What Happens When Art Meets AI

[Music]

i

hate math

i know that might sound strange and

maybe even a little bit controversial

considering that i am employed

at the faculty of mathematics and

natural sciences

and that i’m here to talk to you guys

about artificial intelligence

but it’s true

i am horrible at math

yet i have found a way of dealing with

this which is part of what i want to

share with you all today

now i should perhaps specify

i don’t hate all math

i hate arithmetic

addition subtraction multiplication and

division

growing up i spent countless hours

crying in my mother’s lap as she

attempted to bribe and console me

through my recitation of the

multiplication tables

instead

i was your typical

artsy kid

i loved singing painting dancing

but i also loved science

and logic puzzles i was curious about

life and consciousness and how that all

came about

now what i found was that more often

than not

we tend to think of the world of art and

the world of science

as inherently

separate

opposites even

and for a long time i felt

trapped in the middle

with my math homework standing like a

gatekeeper

making sure that i would never be able

to achieve my goal of working somewhere

in this cross-section

between art and science

so how did i end up here

talking to all of you about art and

artificial intelligence

well

i learned how to code

computers unlike me

are really good at arithmetic

in fact they really only deal in numbers

numbers are their whole world

and for me

programming was like solving little

logic puzzles where the prize was that

my computer would do all the

calculations that i couldn’t

and i loved it

finally i had an accomplice

an aid

a friend even

that would do for me the tasks that i’m

not good at

and allow me to focus on the questions

that i was really interested in

i found that me plus my computer friend

was greater than me alone

and allying myself with my computer

was truly empowering for me

it has been what has enabled me to work

in this fascinating space

between art and science

i was able to begin an academic career

in computer science working with

computational creativity

and now i get to work on all these

really really interesting questions

like can i teach my computer friend

to be

creative

what kind of computational mechanisms

would even support that kind of behavior

can it learn these tasks where

there isn’t a single correct answer or a

goal state to reach

these tasks that are

inherently

subjective

for example

can i teach my computer

to dance

here i’ve shown

my computer a bunch of examples

of dancers improvising in these

beautiful motion capture suits which are

fixed with reflective

markers now as i said my computer only

deals in numbers so it doesn’t get to

see these beautiful polyester suits

all it sees are these giant matrices

and here each cell

contains a number

which represents one of these reflective

markers as it moves through time

it doesn’t know anything else

like the kinematics of human movement

or our body’s natural limitations

all it knows is what it can extract

from this data

and this

is what it comes up with

a little ai dancer

pretty good huh

i mean

it looks

human

you might even say it looks like dancing

but to be honest

it doesn’t always look like this

at first

all it was able to learn was to wiggle

its hands and feet

a lot

and if you think about it this kind of

makes sense

the dancers that it has seen they move

their hands and feet a lot too

not like this

not at this insane speed

but it’s not a completely unreasonable

first step

now this guy was a breakthrough for me

i i know that might seem a little sad

considering that this still looks uh

pretty crazy

but after only getting wiggly guy for

ages

i was

over the moon when i saw this

shapeshifter

smooth movements

kind of beautiful in a way no

of course it’s not very human-like to be

able to grow and shrink at will

but at least it was

learning something different

the method that i’m using here is called

supervised deep learning

i build an artificial neural network

which has recurrent connections

this allows it to hold a sort of

short-term memory

so that it can remember the things that

it has seen

and use that to make predictions about

the future

so by increasing the size of this

artificial neural net and showing it

even more examples

my computer friend was finally able to

generate ai dancers that move in a

smooth and realistic way

but my question

for you guys

is this

do you think

this looks

interesting

did you maybe think that wiggly guy or

the shapeshifter was more interesting to

look at

while this guy might be the most

realistic looking ai dancer

do you think it is creative

is it art

now some of you might say yes yes this

is art

and others will say no

this is one of the biggest challenges in

ai and art

what i would call the subjectivity

challenge

so let’s have another look at this task

that i’ve given my computer friend

i show it a sequence of movement

and then i ask it to predict where each

of these reflective markers will be at

the next frame

i measure its success at that task

by comparing its prediction

to the actual position that the dancer

moved to

so basically i’m saying

there is only one correct answer

what the dancer did

everything else is wrong

so i am

removing

some of this

subjectivity and that

will only ever encourage imitation

now

imitation is definitely part of our

artistic process

but by me never encouraging my computer

to do things differently might i be

forcing it into this mediocrity

now if i do encourage my computer to do

things differently we quickly end up

here

and yes

it’s definitely different

i would say it is uh chaos

so how do i define

this subjective sweet spot between

chaos

and mediocrity

how do i define what is good art

finding a precise metric for evaluating

a creative artifact

is a challenge that raises both

philosophical and technical questions

i mean for the most part we don’t even

agree on what is good art made by humans

and if we can’t agree on this then how

am i supposed to teach my computer what

art is

another

valid question here

is why

why should we teach computers to solve

creative tasks

when we outsource other things to ai it

doesn’t always go as we would hope

and ai is becoming a larger and larger

part of our everyday lives and with it

comes challenges relating to

discrimination alienation data ownership

and bias in our algorithms

and it is actually up to all of us

and especially those of us that work on

or with ai

to work on solutions for mitigating

these issues

still i am

cautiously optimistic

i believe that there is enormous

potential for ai to improve our lives

and our creative practices

but perhaps the solution is not for me

to

try and outsource everything to my

computer friend and have it do

everything on its own

perhaps instead i should teach it to

work with me

teach it to collaborate

take for example how we might use ai in

medical diagnosis

when you’re sick and you visit your

doctor

they can compare your list of symptoms

against hundreds of patients that

they’ve seen before

but if you use an ai

your symptoms could be compared to

millions of patients from around the

world in a matter of seconds

now i don’t imagine that many of you

would prefer the ai over the doctor

well maybe in the future

but for now

i think that the ai and the doctor

together has the potential of giving the

optimal care

and in a similar way ai can improve our

creative practices

by building advanced ai models for

generating creative data we could have

them function as suggestion givers

the ai dancers that my computer made for

example could be used by a dancer to

explore their movement repertoire

offering reflection stimulation

and

feedback

we already have a bunch of examples of

how ai and technology in general can

help improve and even democratize the

world of art

like assistive music software digital

drawing tools

and even something like advanced spell

checkers

but these tools

won’t be making any art without us

my computer was an enormous help for me

to get into science

i literally cannot do my job without it

and it seems that in the case of

artistic endeavors

the tables have turned on me and my

computer friend

here it might never be able to do its

job properly

without me

but i guess my point is

does that really matter

perhaps the solution to the subjectivity

challenge lies in this collaboration

in what we can achieve

together

now

i warned you

arithmetic is not my strong suit

but i think that in this case one plus

one is equal to three

because together my computer friend and

i

are greater than the sum of our parts

thank you

[Applause]

[Music]

you

[音乐]

讨厌数学,

我知道这听起来很奇怪,

甚至可能有点

争议,因为我受雇于

数学和

自然科学学院

,而且我是来和你们

讨论人工智能的,

但这是真的,

我 我的数学很糟糕,

但我找到了解决这个问题的方法,

这是我今天想与大家分享的内容的一部分,

现在我也许应该指定

我不讨厌所有数学

我讨厌算术

加减乘除

我长大了

我在妈妈的腿上哭了无数个小时,因为她

试图通过我背诵乘法表来贿赂和安慰我

而不是

我是典型的

艺术孩子

我喜欢唱歌绘画跳舞

但我也喜欢科学

和逻辑谜题我对

生活和意识很好奇 而这一切

是如何

发生的我现在发现的是,

我们往往倾向于认为

艺术世界和科学世界

本质上是

独立的

甚至在很长一段时间内,我都感到

被困在中间

,我的数学作业就像

守门人一样

,确保我永远

无法实现

在艺术和科学之间的这个交叉领域工作的目标,

所以我是怎么做到的 最后在这里

和大家谈论艺术和

人工智能

我学会了如何编写

计算机不像

我真的很擅长

算术事实上他们真的只处理数字

数字是他们的整个

世界对我来说

编程就像解决小

逻辑难题 奖励是

我的电脑可以完成所有我做不到的

计算

,我很喜欢它

最后我有一个同谋,

一个帮助,

一个朋友,即使

它可以为我完成我不擅长的任务

并让我专注 在

我真正感兴趣的问题上,

我发现我和我的电脑

朋友比我一个人更伟大,

并且将自己与我的电脑

结盟对我来说真正赋予

了我力量 d 我可以

在艺术和科学之间这个迷人的领域

工作

什么样的计算机

制甚至可以支持这种行为

它可以学习这些

没有单一正确答案或

目标状态的

任务来完成这些

本质上是

主观

的任务例如

我可以教我的计算机

这里跳舞吗? 我

给我的电脑看了

一堆舞者在这些

漂亮的动作捕捉套装中

即兴表演的例子

巨大的矩阵

,这里每个单元格都

包含一个数字

,代表这些反射标记之一,

因为它随着时间的推移而移动,

它一无所知 其他

像人类运动的运动学

或我们身体的自然

限制它所知道的就是它可以

从这些数据中提取

什么,这就是它提出的

一个小艾舞者

相当不错嗯

我的意思是

它看起来像

你甚至可以说它看起来 喜欢跳舞,

但老实说,

它一开始并不总是这样

他们的手和脚

也不是这样,

不是以这种疯狂的速度,

但这并不是一个完全不合理的

第一步,

现在这家伙对我来说是一个突破,我

知道

考虑到这看起来仍然

很疯狂,

但这似乎有点难过,但只有 多年来一直摇摆不定的家伙

当我看到这个

变形者

平滑的动作

在某种程度上很漂亮时,我欣喜若狂

,当然,能够随意

增长和收缩并不是很像人类,

但至少它是

le

我在这里使用的方法称为

监督深度学习

我建立了一个具有循环连接的人工神经网络

这使它能够保持一种

短期记忆,

以便它可以记住它所看到的事物

和 用它来

预测未来,

所以通过增加这个

人工神经网络的大小并向它展示

更多的例子,

我的电脑朋友终于能够

生成以流畅和逼真的方式移动的 AI 舞者,

但我

对你们的问题

是这个

你认为

这看起来很

有趣

吗 你是否认为那个摇摆不定的家伙

或变形者看起来更有趣

而这个家伙可能是最

逼真的 AI 舞者

你认为它有

创意吗 现在是艺术

你们中的一些人可能会说是 是 这

是艺术

,其他人会说不,

这是人工智能和艺术中最大的挑战之一

我称之为主观性

挑战,

所以让我们再看看这个

我给我的电脑朋友的任务

我给它展示了一系列运动

,然后我要求它预测每个反射标记中的每一个

在下一帧的位置

通过将其预测

与实际进行比较来衡量它在该任务中的成功 舞者

移动到的位置,

所以基本上我是说

只有一个正确答案

舞者所做的

一切都是错误的,

所以我正在

消除

一些这种

主观性,这

只会鼓励模仿

现在

模仿绝对是我们

艺术过程的一部分

但是我从不鼓励我的

电脑做不同的事情,

如果我鼓励我的电脑做

不同的事情,我现在可能会强迫它变得平庸,我们很快就会在这里结束

,是的,

这绝对是不同的,

我会说这是呃混乱

所以怎么办 我

在混乱和平庸之间定义了这个主观的甜蜜点

我如何定义什么是好的艺术

找到一个精确的指标来评估

一个创造性的人工制品

是一个挑战 在提出了

哲学和技术问题

我的意思是,在大多数情况下,我们甚至不

同意什么是人类创造的好的艺术

,如果我们不能就这一点达成一致,那么

我应该如何教我的计算机什么

另一种

有效的艺术 这里的问题

是,

当我们将其他事情外包给人工智能时,为什么我们要教计算机解决创造性任务它

并不总是像我们希望的那样发展

,人工智能正在

成为我们日常生活中越来越大的一部分,

随之而来的挑战是

我们算法中的歧视异化数据所有权和偏见

实际上取决于我们所有人

,尤其是我们这些从事

人工智能工作或使用人工智能的人

来解决

这些问题,

我仍然

谨慎乐观,

我相信有巨大的

潜力 人工智能来改善我们的生活

和我们的创造性实践,

但也许解决方案不是让我

尝试将所有内容外包给我的

电脑朋友,让它

自己做所有事情,

也许是 inst 我应该教它

与我一起工作

教它

协作 例如,

当您生病并且去看医生时,我们如何在医疗诊断中使用人工智能,

他们可以将您的症状列表

与他们看过的数百名患者进行比较

以前,

但是如果您使用人工智能,

您的症状可以在

几秒钟内与来自世界各地的数百万患者进行比较

现在我不认为你们中的许多人

可能会更喜欢人工智能而不是医生

但现在

我认为人工智能和医生

一起具有提供最佳护理的潜力,

并且以类似的方式,人工智能可以

通过构建用于

生成创意数据的高级人工智能模型来改进我们的创造性实践,我们可以让

它们作为建议提供者发挥作用

例如,舞者可以使用制造的计算机来

探索他们的动作曲目,

提供反射刺激

反馈

我们已经有了一堆

关于人工智能和技术如何在 g eneral 可以

帮助改善甚至使艺术世界民主化,

例如辅助音乐软件数字

绘图工具

,甚至像高级拼写

检查器

这样的工具,但

如果没有我们,这些工具将无法制作任何艺术

我的电脑对我进入科学领域有很大帮助

没有它真的无法完成我的工作

,似乎在艺术努力的情况下,

桌子已经转向我和我的

电脑

朋友,如果没有我,它可能永远无法正常完成它的

工作,

但我想我的意思

是这真的重要吗

也许主观性挑战的解决方案

在于这种合作

,我们可以一起实现

现在

我警告你

算术不是我的强项,

但我认为在这种情况下,一加

一等于三,

因为我和我的电脑朋友在一起

更大 比我们各部分的总和

谢谢你

[鼓掌]

[音乐]