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

翻译人员: Yip Yan Yeung 校对人员: Grace Man

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

我想先请你们想象一种

that you've never seen before.

从未见过的颜色。

Just for a second give this a try.

姑且来试一下吧。

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

你能想象出一种无法感知的颜色吗？

I never seem to get tired of trying this

我对此乐此不疲，

although I know it's not an easy challenge.

即使我知道这并不容易。

And the thing is,

问题是，

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

如果不基于我们自身的经历， 我们就无法想象出某些东西。

A color we haven't yet seen

我们从未见过的颜色，

outside the spectrum we can perceive

超出了我们可以感知的光谱，

is outside our ability to conjure up.

也就超出了我们的想象力。

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

这就感觉像是 我们的想象力有一个边界，

where all the colors we can imagine

我们可以想象出的所有颜色

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

只能是我们之前见过的 颜色的不同变体。

Yet we know for a fact

但是我们知道有这么一个事实，

that those color frequencies outside our visible spectrum are there.

超出可见光频谱的 颜色频率确实存在。

And scientists believe that there are species

科学家们相信有些物种

that have many more photo receptors

拥有更多的光感受器，

than just the three color ones we humans have.

比我们人类拥有的 三色光感受器要多。

Which, by the way,

顺带一提，

not all humans see the world in the same way.

不是所有人眼中的世界 都是一样的。

Some of us are colorblind to various degrees,

有些人会有不同程度的色盲，

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

我们经常会 就一些小事争论不休，

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

比如网上的一条裙子是 蓝黑还是白金。

But my favorite creature, one of my favorite creatures,

但是我最喜欢的生物， 最喜欢的生物之一，

is the peacock mantis shrimp,

是雀尾螳螂虾，

which is estimated to have 12 to 16 photo receptors.

估计有 12 至 16 个光感受器。

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

意味着它们眼中的世界 会更加色彩斑斓。

So what about artificial intelligence?

人工智能（AI）呢？

Can AI help us see beyond our human capabilities?

AI 可以帮助我们 超越人类的视觉吗？

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

我在过去的五年里 一直在用 AI 工作，

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

根据我的经验，它能看到的内容 取决于输入的数据。

But then you might be wondering, OK,

但是你可能会想，好吧，

if AI can't help imagine anything new,

如果 AI 不能想象出新东西，

why would an artist see any point in using it?

艺术家为什么要用它呢？

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

我的答案是，我认为 AI 可以增强我们的创造力，

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

通过组合既有的元素，创造新元素， 它就产生了价值。

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

基于我们的经历形成的 想象力的边界

is the place that I have been exploring.

是我探索的领域。

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

对我来说，一切始于 水族馆屏幕上的水母，

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

我戴着那种老式 3D 眼镜， 但愿你还有印象，

the ones with the blue and red lens.

镜片一片蓝一片红的那种。

And this experience made me want to recreate their textures.

这个体验让我想重现它们的质感。

But not just that,

但是不仅如此，

I also wanted to create new jellyfish

我还想创造我从未见过的

that I hadn't seen before, like these.

新型水母，就像这种。

And what started with jellyfish,

我从水母开始，

very quickly escalated to other sea creatures

迅速扩展到了其他海洋生物，

like sea anemone, coral and fish.

比如海葵、珊瑚和鱼类。

And then from there came amphibians, birds and insects.

然后是两栖动物、鸟类和昆虫。

And this became a series called “Neural Zoo”.

它们组成了一个系列，名为 《神经动物园》（Neural Zoo）。

But when you look closely, what do you see?

但是如果你仔细看看， 你会看见什么呢？

There's no single creature in these images.

这些图片里的都不是单一生物。

And AI augments my creative process

AI 增强了我发挥创意的过程，

by allowing me to distill and recombine textures.

让我可以提取、 重组它们的质感。

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

如果我要手绘， 可能要花上好几个月。

Plus I'm actually terrible at drawing.

而且我画画水平太差了。

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

你可以这么说， 从某种角度来看，我在做的

is a contemporary version of something

是早在相机出现之前，

that humans have already been doing for a long time,

人类就一直在做的事，

even before cameras existed.

但是是它的现代版本。

In medieval times,

中世纪，

people went on expeditions,

人们踏上远征，

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

回来之后他们会向画师 描述他们的见闻。

to an illustrator.

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

画师从未见过描绘之物，

would end up drawing

于是他们会按照

based on the creatures that they had previously seen

他们之前见过的生物绘制，

and in the process creating hybrid animals of some sort.

再自创一些合成动物。

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

也许出行者描述了一只海狸， 但是画师从没见过海狸，

the illustrator might give it the head of a rodent,

于是画了啮齿动物的头部，

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

狗的身子和类似鱼类的尾巴。

In the series “Artificial Natural History”,

我在《人工自然历史》 （Artificial Natural History）系列中

I took thousands of illustrations from a natural history archives,

从自然历史档案中 提取了几千张插图，

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

输入神经网络， 产生新版本的插图。

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

但是时至今日，我的所有作品 都是以 2D 的形式完成的。

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

在我的工作室伙伴菲力肯·麦考密克 （Feileacan McCormick）的帮助下，

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

我们打算基于经 3D 扫描的 甲虫的数据集训练一个神经网络。

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

但是我要提醒你的是 我们的第一批结果非常模糊，

and they looked like the blobs you see here.

如图所示的一坨。

And this could be due to many reasons,

导致这种情况的原因可能有很多，

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

但是其中一个原因是公开可用的 3D 昆虫数据集比较有限。

of openly available data sets of 3D insects.

And also we were repurposing

而且我们也改变了神经网络的功能，

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

由常见的生成图片 转向了生成 3D 结果。

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

无论如何，它们都是 令我们激动的一坨物体。

But with time and some very hacky solutions

随着时间的推移和 一些另辟蹊径的产品的出现，

like data augmentation,

如数据增强，

where we threw in ants and other beetle-like insects

如果我们输入蚂蚁 或者其他类似甲虫的昆虫

to enhance the data set,

增强了数据集，

we ended up getting this,

我们会得到这样的结果，

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

有人说它们长得像烤鸡。

(Laughter)

（笑声）

But hungry for more, we pushed our technique,

但是我们的野心不止于此， 我们改进了技术，

and eventually they ended up looking like this.

最终得到了这样的结果。

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

我们用了一项叫做 “3D 风格迁移”的技术

把纹理附着到输出结果上，

and we also trained a natural language model

我们还训练了一个自然语言模型，

to generate scientific-like names

生成科研风的名字和

and anatomical descriptions.

生物结构描述。

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

我们最后甚至找到了一个 网络体系结构处理 3D 网格。

So they ended up looking like this.

最后的结果是这样的。

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

对我们来说，这已经类似于 一种推测性研究……

(Applause)

（掌声）

A speculative study of creatures that never existed,

关于不存在的生物的 推测性研究，

kind of like a speculative biology.

类似猜想生物。

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

但是我并不想谈论 AI 和它的潜力，

unless it brought me closer to a real species.

除非 AI 产生的结果 接近真实存在的物种。

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

图上的两种生物，你觉得哪个 更容易在网上找到数据？

(Laughter)

（笑声）

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

你猜的没错，小熊猫。

And this maybe could be due to many reasons,

可能会有很多原因，

but one of them being how cute they are,

但是其中一个原因就是 它们太可爱了，

which means we photograph and talk about them a lot,

所以我们总是会给它们拍照， 讨论它们，

unlike the boreal felt lichen.

北方毡状地衣就没这种待遇了。

But both of them are classified as endangered.

但是这两种生物 都被定为了频危物种，

So I wanted to bring visibility to other endangered species

我想请大家都去关注一下 其他的濒危物种，

that don't get the same amount of digital representation

它们不一定会像 毛茸茸的可爱小熊猫那样

as a cute, fluffy red panda.

拥有那么高的数字化曝光度。

And to do this,

为了达成这个目标，

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

我们用自然界的百万张图片 训练了 AI，

and then we prompted with text

然后我们用文字提示，

to generate some of these creatures.

生成一些生物。

So when prompted with a text,

如果我们输入这样的提示：

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

“蓝宝石华丽雨林—— 一种极度濒危的蜘蛛的图片”

and its scientific name,

和它的学名，

our model generated this.

我们的模型会输出这样的结果。

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

这是一张真实的 蓝宝石华丽雨林的照片，

which is a wonderful spider endemic to India.

它是一种分布于 印度本土的华丽蜘蛛。

But when prompted with a text

如果我们输入这样的提示：

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

“红树林雀—— 一种极度濒危的鸟类的图片”，

our model generated this.

我们的模型会生成这样的结果。

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

这是真实的红树林雀的照片。

Both these creatures exist in the wild,

这两种动物都在自然界中真实存在，

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

但是生成图片的准确度 完全取决于可用的数据。

These chimeras of our everyday data

由我们日常数据产生的“怪物”

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

对我来说是对未来的另一种畅想。

Not in a literal sense, perhaps,

也许不是字面意思，

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

但是通过扩展我们对

我们所处生态环境的想象，

about the ecosystems we are a part of,

we might just be better equipped to recognize new opportunities

我们也许更有可能 会发现新机会和新潜力。

and potential.

Knowing that there's a boundary to our imagination

清楚地知道 我们的想象力是有界限的，

doesn't have to feel limiting.

并不代表我们得束手束脚。

On the contrary,

相反，

it can help motivate us to expand that boundary further

这可以鼓励我们拓展边界，

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

寻找我们没有见过的颜色和事物，

and perhaps enrich our imagination as a result.

也许最终会丰富我们的想象。

So thank you.

谢谢。

(Applause)

（掌声）