[Music]

artificial intelligence

this term has been buzzwords these days

right

but what's on your mind when you heard

about this term digital intelligence

for me it's going to be like a super

highly intelligent

machine like wally

what is a what form of the robot that's

very smart

funny and also likes robot so i really

like alienware right

and if we are on the same generation our

childhood might be accompanied by

this robot it's a highly super

intelligent robot

robotic cat from the future doraemon yup

you heard the rhyme

if you think about it it's actually

super intelligent

robotics because it can

solve any kind of problems it can

communicate with the humans very well

it also has feelings so

it's one of the form that's a super

intelligent robot right

so to be honest when i was a childhood i

was

i dream of having a door i want by my

side yeah

and another thing about ai could be

jarvis

it's the ai that helps tony stark build

iron man right it also manages house

and another ai is also maybe if you

think

from another science fiction it's like

terminator the human killing machines

a network that they try uh try to kill

the whole humanities

is this kind of so what are the same

things between all of these robots

all of them are highly intelligent ai

right they have specific feelings

they are general uh overall they are

very smart

but what about our current state of ai

in the world

if i quote from john kuhn he's a

facebook ai director

and he's one of the uh one of from the

three

publishers of the deep learning the

godfather of deep learning together with

joseo bengio

and joffrey hinson he said that we're

very far from building truly intelligent

machines

and in fact these experts say that our

current ai

is not even as smart as a rat yup you

heard it right

not even a smart ass thread because the

current ai can only do very specific

things

let me give you one example right so

imagine that we already have a tons

of data of pictures right so here i have

pictures that label as airplane

automobile bird cat etc and i wanted to

build a neural network model

so it is one form of of ai

i wanted to build an image

classification so i trained this model

basically to understand what kind of

picture is it

right so from all of this model i fit

into this neural network

and i get a model trained and if i give

a new picture

to this neural network model it will try

to classify

what kind of picture it is okay it

predicts tracks

perfect right but what if i give it

another picture

a handwritten image picture can it still

detect what kind of vector is this no

because

we train this model using a truck data

card data

births etc but not a handwritten image

so this neural network cannot predict

something that they are not trained with

right

so back again our current

eyes can only do very specific thing

very specific thing that's first

and it's only as good as what kind of

data you fit it into

but given of this specifics ai actually

has saved us

a lot of time what i mean by a lot it's

really actually a lot

even if i have to accumulate it's like a

thousand years

i'm not excited let me tell you in uh

after a bit yep this

receives it saves us time a lot

okay let's take a look back at this

image classification model right

uh what about this classification model

uh now i try to train a toddler i would

like to

train him or her uh basically to

classify what kind of image is this

imagine how long that you uh you will

take to train this toddler

to basically to classify whether this is

a airplane automobile

car trucks could it take weeks

maybe depends on the age right could it

take months

could be or maybe even years but the ai

can train them in just a very snap just

very

very instantly let me give one one

another example

so another example of ai is this one

a machine learning model called alphago

it was built by deepmind

in 2016 and in 2016

this ai able to beat the human world

champion of go

lizido and are you curious how he can

train to beat the

human world champion basically it learns

about a hundred thousand human games of

gold

amazing right and so if we have

if the one average of the human game of

go

is played uh maybe 90 minutes 90 minutes

to what two hours

so it's about 22 years worth of

experience of playing go

can you imagine in 22 years i'm able to

complete my

elementary high school elementary school

i go to junior high school

senior high school i complete my

bachelor degree

i even able to complete my master's

degree in that time

and yes so uh but

imagine can you guess how long does it

take for this

ai to trend these 22 years

in just three weeks it's amazing right

the 22 years worth of experience of

playing go

can be learned by this ai just in three

weeks imagine that

uh so it's until it able to uh

it's able to be the human world champion

amazing right

let me give you one another example an

example that also built by this deepmind

is open ai5 it tried to learn dota 2

about 10 000 years of experience just in

10 months

and it has been tested that this open i5

it has a winning rate about 99.4 percent

against the professional player of dota

it's so amazing though this is the thing

that ai can

really be have a superhuman performance

if they are given a very specific task

if they're given a very good amount of

data

okay but ai is actually not just playing

games

it's actually everywhere around us even

in your pocket

me give an example instagram our daily

apps it's newspapers

actually tailored to the preferences how

can how

how does it do that right because

it knows what kind of pictures that you

like it knows what kind of post that you

commented on

it knows how long do you stare at this

kind of picture

oh yes it knows how long you start at

one picture if you look

into that puppy picture for one minute

it knows if you look into it

watch this video about uh a cat get

swimming or maybe

poppy swimming it knows how long that

you stare at this video

so all of these are tailored they built

an ai

basically to make you more engaged into

this app right

so it's it's amazing that's making you

more engaged on this app

so have you if you if you're like me

have you ever felt that you just plan to

scroll instagram before you sleep

initially you try to uh just scrolling

for like 15 minutes

and then turns out it's been hours and

sometimes maybe

it's already morning and then did you

just scrolling for instagram

right so you're not even aware that

you have been that engaged to this app

because their ai has

makes you uh really engaged with

your feed because it's been tailored to

your preferences

but not just instagram let me give give

you one another example

from one of the right healing app from a

company

from a country is going so in 2010

when when project first launched it's

very manual right

when there is a customer need to order a

booking

they have to call the customer agent

right hey

i need uh i need someone to pick me up

at that's a certain location

okay and then the customer agents reply

and say okay let me

allocate one of the driver for you and

yeah

then the the driver can come to you pick

you up

and done and in 2016 actually

change uh they launched the first mobile

app right

so at that time they changed from the

human agent

into a more automated and nx

they're using ai to allocate drivers to

customers

so imagine how how much time do they

save even this

let's say there are millions of bookings

every day and for each booking

it's handled about 20 to 30 minutes by a

human agent

right and so if you in total it takes

every day takes like 50 years if there's

only one person who manages all of the

spooky

yeah so imagine that there's a team of

50 agents it's going to take them

the whole year just to allocate the

booking on that day

you're getting it see so ai has saved us

50 years every day so imagine this it

just

lasts for 20 days it's been thousand

years it's amazing right

another one example is uh they use ai

to make the platform safe for everyone

so they use the data basically to

look into the human behavior one of the

examples is

the human movement so you collect from

the gps data points

there's about billions of data points

every day right

so let me give you an example does it

make sense if

one day there is a user in command and

then on the next minute

suddenly it jumped to anchor just in one

minute they jump about 20 kilometers

away

it doesn't make sense right uh

definitely it's not

uh honest it's not an honest user uh

they may be trying to try

to uh scam or they're trying to do bad

things to

uh or the users on the platform and

so here we have to tackle all of these

kind of things right

but imagine that this kind of data

points have to be eyeballed by

human validator one by one can you

imagine how many years will it take

right and with an ai it actually saves

us about

19 years every day to protect millions

of users

it's amazing right so if you combine it

with

how ai has allocated drivers

it saved about 70 years one day for one

single day

so it's amazing right i i actually saved

us

uh more times than you can ever imagine

actually and

and not just saving time it's also saved

lives

so recently there is a news from mit

that

they tried to collect about 70 000 of

audio recordings

these every audio recordings have the

audio of people caffeine yeah

your hair correctly so caffeine so from

the 70 000 recordings

actually there was about 200 thousands

of

caffeine sampled audio from this data

they tried to

create an ai that is able to predict

whether someone

is uh coffee positive or not

and guess what their ai is even able to

predict what 98.5

whether someone has coveted or not just

by the cough ideology

it's amazing right so imagine that

including asymptomatic

so it's not just a symptom as a guy that

already has shown symptoms but also this

asymptomatic

people so imagine that this kind of ai

has been implemented in your device

imagine how uh it can detect the covet

this earlier than than we possibly think

right

so if if someone can be detected much

earlier they can be

make a self-isolation hence uh they

don't need

they can prevent to spread the disease

uh

more wider so it's pretty amazing

another example is the heart attack uh

detection so one company

in copenhagen called corti they analyzed

about 161 thousand

of emergency calls so those are the

audio the call to the ambulance hey i'm

actually for ambulance

and based on that call they're able to

create an ai

and reduce about 42 percent of

undetected out of hospital

cardiac arrest it's so amazing right it

already saved thousands of lives

but given all of the achievement by ai

so what do you think that we should

really be aware of do you think that we

still need to worry about the human

killing machines

that still in our image that's actually

it's on in our imaginary

no if i can tell you one thing it should

be something like this

yep uh you look it correctly it's a

social media news feed

like what i said before on instagram

social media our social media nisbet

is actually telling your preference no

matter what

if it's a fun things like you like

puppies or not okay so maybe your

newsfeed

filled with puppies i like to scuba dive

so my newsfeed will be filled with

scuba diving pictures so it's it's

pretty awesome

but imagine that that is something

sensitive like

politics or something else

it's quite more sensitive and

if i look into the research in 2014 by

peer research center

we are actually today we are more

divided than ever than what happened in

the past

so this look let's take a look at this

chart right

uh in 2014 there is a graph about the

uh they surveyed about the democrats

supporters and also republican

supporters

and somehow compared to the 10 years ago

or 20 20 years ago

it's become more polarized so

we're more polarized on this kind of

views because

someone who supports democrat on their

newsfeed will be

they will see many things about the

democrats

and vice versa and another survey told

us that

these people uh these supporters in

democrat

uh many more people that think badly

about people that support republican

and and vice versa as well can you

imagine that

if uh if you do you remember that what

happens in

indonesia political situation a couple

of uh a couple of months couple of years

back

it's quite similar isn't it if you even

imagine that

someone who supports candidate a they

think very badly about

people who support candidate b and vice

versa

so let me ask you one thing

who whoever watched this that

has left the our chat group

of families or long friends because have

a different views of this

right can you imagine so actually where

today

is way more polarized than ever we're

divided more than ever

because because of our social media is

territory our preferences

if i support candidate a my newspaper

will be filled with

news from candidate a and i will think

that all supports from negative b is

very bad they're wrong right and vice

versa

that makes us uh more divided

and so my message to you is

if you look into your news feed in our

social media or in

any news you might want to want to think

or want to double check

hey is this correct does it does it

makes sense

so uh you cannot think that

your site is the one that is true you

have to think that

you could be wrong right or maybe the

other side couldn't be right

or vice versa so this kind of things

that uh

hopefully you can be more open-minded

that everything that you see in your

social media

is actually tailored by ai according to

your preferences

so please be careful please be

open-minded

about whatever you see on the social

media and

there is one thing that ai still don't

have we just miss them

lastly if i have to quote from stephen

hawking he said that

we all have potential to boost our

boundaries through technology

and to think big even we are now at a

very exciting times and place

because we see all of this kind of power

right

and actually our generation are the

future are the pioneers of that

technologies like what i mentioned

before the neural network

uh the system to detect uh

heart attack is them to predict a covet

everything is

a system that is able to

beat the human world champion it's all

made in our generation

it's so exciting right but there's one

thing if we see

our future our features actually are

raised between the growing power of this

technology

and the wisdom with which we use it

let us all hope that the wisdom wins in

this race

thank you very much

{{}}

[音乐]

人工智能

这个词最近一直是流行语，

但是当你听到这个词时，你在想

什么 对我来说，数字智能就像一个超级

智能

机器，像沃利

什么是机器人的什么形式？

聪明

有趣，也喜欢机器人，所以我真的很

喜欢外星人

，如果我们是同一代人，我们的

童年可能会伴随着

这个机器人，它是

来自未来哆啦A梦的高度超级智能的机器人机器猫是的

，

如果你仔细想想，你听到了押韵 它实际上是

超级智能

机器人，因为它可以

解决任何类型的问题，它可以

很好地与人类交流

它也有感觉，所以

它是超级智能机器人的一种形式，

所以说实话，当我小时候，我

是

我的梦想 我想要一扇门

是的

房子和另一个人工智能也可能如果你从另一个科幻小说中想到它就像终结者人类杀人机器他们尝试的网络呃试图杀死整个人文学科就是这样的那么所有这些机器人之间有什么相同的东西所有他们是非常聪明的人工智能，他们有特定的感受，他们很一般，嗯，总的来说，他们非常聪明，但是如果我引用约翰库恩的话，那么我们目前在世界上的人工智能状态呢？他是 Facebook 人工智能总监，他是来自深度学习的三位出版商深度学习的教父与 joseo bengio 和 joffrey hinson 他说我们离制造真正的智能机器还很远，事实上这些专家说我们现在的人工智能甚至不如老鼠聪明是的，你没听错，甚至不是一个自作聪明线程，因为当前的人工智能只能做非常具体的事情让我给你一个例子吧，想象一下我们已经有很多

正确的图片数据，所以这里我

有标签为飞机

汽车鸟猫等的图片，我想

建立一个神经网络模型，

所以它是 AI 的一种形式，

我想建立一个图像

分类，所以我训练这个模型

基本上是为了理解 什么样的

图片是

正确的，所以从所有这个模型中，我

适合这个神经网络

，我得到一个训练好的模型，如果我

给这个神经网络模型一个新的图片，它会

尝试分类

什么样的图片是好的

预测轨道

完美正确但是如果我给它

另一张

图片手写图像图片它仍然可以

检测到什么样的向量这是不，

因为

我们使用卡车数据

卡数据

出生等而不是手写图像训练这个模型

所以这个神经网络不能 预测

一些他们没有接受过正确训练的东西，

所以我们现在的

眼睛只能做非常具体的事情，

非常具体的事情，这是第一件事

，它只取决于你适合什么样的

数据

但是考虑到这个细节，人工智能实际上

为我们节省

了很多时间，我的意思是很多，

实际上实际上很多，

即使我必须积累它就像一

千年，

我并不兴奋，让我在 uh 之后告诉你

是的，这个

接收它为我们节省了很多时间，

好吧，让我们回顾一下这个

图像分类模型吧，

呃，这个分类模型怎么样，

呃，现在我试着训练一个蹒跚学步的孩子，我想

训练他或她，呃，基本上是为了

分类什么类型 图片是这个

想象一下，你，呃，你要花多长时间

来训练这个蹒跚学步的孩子

，基本上来分类这是否

是飞机，汽车，

汽车，卡车，可能需要几周，

可能取决于年龄，可能

需要几个月

，甚至几年，但是 人工智能

可以在很短的时间内训练它们 非常非常快速

让我

再举一个例子

所以另一个人工智能的例子是这个 一个

叫做 alphago 的机器学习模型

它是由 deepmind

在 2016 年和 2016 年建立

的 ai 能够击败 go lizido 的人类世界

冠军

，你是否好奇他如何

训练以击败

人类世界冠军基本上它了解了

大约十万人类游戏的

黄金

惊人的权利，所以如果我们有

一个人类的平均水平 下

围棋

呃 大概是 90 分钟 90 分钟

到 2 小时

左右 所以这大约是 22 年

的围棋经验

你能想象在 22 年内我能够

完成我的

小学 小学

我去初中

高中 我完成了我的

学士学位

我什至能够在那个时候完成我的硕士学位

，是的，但是

想象一下，你能猜到

这个

人工智能需要多长时间才能在短短三周内成为这 22 年的趋势，

这真是太棒

了 22

这个AI可以在三

周内

学会多年的围棋经验

举一个同样由这个 deepmind 构建的例子

是 open ai5 它试图在 10 个月内学习 dota 2

大约 10 000 年的经验

，

并且经过测试，这个 open i5

它对职业玩家的胜率约为 99.4%

dota 太神奇了，虽然这

就是人工智能

真的可以拥有超人的表现，

如果他们被赋予非常具体的任务，

如果他们得到大量的

数据，

好吧，但人工智能实际上不仅仅是玩

游戏

它实际上无处不在 我们甚至

在你的口袋里

我举个例子 instagram 我们的日常

应用 它是报纸

实际上是根据偏好量身定制的

它如何做到这一点 因为

它知道你喜欢什么样的图片

它知道你评论过什么样的帖子

知道你会盯着

这种照片

多久哦是的它知道你从一张照片开始看多长时间

如果你

看着那张小狗的照片一分钟

它知道你是否

看着它 一个关于猫

游泳或者

罂粟游泳的视频 它知道

你会盯着这个视频多久，

所以所有这些都是量身定制

的 参与了这个应用程序，

所以如果你像我一样，

你有没有觉得你只是打算

在睡觉前滚动 Instagram

最初你尝试呃只是

滚动 15 分钟

，然后结果是几个小时，

有时可能

现在已经是早上了，然后您是否

刚刚滚动浏览 Instagram，

所以您甚至不知道

您对这个应用程序的参与度很高，

因为他们的人工智能

让您真正参与到

您的提要中，因为它是根据您的喜好量身定制的，

但不仅仅是 instagram 让我

再举一个例子

，来自一个国家的一家公司的正确治疗应用程序之一，

所以在 2010 年

项目第一次启动时，它是

非常手动的，

当有 客户需要

预订，

他们必须打电话给客户代理，

嘿，

我需要，嗯，我需要有人在某个地点接我，

好的，然后客户代理

回复说好的，让我

为您分配一名司机

是的，

然后司机可以来接你

并完成，并且在

2016 年实际上发生

了变化 将司机分配给

客户，

所以想象一下，即使这样，他们也能节省多少时间，

假设每天有数百万个预订

，每个

预订由人工代理处理大约 20 到 30 分钟

，所以如果你总共需要

每天 如果

只有一个人管理所有

令人毛骨悚然的事情，大概需要 50 年，是的，所以想象一下，有一个由

50 名代理商组成的团队，他们将花费

一整年的时间来分配

预订那天

你看到的 所以人工智能每天为我们节省了

50 年，所以想象一下，它只

持续了 20 天，已经有几千年

了 人类行为

示例之一

是人类运动，因此您

从 gps 数据点收集

每天大约有数十亿个数据点，

所以让我给您举个例子，

如果

有一天有一个用户在指挥

然后继续，这是否有意义 下一分钟，

它突然跳到锚点，一

分钟后他们跳了大约 20

公里，

这没有任何意义，嗯，

绝对不是，

诚实，这不是一个诚实的用户

试图对

呃或平台上的用户做坏事，

所以在这里我们必须正确处理所有这些

事情，

但想象一下这种数据

点必须由

人工验证者一个一个地观察，你

能想象吗？ 需要多少年才能

正确，而有了人工智能，它实际上

每天为我们节省大约 19 年的时间来保护数

百万用户，

这真是太棒了，所以如果你将它

与

人工智能如何分配驱动程序结合起来，

它一天节省了大约 70 年的

时间

所以这真是太棒了 ii 实际上拯救了

我们的

次数比你想象的要多

，

而且不仅仅是节省时间，它还拯救了

生命

所以最近有一个来自 mit 的消息

，

他们试图收集大约 70 000 个

录音

这些每一个录音 有

人们咖啡因的音频是的，

你的头发是正确的所以咖啡因所以

从 70 000 条录音中

实际上有大约 20 万

个

咖啡因采样音频从这些数据中

他们试图

创建一个能够预测某人是否是咖啡因的人工智能

猜猜他们的人工智能甚至能够

预测 98.5

是否有人垂涎

于咳嗽意识形态，

这太棒了，所以想象一下，

包括 asym ptomatic

所以这不仅仅是一个

已经表现出症状的人的症状，而且还有这种

无症状的

人所以想象一下这种人工智能

已经在你的设备中实现了

想象一下它如何能够

比我们可能认为的更早地检测到这种渴望

所以 如果可以更早地检测到某人，

他们可以

进行自我隔离，因此，他们

不需要，

他们可以防止将疾病传播

得更广泛，所以这非常令人惊讶

另一个例子是心脏病发作，呃

检测所以

哥本哈根的一家公司 打电话给 corti，他们分析了

大约 161,000

个紧急呼叫，所以这些是

呼叫救护车的音频，嘿，我

实际上是

在呼叫救护车，基于该呼叫，他们能够

创建一个人工智能，

并减少约 42%

未被发现出院的人

心脏骤停，这太神奇了，它

已经挽救了数千人的生命，

但考虑到人工智能的所有成就，

你认为我们应该

真正意识到什么，你认为

我们是 直到需要

担心仍然在我们的形象中的人类杀人机器实际上

它在我们的想象中

没有如果我可以告诉你一件事它

应该是这样的

是的，你看对了它是一个

社交媒体新闻提要，

就像我一样 之前在 instagram 社交媒体上说过，

我们的社交媒体

nisbet 实际上是在告诉你的偏好，

不管它是不是很有趣，比如你喜欢

小狗还是不好，所以也许你的

新闻源

充满了

小狗 潜水照片，所以它

非常棒，

但想象一下，这是

敏感的事情，比如

政治或其他事情，

它会更加敏感，

如果我研究 2014 年同行研究中心的研究，

我们实际上

比以往任何时候都更加分裂

过去

所以这个样子让我们看一下这张

图表吧

呃在 2014 年有一张关于

呃他们调查的关于民主党

支持者和代表的图表 ublican 的

支持者

，不知何故，与 10 年前

或 20 20 年前相比，

它变得更加两极分化，所以

我们对这种观点更加两极分化，

因为

在他们的新闻源上支持民主党的人

会看到很多关于

民主党

的事情 反之亦然，另一项调查告诉

我们，

这些人，嗯，这些民主党的支持者，

嗯，更多的人对支持共和党的人有不好的看法

，反之亦然，你

能想象，

如果你记得

在

印度尼西亚的政治局势中发生了什么

几个 呃 几个月 几年

前

很相似 不是吗 如果你甚至

想象

支持候选人 a 的

人对支持候选人 b 的人看法很差 反之亦然

所以让我问你一件事

谁 谁看了

这个离开了我们

的家庭或老朋友的聊天群，因为

对这个权利有不同的看法，

你能想象到哪里

今天

比以往任何时候都更加两极分化，我们比以往任何时候都

更加分裂，

因为因为我们的社交媒体是

我们的领地，

如果我支持候选人 a，我的报纸

将充满

来自候选人 a 的新闻，我

认为来自负面 b 的所有支持都是

非常糟糕，他们是对的，反之亦然

，这让我们更加分裂

，所以我给你的信息是，

如果你在我们的

社交媒体或

任何你可能想要思考

或想要仔细检查的新闻中查看你的新闻提要

嘿，这是正确的吗？它是否

有意义，

所以你不能认为

你的网站是真实的，你

必须认为

你可能是错的，或者

另一方不可能是对的

，反之亦然，所以这种

呃

希望你能更开放

的事情，你在

社交媒体

上看到的一切实际上都是由人工智能根据你的喜好量身定制的，

所以请小心，请

对你在社交媒体上看到的任何内容保持开放的态度

还有一件事，人工智能仍然

没有我们最后想念他们，

如果我不得不引用斯蒂芬

霍金的话，他说

我们都有潜力通过技术来扩大我们的

界限

，即使我们现在正处于一个

非常令人兴奋的时期 时间和地点，

因为我们正确地看到了所有这种力量

，实际上我们这一代人

是这些技术的先驱，

就像我

之前提到的神经网络，

呃，检测呃

心脏病发作的系统是他们预测一个觊觎的

一切都是

一个能够

击败人类世界冠军的系统，这一切

都是在我们这一代人创造

出来的，这太令人兴奋了

让我们都希望智慧在

这场比赛中获胜

非常感谢