這個東西

which we've been told they look like grilled chicken.

我知道，看起來很像烤雞吧

(Laughter)

（笑聲）

But hungry for more, we pushed our technique,

我們不想止步於此 於是繼續精進

and eventually they ended up looking like this.

而最終的結果看起來像這樣

We use something called 3D style transfer to map textures onto them,

我們用「三維風格轉換」這個技術 來繪製它們的質地

and we also trained a natural language model

我們也訓練自然語言模型

to generate scientific-like names

來產生類似學名的名字

and anatomical descriptions.

和內在結構描述

And eventually we even found a network architecture that could handle 3D meshes.

最後我們甚至找到一個 能夠處理三維網格的網路架構

So they ended up looking like this.

所以最後的成品就像這樣

And for us, this became a way of creating kind of a speculative study --

對我們來說這成為了 創造某種推想研究的方式

(Applause)

（掌聲）

A speculative study of creatures that never existed,

一個與前所未見的生物有關的推想研究

kind of like a speculative biology.

有點像是推想生物學

But I didn't want to talk about AI and its potential

但我並不想空聊 人工智慧和它的潛能

unless it brought me closer to a real species.

它需要更貼近真實物種

Which of these do you think is easier to find data about online?

你們覺得哪一個 更容易在網路上找到資料？

(Laughter)

（笑聲）

Yeah, well, as you guessed correctly, the red panda.

沒錯，不出你們所料 答案是小熊貓

And this maybe could be due to many reasons,

這可歸結於很多原因

but one of them being how cute they are,

而其中一個原因是牠們真的很可愛

which means we photograph and talk about them a lot,

這代表我們喜歡拍牠們的照片 並常常討論這個物種

unlike the boreal felt lichen.

而北方氈狀地衣就沒有這種福氣

But both of them are classified as endangered.

但實際上兩者都是瀕危物種

So I wanted to bring visibility to other endangered species

所以我想提升其他瀕危物種的能見度

that don't get the same amount of digital representation

讓在數位世界中，曝光度 無法和可愛又毛茸茸的小熊貓相比的

as a cute, fluffy red panda.

那些生物也能被看見

And to do this,

要做到這一點

we trained an AI on millions of images of the natural world,

我們訓練人工智慧瀏覽 數百萬張自然世界的圖片

and then we prompted with text

然後我們給人工智慧提詞

to generate some of these creatures.

來產生其中幾種生物

So when prompted with a text,

所以當我們的文字敘述是

"an image of a critically endangered spider, the peacock tarantula"

「一張極危蜘蛛『孔雀蜘蛛』的圖片」

and its scientific name,

及牠的學名

our model generated this.

我們的模型產生了這個結果

And here's an image of the real peacock tarantula,

而這是孔雀蜘蛛的真實模樣

which is a wonderful spider endemic to India.

這是一種印度獨有的絕妙蜘蛛

But when prompted with a text

當我們換個文字敘述

"an image of a critically endangered bird, the mangrove finch,"

「一張極危鳥類『紅樹林雀』的圖片」

our model generated this.

這是模型產生的成品

And here's a photo of the real mangrove finch.

而這是紅樹林雀的真實樣貌

Both these creatures exist in the wild,

這兩個物種都生活在野外

but the accuracy of each generated image is fully dependent on the data available.

但最終成像的精確度 完全取決於能夠得到多少資料

These chimeras of our everyday data

這些從日常資料中產生的嵌合體

to me are a different way of how the future could be.

對我來說是一種 預示未來可能的不同方式

Not in a literal sense, perhaps,

或許不是那麼直觀

but in the sense that through practicing the expanding of our own imagination

但轉念想想，經由練習擴張我們

about the ecosystems we are a part of,

對所處生態系統的想像力

we might just be better equipped to recognize new opportunities

我們或許就能提升自己 在察覺新的機運和潛能

and potential.

這方面的能力

Knowing that there's a boundary to our imagination

我們的想像空間確有界限

doesn't have to feel limiting.

但不必因此感到受限

On the contrary,

相反地

it can help motivate us to expand that boundary further

這能激勵我們擴張自我界限

and to seek out colors and things we haven't yet seen

尋找那些我們從未見過的顏色和事物

and perhaps enrich our imagination as a result.

或許還能讓我們的想像力變得更加豐富

So thank you.

謝謝大家

(Applause)

（掌聲）