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

Prevoditelj: Nina Bassi Recezent: Sanda Liker

So, artificial intelligence

Dakle, umjetna inteligencija

is known for disrupting all kinds of industries.

poznata je po remećenju svih vrsta industrija.

What about ice cream?

Što je sa sladoledima?

What kind of mind-blowing new flavors could we generate

Koje nevjerojatne vrste novih okusa bismo mogli napraviti

with the power of an advanced artificial intelligence?

uz sposobnosti napredne umjetne inteligencije?

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

Dakle, udružila sam se s timom programera iz Srednje škole "Kealing"

to find out the answer to this question.

kako bih pronašla odgovor na ovo pitanje.

They collected over 1,600 existing ice cream flavors,

Oni su skupili preko 1600 postojećih okusa sladoleda

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

i zajedno smo ih stavili u algoritam kako bismo vidjeli što će proizvesti.

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

I evo nekoliko okusa koje je UI smislila.

[Pumpkin Trash Break]

[Pauza za smeće od bundeve]

(Laughter)

(Smijeh)

[Ljiga od kikiriki maslaca]

[Peanut Butter Slime]

[Strawberry Cream Disease]

[Bolest kreme od jagoda]

(Laughter)

(Smijeh)

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

Ovi okusi nisu ukusni onoliko koliko smo se nadali da bi mogli biti.

So the question is: What happened?

Dakle, pitanje je: Što se dogodilo?

What went wrong?

Što je pošlo po zlu?

Is the AI trying to kill us?

Pokušava li nas UI ubiti?

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

Ili pokušava napraviti ono što smo tražili, ali se pojavio problem?

In movies, when something goes wrong with AI,

U filmovima, kada nešto s UI pođe po zlu,

it's usually because the AI has decided

obično je to zato što je UI odlučila

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

kako ne želi više izvršavati naredbe ljude

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

i kako ima svoje ciljeve, molim lijepo.

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

U stvarnosti, ipak, UI koju imamo

is not nearly smart enough for that.

nije ni blizu toliko pametna za takvo nešto.

It has the approximate computing power

Računalna moć joj je otprilike veličine

of an earthworm,

gliste

or maybe at most a single honeybee,

ili možda najviše jedne pčele,

and actually, probably maybe less.

a zapravo, vjerojatno i manja.

Like, we're constantly learning new things about brains

Stalno učimo nove stvari o mozgu

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

koje potvrđuju koliko zapravo naša UI nije ni blizu pravog mozga.

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

Današnja UI može obaviti zadatak kao što je identificiranje pješaka na slici,

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

ali nema predodžbu toga što je pješak,

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

osim što je skup linija, tekstura i stvari.

It doesn't know what a human actually is.

Ne zna što je zapravo čovjek.

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

Dakle, hoće li današnja UI učiniti ono što od nje tražimo?

It will if it can,

Hoće ako može,

but it might not do what we actually want.

ali možda neće moći napraviti ono što mi zapravo želimo.

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

Recimo da pokušavate učiniti da UI

to take this collection of robot parts

uzme ovu skupinu dijelova robota

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

i sastavi ih u nekakvog robota da dođe od točke A do točke B.

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

Ako pokušate riješiti problem

tako da napišete tradicionalan kompjutorski program,

by writing a traditional-style computer program,

you would give the program step-by-step instructions

dali biste programu upute korak po korak

on how to take these parts,

kako da uzme dijelove

how to assemble them into a robot with legs

i sastavi ih u robota s nogama,

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

a onda kako da upotrijebi te noge da dođe do točke B.

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

Ali kada koristite UI za rješavanje problema,

it goes differently.

to ide drugačije.

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

Ne kažete joj kako da riješi problem,

you just give it the goal,

samo joj date cilj,

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

a ona mora sama zaključiti, kroz sustav pokušaja i pogrešaka,

how to reach that goal.

kako doći do tog cilja.

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

Ispada kako UI ovaj problem nastoji riješiti

is by doing this:

radeći ovo:

it assembles itself into a tower and then falls over

sastavi se u toranj i onda se sruši

and lands at Point B.

i sleti na točku B.

And technically, this solves the problem.

Tehnički, ovo rješava problem.

Technically, it got to Point B.

Tehnički, došla je do točke B.

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

Opasnost od UI nije što će se pobuniti protiv nas,

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

nego što će napraviti točno ono što od nje tražimo.

So then the trick of working with AI becomes:

Tako da pitanje rada s UI postaje:

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

Kako postaviti problem tako da zapravo napravi ono što mi želimo?

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

Ovim malim robotom ovdje upravlja UI.

The AI came up with a design for the robot legs

UI smislila je dizajn za noge robota

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

i onda pronašla način kako ih iskoristiti da prijeđe sve ove prepreke.

But when David Ha set up this experiment,

Ali kada je David Ha postavio ovaj eksperiment,

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

morao ga je postaviti s veoma, veoma čvrstim ograničenjima

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

u vezi toga koliko velike noge UI smije napraviti,

because otherwise ...

inače...

(Laughter)

(Smijeh)

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

I tehnički, došla je do kraja tog slijeda prepreka.

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

Dakle, vidite koliko je teško dobiti da UI napravi nešto jednostavno kao hodanje.

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

Gledajući kako UI ovo radi možete reći, OK, nije fer,

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

ne možeš biti samo visoki toranj i srušiti se,

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

moraš zapravo upotrijebiti noge za hodanje.

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

A ispada kako ni to ne upali svaki puta.

This AI's job was to move fast.

Posao je ove UI da se kreće brzo.

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

Nisu joj rekli da mora trčati dok je okrenuta prema naprijed

or that it couldn't use its arms.

ili da ne smije koristiti ruke.

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

Ovo dobijete kada kažete UI da se kreće brzo,

you get things like somersaulting and silly walks.

dobijete salta i čudna hodanja.

It's really common.

To je uobičajeno.

So is twitching along the floor in a heap.

Kao i što je trzanje po podu dok je skupljena na hrpu.

(Laughter)

(Smijeh)

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

Tako da po mom mišljenju, znate što bi trebalo biti puno čudnije?

is the "Terminator" robots.

"Terminator" roboti.

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

Hakiranje "Matrice" još je jedna stvar koju će UI napraviti ako joj date priliku.

So if you train an AI in a simulation,

Tako da ako stavite UI u simulaciju,

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

naučit će kako napraviti stvari kao što su hakiranje u matematičke pogreške simulacije

and harvest them for energy.

i upotrijebiti ih za energiju.

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

Ili će skužiti kako se kretati brže tražeći greške kako bi prošla ispod površine.

When you're working with AI,

Kada radite s UI,

it's less like working with another human

nije kao da radite s drugim čovjekom,

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

više je kao da radite s nekakvom čudnom silom prirode.

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

I veoma je jednostavno slučajno dati UI da riješi krivi problem,

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

a to često ne shvatimo dok nešto ne pođe po zlu.

So here's an experiment I did,

Evo eksperimenta koji sam napravila

where I wanted the AI to copy paint colors,

u kojem sam htjela da UI kopira boje za slikanje,

to invent new paint colors,

kako bi izmislila nove boje,

given the list like the ones here on the left.

kada joj damo popis kao što je ovaj lijevo.

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

I evo što je UI smislila.

[Sindis Poop, Turdly, Suffer, Gray Pubic]

[Sindis kakica, Govnasto, Patiti, Siva stidna] (okvirna značenja)

(Laughter)

(Smijeh)

So technically,

Dakle tehnički,

it did what I asked it to.

napravila je ono što sam je tražila.

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

Mislila sam da sam je tražila lijepa imena za boje,

but what I was actually asking it to do

ali ono što sam je zapravo tražila

was just imitate the kinds of letter combinations

je da samo imitira vrste kombinacije slova

that it had seen in the original.

koje je vidjela u originalu.

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

I nisam joj rekla ništa o tome što riječi znače

or that there are maybe some words

ili o tome kako bi mogle postojati riječi

that it should avoid using in these paint colors.

koje bi trebala izbjegavati u ovim bojama za slikanje.

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

Dakle njezin cijeli svijet sastoji se od podataka koje joj dam.

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

Kao i s okusima sladoleda, ne zna ni za što drugo.

So it is through the data

Tako da zapravo kroz podatke

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

često UI slučajno govorimo da napravi krivu stvar.

This is a fish called a tench.

Ovo je riba linjak.

And there was a group of researchers

Bila je grupa istraživača

who trained an AI to identify this tench in pictures.

koja je trenirala UI da pronađe linjaka na slikama.

But then when they asked it

Ali kada su je upitali

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

koji je dio slike zapravo koristila da pronađe ribu,

here's what it highlighted.

evo što je pokazala.

Yes, those are human fingers.

Da, to su ljudski prsti.

Why would it be looking for human fingers

Zašto bi tražila ljudske prste

if it's trying to identify a fish?

ako nastoji pronaći ribu?

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

Pa, ispada kako je linjak trofejna riba,

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

tako da je na većini slika riba koje je UI vidjela

during training,

tijekom treninga,

the fish looked like this.

ova riba izgledala ovako.

(Laughter)

(Smijeh)

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

I nije znala kako prsti nisu dio ribe.

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

Tako da vidite zašto je toliko teško dizajnirati UI

that actually can understand what it's looking at.

koja zapravo razumije u što gleda.

And this is why designing the image recognition

I zato je dizajniranje prepoznavanja slike

in self-driving cars is so hard,

u samovozećim autima toliko teško,

and why so many self-driving car failures

i zašto je toliko pogrešaka u samovozećim autima

are because the AI got confused.

zato što se UI zbunila.

I want to talk about an example from 2016.

Želim vam pričati o primjeru iz 2016.

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

Dogodila se smrtna nesreća kada je netko koristio Teslin autopilot,

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

ali umjesto da su ga koristili na autocesti za što je i bio napravljen,

they used it on city streets.

koristili su ga na gradskim ulicama.

And what happened was,

I ono što se dogodilo je

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

da je kamion izletio pred auto i auto nije zakočio.

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

UI definitivno je bila trenirana da prepozna kamion na slikama.

But what it looks like happened is

Ali izgleda kako je ono što se dogodilo bilo

the AI was trained to recognize trucks on highway driving,

da je UI trenirana da prepozna kamione u vožnji autocestom

where you would expect to see trucks from behind.

gdje biste očekivali vidjeti kamion sa stražnje strane.

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

Stranice kamiona nisu ono što bi se trebalo vidjeti na autocesti,

and so when the AI saw this truck,

tako da kad je UI vidjela ovaj kamion,

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

izgleda kako ga je vjerojatno prepoznala kao znak na cesti

and therefore, safe to drive underneath.

i zbog toga, kao sigurno za proći ispod.

Here's an AI misstep from a different field.

Evo pogreška UI na drugom polju.

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

Amazon je nedavno morao odustati od algoritma za razvrstavanje životopisa

that they were working on

na kojem su radili,

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

kada su otkrili kako je algoritam naučio diskriminirati žene.

Ono što se dogodilo je da su ga trenirali na primjerima životopisa

What happened is they had trained it on example résumés

ljudi koje su ranije zaposlili.

of people who they had hired in the past.

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

A iz tih je primjera UI naučila izbjegavati životopise ljudi

who had gone to women's colleges

koji su išli na ženske fakultete

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

ili koji su imali riječ "žena" negdje unutar životopisa,

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

kao u "ženska nogometna momčad" ili "Društvo žena inženjera".

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

UI nije znala kako nije trebala kopirati ovu osobitu stvar

that it had seen the humans do.

koju je vidjela da ljudi rade.

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

I tehnički, učinila je ono što su je tražili.

They just accidentally asked it to do the wrong thing.

Samo su je slučajno tražili da napravi krivu stvar.

And this happens all the time with AI.

A ovo se s UI stalno događa.

AI can be really destructive and not know it.

Može biti destruktivna a da i ne zna.

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

Tako da UI koje preporučuju nove sadržaje na Facebooku, YouTubeu,

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

optimizirane su da povećaju broj klikova i pregleda.

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

A nažalost, jedan način na koji se ovo može raditi

is to recommend the content of conspiracy theories or bigotry.

je preporučiti sadržaj teorija urote ili netrpeljivosti.

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

UI same po sebi nemaju predodžbu što taj sadržaj zapravo je

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

i nemaju predodžbu o tome koje bi posljedice mogle biti

of recommending this content.

kada se preporučuje ovaj sadržaj.

So, when we're working with AI,

Tako da kada radimo s UI,

it's up to us to avoid problems.

na nama je da izbjegavamo probleme.

And avoiding things going wrong,

A izbjegavanjem toga da stvari krenu u krivom smjeru

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

može doći do drevnog problema komunikacije

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

gdje mi kao ljudi moramo naučiti kako komunicirati s UI.

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

Moramo naučiti za što je UI sposobna, a za što nije,

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

i razumjeti kako, sa svojim minijaturnim mozgom gliste,

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

UI zapravo ne razumije što želimo od nje da napravi.

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

Dakle, drugim riječima, moramo se pripremiti na rad s UI

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

koja nije svemoguća i sveznajuća UI iz znanstvene fantastike.

We have to be prepared to work with an AI

Moramo se pripremiti na rad s UI

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

koju zapravo imamo u sadašnjosti.

And present-day AI is plenty weird enough.

A sadašnja UI je već dovoljno čudna.

Thank you.

Hvala.

(Applause)

(Pljesak)