The danger of AI is weirder than you think | Janelle Shane

Translator: Marsya Firina Reviewer: Ade Indarta

So, artificial intelligence

Kecerdasan buatan (AI)

is known for disrupting all kinds of industries.

dikenal mengganggu berbagai macam industri.

What about ice cream?

Bagaimana dengan es krim?

What kind of mind-blowing new flavors could we generate

Rasa baru mencengangkan seperti apa yang bisa kita ciptakan

with the power of an advanced artificial intelligence?

dengan kecanggihan kecerdasan buatan?

So I teamed up with a group of coders from Kealing Middle School

Saya bekerja sama dengan pembuat kode dari SMP Kealing

to find out the answer to this question.

untuk mencari jawaban pertanyaan ini.

They collected over 1,600 existing ice cream flavors,

Mereka mengumpulkan 1.600 lebih rasa es krim yang sudah ada,

and together, we fed them to an algorithm to see what it would generate.

lalu kami memasukkannya ke algoritme untuk melihat hasilnya.

And here are some of the flavors that the AI came up with.

Ini beberapa rasa yang diciptakan AI.

[Pumpkin Trash Break]

[Patahan Sampah Labu]

(Laughter)

(Tertawa)

[Peanut Butter Slime]

[Lendir Selai Kacang]

[Strawberry Cream Disease]

[Penyakit Es Krim Stroberi]

(Laughter)

(Tertawa)

These flavors are not delicious, as we might have hoped they would be.

Rasa-rasa ini tidak seenak harapan kami.

So the question is: What happened?

Pertanyaannya, apa yang terjadi?

What went wrong?

Apa yang salah?

Is the AI trying to kill us?

Apakah AI ingin membunuh kita?

Or is it trying to do what we asked, and there was a problem?

Atau hanya menuruti perintah, tapi ada masalah?

In movies, when something goes wrong with AI,

Dalam film, jika AI bermasalah,

it's usually because the AI has decided

itu biasanya karena kecerdasan buatan itu

that it doesn't want to obey the humans anymore,

tidak ingin patuh pada manusia lagi.

Dia memiliki tujuan sendiri, terima kasih.

and it's got its own goals, thank you very much.

In real life, though, the AI that we actually have

Namun di kehidupan nyata, AI yang kita miliki

is not nearly smart enough for that.

tidak cukup pintar untuk itu.

It has the approximate computing power

AI yang ada memiliki daya komputasi

of an earthworm,

seperti cacing tanah

or maybe at most a single honeybee,

atau setidaknya seperti lebah madu.

and actually, probably maybe less.

Mungkin kurang dari itu.

Like, we're constantly learning new things about brains

Kita terus mempelajari hal baru tentang otak

that make it clear how much our AIs don't measure up to real brains.

dan menunjukkan bahwa AI tak sebanding dengan otak kita.

So today's AI can do a task like identify a pedestrian in a picture,

AI saat ini bisa mengenali pejalan kaki dalam foto,

but it doesn't have a concept of what the pedestrian is

tapi tidak paham apa itu pejalan kaki

beyond that it's a collection of lines and textures and things.

selain bahwa itu hanya kumpulan garis, tekstur, dan lainnya.

It doesn't know what a human actually is.

AI tak tahu apa itu manusia.

So will today's AI do what we ask it to do?

Apakah AI saat ini akan menuruti perintah kita?

It will if it can,

Ya, jika bisa.

but it might not do what we actually want.

Namun, mungkin tak sesuai keinginan kita.

So let's say that you were trying to get an AI

Anggap Anda ingin kecerdasan buatan

to take this collection of robot parts

mengambil suku cadang robot,

and assemble them into some kind of robot to get from Point A to Point B.

menyusunnya menjadi robot, dan pindah dari Poin A ke Poin B.

Now, if you were going to try and solve this problem

Jika Anda ingin selesaikan masalah ini

dengan membuat program komputer biasa,

by writing a traditional-style computer program,

you would give the program step-by-step instructions

Anda memberi instruksi langkah demi langkah pada program

on how to take these parts,

untuk mengambil suku cadang,

how to assemble them into a robot with legs

menyusunnya menjadi robot berkaki,

and then how to use those legs to walk to Point B.

dan cara menggunakan kakinya untuk berjalan ke Poin B.

But when you're using AI to solve the problem,

Namun, jika memakai kecerdasan buatan,

it goes differently.

caranya berbeda.

You don't tell it how to solve the problem,

Tak perlu menjelaskan cara menyelesaikan masalah,

you just give it the goal,

Anda hanya memberi tujuan.

and it has to figure out for itself via trial and error

AI harus mencoba mencari cara sendiri

how to reach that goal.

untuk mencapai tujuan itu.

And it turns out that the way AI tends to solve this particular problem

Ternyata, AI cenderung mengatasi masalah ini

is by doing this:

seperti ini:

it assembles itself into a tower and then falls over

dia membentuk dirinya menjadi menara, lalu jatuh

and lands at Point B.

dan mendarat di Poin B.

And technically, this solves the problem.

Secara teknis, masalahnya terpecahkan.

Technically, it got to Point B.

Dia sampai ke Poin B.

The danger of AI is not that it's going to rebel against us,

Bahaya kecerdasan buatan bukan kemungkinan dia memberontak,

it's that it's going to do exactly what we ask it to do.

tapi AI akan benar-benar mematuhi kita.

So then the trick of working with AI becomes:

Maka, trik menggunakan AI adalah

How do we set up the problem so that it actually does what we want?

bagaimana kita menyusun masalah agar AI berfungsi sesuai harapan?

So this little robot here is being controlled by an AI.

Robot kecil ini dikontrol oleh AI.

The AI came up with a design for the robot legs

AI membuat desain untuk kaki robotnya

and then figured out how to use them to get past all these obstacles.

dan bisa menggunakannya untuk melalui rintangan ini.

But when David Ha set up this experiment,

Namun, saat David Ha membuat eksperimen ini,

he had to set it up with very, very strict limits

dia harus membuat batasan tegas

on how big the AI was allowed to make the legs,

seberapa besar kaki yang bisa dibuat oleh AI.

because otherwise ...

Jika tidak...

(Laughter)

(Tertawa)

And technically, it got to the end of that obstacle course.

Secara teknis, rintangan itu berhasil dilalui.

So you see how hard it is to get AI to do something as simple as just walk.

Sangat sulit membuat AI melakukan hal sederhana seperti berjalan.

So seeing the AI do this, you may say, OK, no fair,

Saat melihat ini, Anda mungkin berkata itu tidak adil.

you can't just be a tall tower and fall over,

Anda tak boleh menjadi menara tinggi, lalu jatuh.

you have to actually, like, use legs to walk.

Anda harus menggunakan kaki untuk berjalan.

And it turns out, that doesn't always work, either.

Ternyata, itu tidak selalu berhasil.

This AI's job was to move fast.

Tugas AI ini adalah bergerak cepat.

They didn't tell it that it had to run facing forward

Kami tak memintanya berlari menghadap ke depan

or that it couldn't use its arms.

atau tak boleh menggunakan tangan.

So this is what you get when you train AI to move fast,

Ini yang terjadi saat Anda melatih AI bergerak cepat.

you get things like somersaulting and silly walks.

Hasilnya adalah jungkir balik dan cara berjalan aneh.

It's really common.

Ini biasa terjadi.

So is twitching along the floor in a heap.

Juga bergeliat di lantai dalam bentuk gumpalan.

(Laughter)

(Tertawa)

So in my opinion, you know what should have been a whole lot weirder

Menurut pendapat saya, yang semestinya lebih aneh

adalah robot "Terminator".

is the "Terminator" robots.

Hacking "The Matrix" is another thing that AI will do if you give it a chance.

AI mungkin akan meretas "The Matrix" jika dicoba.

So if you train an AI in a simulation,

Jika AI dilatih dalam sebuah simulasi,

it will learn how to do things like hack into the simulation's math errors

dia akan belajar cara meretas kesalahan matematis simulasi itu

and harvest them for energy.

dan menjadikannya energi.

Or it will figure out how to move faster by glitching repeatedly into the floor.

Atau dia akan bisa bergerak lebih cepat dengan bergeliat di lantai.

When you're working with AI,

Bekerja dengan AI

tidak seperti bekerja dengan manusia,

it's less like working with another human

and a lot more like working with some kind of weird force of nature.

tapi seperti bekerja dengan kekuatan alam yang aneh.

And it's really easy to accidentally give AI the wrong problem to solve,

Kesalahan sangat mudah terjadi saat memberi perintah pada AI,

and often we don't realize that until something has actually gone wrong.

dan kami sering kali tak sadar sampai ada yang salah.

So here's an experiment I did,

Ini eksperimen yang saya lakukan.

where I wanted the AI to copy paint colors,

Saya ingin AI menyalin warna cat

to invent new paint colors,

untuk menciptakan warna cat baru

given the list like the ones here on the left.

dengan daftar seperti gambar di sebelah kiri.

And here's what the AI actually came up with.

Inilah yang diciptakan oleh AI.

[Sindis Poop, Turdly, Suffer, Gray Pubic]

[Kotoran Sindi, Kerdil, Menderita, Kemaluan Abu-Abu]

(Laughter)

(Tertawa)

So technically,

Secara teknis,

it did what I asked it to.

perintah saya dipatuhi.

I thought I was asking it for, like, nice paint color names,

Saya sudah memintanya membuat nama cat yang bagus,

but what I was actually asking it to do

tapi yang sebenarnya saya perintahkan

was just imitate the kinds of letter combinations

adalah meniru kombinasi huruf

that it had seen in the original.

dari daftar aslinya.

And I didn't tell it anything about what words mean,

Saya tak menjelaskan apa arti kata-kata

or that there are maybe some words

atau mungkin ada kata-kata

that it should avoid using in these paint colors.

yang harus dihindari dalam warna cat ini.

So its entire world is the data that I gave it.

Seluruh dunia AI ini adalah data yang saya berikan.

Like with the ice cream flavors, it doesn't know about anything else.

Seperti rasa es krim, AI itu tidak tahu tentang hal lain.

So it is through the data

Jadi, melalui data,

that we often accidentally tell AI to do the wrong thing.

kami sering salah memberi perintah secara tak sengaja.

This is a fish called a tench.

Ini adalah ikan mas tinca.

And there was a group of researchers

Ada sekelompok peneliti

who trained an AI to identify this tench in pictures.

yang melatih AI untuk mengenali ikan ini dalam foto.

But then when they asked it

Saat mereka bertanya

what part of the picture it was actually using to identify the fish,

bagian foto mana yang digunakan untuk mengenali ikannya,

here's what it highlighted.

ini bagian yang disorot.

Yes, those are human fingers.

Ya, itu jari manusia.

Why would it be looking for human fingers

Kenapa AI mencari jari manusia

if it's trying to identify a fish?

untuk mengenali ikan?

Well, it turns out that the tench is a trophy fish,

Ternyata, ikan mas tinca adalah ikan trofi.

and so in a lot of pictures that the AI had seen of this fish

Jadi, dalam banyak foto yang dilihat AI

during training,

selama pelatihan,

the fish looked like this.

ikannya tampak seperti ini.

(Laughter)

(Tertawa)

And it didn't know that the fingers aren't part of the fish.

AI tak tahu jari itu bukan bagian dari ikan.

So you see why it is so hard to design an AI

Itulah sebabnya sangat sulit merancang AI

that actually can understand what it's looking at.

yang bisa mengerti apa yang dilihatnya.

And this is why designing the image recognition

Inilah alasan merancang pengenalan gambar

in self-driving cars is so hard,

dalam mobil tanpa pengemudi sangat sulit.

and why so many self-driving car failures

Banyak mobil tanpa pengemudi gagal,

are because the AI got confused.

karena AI menjadi bingung.

I want to talk about an example from 2016.

Saya ingin membahas contoh dari tahun 2016.

There was a fatal accident when somebody was using Tesla's autopilot AI,

Terjadi kecelakaan fatal saat seseorang mengendarai Tesla dengan autopilot.

but instead of using it on the highway like it was designed for,

Namun, alih-alih memakainya di jalan tol seperti tujuan awalnya,

they used it on city streets.

dia menggunakannya di jalanan kota.

And what happened was,

Yang terjadi adalah

a truck drove out in front of the car and the car failed to brake.

truk melaju ke depan mobil dan mobilnya tak mengerem.

Now, the AI definitely was trained to recognize trucks in pictures.

AI itu sudah dilatih untuk mengenali truk dalam foto.

But what it looks like happened is

Namun, sepertinya yang terjadi adalah

the AI was trained to recognize trucks on highway driving,

AI itu dilatih untuk mengenali truk di jalan tol,

where you would expect to see trucks from behind.

di mana biasanya truk datang dari belakang.

Trucks on the side is not supposed to happen on a highway,

Truk di samping tidak seharusnya ada di jalan tol.

and so when the AI saw this truck,

Saat AI melihat truk itu,

it looks like the AI recognized it as most likely to be a road sign

sepertinya AI mengenalinya sebagai rambu lalu lintas,

and therefore, safe to drive underneath.

dan melaju di bawahnya akan aman.

Here's an AI misstep from a different field.

Ini kesalahan AI di bidang yang berbeda.

Amazon recently had to give up on a résumé-sorting algorithm

Baru-baru ini, Amazon menghentikan algoritme penyortiran CV

that they were working on

yang mereka kerjakan

when they discovered that the algorithm had learned to discriminate against women.

saat mereka tahu algoritme itu mendiskriminasi wanita.

Itu karena mereka melatih AI dengan contoh CV

What happened is they had trained it on example résumés

dari orang-orang yang pernah mereka terima.

of people who they had hired in the past.

And from these examples, the AI learned to avoid the résumés of people

Dari contoh itu, AI belajar menghindari CV orang-orang

who had gone to women's colleges

yang pernah belajar di universitas putri

or who had the word "women" somewhere in their resume,

atau yang berisi kata "wanita" di CV-nya,

as in, "women's soccer team" or "Society of Women Engineers."

seperti "tim sepak bola wanita" atau "Perhimpunan Insinyur Wanita".

The AI didn't know that it wasn't supposed to copy this particular thing

AI itu tidak tahu dia tak seharusnya menyalin hal tertentu

that it had seen the humans do.

dari yang dilakukan manusia.

And technically, it did what they asked it to do.

Secara teknis, AI itu sudah mematuhi perintah.

They just accidentally asked it to do the wrong thing.

Mereka hanya tak sengaja memberi perintah yang salah.

And this happens all the time with AI.

Ini sering terjadi dengan AI.

AI can be really destructive and not know it.

AI bisa sangat merusak tanpa menyadarinya.

So the AIs that recommend new content in Facebook, in YouTube,

Jadi, AI yang menyarankan konten baru di Facebook dan YouTube

they're optimized to increase the number of clicks and views.

dioptimalkan untuk meningkatkan jumlah klik dan tontonan.

And unfortunately, one way that they have found of doing this

Sayangnya, cara yang mereka pelajari untuk melakukan itu

is to recommend the content of conspiracy theories or bigotry.

adalah menyarankan konten teori konspirasi atau kefanatikan.

The AIs themselves don't have any concept of what this content actually is,

AI itu sendiri tak mengerti apa itu konten.

and they don't have any concept of what the consequences might be

AI juga tak mengerti konsekuensi

of recommending this content.

dari menyarankan konten seperti itu.

So, when we're working with AI,

Saat kita bekerja dengan AI,

it's up to us to avoid problems.

kita yang menjadi penentu untuk menghindari masalah.

And avoiding things going wrong,

Menghindari kesalahan

that may come down to the age-old problem of communication,

mungkin mengarah pada masalah komunikasi yang sudah ada sejak lama,

where we as humans have to learn how to communicate with AI.

di mana manusia harus belajar cara berkomunikasi dengan AI.

We have to learn what AI is capable of doing and what it's not,

Kita harus mempelajari yang bisa dilakukan AI dan tidak.

and to understand that, with its tiny little worm brain,

Juga memahami, dengan kemampuan berpikirnya,

AI doesn't really understand what we're trying to ask it to do.

AI tidak begitu mengerti perintah kita.

So in other words, we have to be prepared to work with AI

Dengan kata lain, kita harus siap bekerja dengan AI

that's not the super-competent, all-knowing AI of science fiction.

yang tak begitu kompeten dan canggih seperti dalam fiksi ilmiah.

We have to be prepared to work with an AI

Kita harus siap bekerja dengan AI

that's the one that we actually have in the present day.

yang kita miliki saat ini.

And present-day AI is plenty weird enough.

Dan AI masa kini agak aneh.

Thank you.

Terima kasih.

(Applause)

(Tepuk tangan)