So, artificial intelligence is known for disrupting all kinds of industries. What about ice cream? What kind of mind-blowing new flavors could we generate with the power of an advanced artificial intelligence? So I teamed up with a group of coders from Kealing Middle School to find out the answer to this question. They collected over 1,600 existing ice cream flavors, and together, we fed them to an algorithm to see what it would generate. And here are some of the flavors that the AI came up with.
Dakle, umjetna inteligencija poznata je po remećenju svih vrsta industrija. Što je sa sladoledima? Koje nevjerojatne vrste novih okusa bismo mogli napraviti uz sposobnosti napredne umjetne inteligencije? Dakle, udružila sam se s timom programera iz Srednje škole "Kealing" kako bih pronašla odgovor na ovo pitanje. Oni su skupili preko 1600 postojećih okusa sladoleda i zajedno smo ih stavili u algoritam kako bismo vidjeli što će proizvesti. I evo nekoliko okusa koje je UI smislila.
[Pumpkin Trash Break]
[Pauza za smeće od bundeve]
(Laughter)
(Smijeh)
[Peanut Butter Slime]
[Ljiga od kikiriki maslaca]
[Strawberry Cream Disease]
[Bolest kreme od jagoda]
(Laughter)
(Smijeh)
These flavors are not delicious, as we might have hoped they would be. So the question is: What happened? What went wrong? Is the AI trying to kill us? Or is it trying to do what we asked, and there was a problem?
Ovi okusi nisu ukusni onoliko koliko smo se nadali da bi mogli biti. Dakle, pitanje je: Što se dogodilo? Što je pošlo po zlu? Pokušava li nas UI ubiti? Ili pokušava napraviti ono što smo tražili, ali se pojavio problem?
In movies, when something goes wrong with AI, it's usually because the AI has decided that it doesn't want to obey the humans anymore, and it's got its own goals, thank you very much. In real life, though, the AI that we actually have is not nearly smart enough for that. It has the approximate computing power of an earthworm, or maybe at most a single honeybee, and actually, probably maybe less. Like, we're constantly learning new things about brains that make it clear how much our AIs don't measure up to real brains. So today's AI can do a task like identify a pedestrian in a picture, but it doesn't have a concept of what the pedestrian is beyond that it's a collection of lines and textures and things. It doesn't know what a human actually is. So will today's AI do what we ask it to do? It will if it can, but it might not do what we actually want.
U filmovima, kada nešto s UI pođe po zlu, obično je to zato što je UI odlučila kako ne želi više izvršavati naredbe ljude i kako ima svoje ciljeve, molim lijepo. U stvarnosti, ipak, UI koju imamo nije ni blizu toliko pametna za takvo nešto. Računalna moć joj je otprilike veličine gliste ili možda najviše jedne pčele, a zapravo, vjerojatno i manja. Stalno učimo nove stvari o mozgu koje potvrđuju koliko zapravo naša UI nije ni blizu pravog mozga. Današnja UI može obaviti zadatak kao što je identificiranje pješaka na slici, ali nema predodžbu toga što je pješak, osim što je skup linija, tekstura i stvari. Ne zna što je zapravo čovjek. Dakle, hoće li današnja UI učiniti ono što od nje tražimo? Hoće ako može, ali možda neće moći napraviti ono što mi zapravo želimo.
So let's say that you were trying to get an AI to take this collection of robot parts and assemble them into some kind of robot to get from Point A to Point B. Now, if you were going to try and solve this problem by writing a traditional-style computer program, you would give the program step-by-step instructions on how to take these parts, how to assemble them into a robot with legs and then how to use those legs to walk to Point B. But when you're using AI to solve the problem, it goes differently. You don't tell it how to solve the problem, you just give it the goal, and it has to figure out for itself via trial and error how to reach that goal. And it turns out that the way AI tends to solve this particular problem is by doing this: it assembles itself into a tower and then falls over and lands at Point B. And technically, this solves the problem. Technically, it got to Point B. The danger of AI is not that it's going to rebel against us, it's that it's going to do exactly what we ask it to do. So then the trick of working with AI becomes: How do we set up the problem so that it actually does what we want?
Recimo da pokušavate učiniti da UI uzme ovu skupinu dijelova robota i sastavi ih u nekakvog robota da dođe od točke A do točke B. Ako pokušate riješiti problem tako da napišete tradicionalan kompjutorski program, dali biste programu upute korak po korak kako da uzme dijelove i sastavi ih u robota s nogama, a onda kako da upotrijebi te noge da dođe do točke B. Ali kada koristite UI za rješavanje problema, to ide drugačije. Ne kažete joj kako da riješi problem, samo joj date cilj, a ona mora sama zaključiti, kroz sustav pokušaja i pogrešaka, kako doći do tog cilja. Ispada kako UI ovaj problem nastoji riješiti radeći ovo: sastavi se u toranj i onda se sruši i sleti na točku B. Tehnički, ovo rješava problem. Tehnički, došla je do točke B. Opasnost od UI nije što će se pobuniti protiv nas, nego što će napraviti točno ono što od nje tražimo. Tako da pitanje rada s UI postaje: Kako postaviti problem tako da zapravo napravi ono što mi želimo?
So this little robot here is being controlled by an AI. The AI came up with a design for the robot legs and then figured out how to use them to get past all these obstacles. But when David Ha set up this experiment, he had to set it up with very, very strict limits on how big the AI was allowed to make the legs, because otherwise ...
Ovim malim robotom ovdje upravlja UI. UI smislila je dizajn za noge robota i onda pronašla način kako ih iskoristiti da prijeđe sve ove prepreke. Ali kada je David Ha postavio ovaj eksperiment, morao ga je postaviti s veoma, veoma čvrstim ograničenjima u vezi toga koliko velike noge UI smije napraviti, inače...
(Laughter)
(Smijeh)
And technically, it got to the end of that obstacle course. So you see how hard it is to get AI to do something as simple as just walk.
I tehnički, došla je do kraja tog slijeda prepreka. 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, you can't just be a tall tower and fall over, you have to actually, like, use legs to walk. And it turns out, that doesn't always work, either. This AI's job was to move fast. They didn't tell it that it had to run facing forward or that it couldn't use its arms. So this is what you get when you train AI to move fast, you get things like somersaulting and silly walks. It's really common. So is twitching along the floor in a heap.
Gledajući kako UI ovo radi možete reći, OK, nije fer, ne možeš biti samo visoki toranj i srušiti se, moraš zapravo upotrijebiti noge za hodanje. A ispada kako ni to ne upali svaki puta. Posao je ove UI da se kreće brzo. Nisu joj rekli da mora trčati dok je okrenuta prema naprijed ili da ne smije koristiti ruke. Ovo dobijete kada kažete UI da se kreće brzo, dobijete salta i čudna hodanja. To je uobičajeno. 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 is the "Terminator" robots. Hacking "The Matrix" is another thing that AI will do if you give it a chance. So if you train an AI in a simulation, it will learn how to do things like hack into the simulation's math errors and harvest them for energy. Or it will figure out how to move faster by glitching repeatedly into the floor. When you're working with AI, it's less like working with another human and a lot more like working with some kind of weird force of nature. And it's really easy to accidentally give AI the wrong problem to solve, and often we don't realize that until something has actually gone wrong.
Tako da po mom mišljenju, znate što bi trebalo biti puno čudnije? "Terminator" roboti. Hakiranje "Matrice" još je jedna stvar koju će UI napraviti ako joj date priliku. Tako da ako stavite UI u simulaciju, naučit će kako napraviti stvari kao što su hakiranje u matematičke pogreške simulacije i upotrijebiti ih za energiju. Ili će skužiti kako se kretati brže tražeći greške kako bi prošla ispod površine. Kada radite s UI, nije kao da radite s drugim čovjekom, više je kao da radite s nekakvom čudnom silom prirode. I veoma je jednostavno slučajno dati UI da riješi krivi problem, a to često ne shvatimo dok nešto ne pođe po zlu.
So here's an experiment I did, where I wanted the AI to copy paint colors, to invent new paint colors, given the list like the ones here on the left. And here's what the AI actually came up with.
Evo eksperimenta koji sam napravila u kojem sam htjela da UI kopira boje za slikanje, kako bi izmislila nove boje, kada joj damo popis kao što je ovaj lijevo. I evo što je UI smislila.
[Sindis Poop, Turdly, Suffer, Gray Pubic]
[Sindis kakica, Govnasto, Patiti, Siva stidna] (okvirna značenja)
(Laughter)
(Smijeh)
So technically, it did what I asked it to. I thought I was asking it for, like, nice paint color names, but what I was actually asking it to do was just imitate the kinds of letter combinations that it had seen in the original. And I didn't tell it anything about what words mean, or that there are maybe some words that it should avoid using in these paint colors. So its entire world is the data that I gave it. Like with the ice cream flavors, it doesn't know about anything else.
Dakle tehnički, napravila je ono što sam je tražila. Mislila sam da sam je tražila lijepa imena za boje, ali ono što sam je zapravo tražila je da samo imitira vrste kombinacije slova koje je vidjela u originalu. I nisam joj rekla ništa o tome što riječi znače ili o tome kako bi mogle postojati riječi koje bi trebala izbjegavati u ovim bojama za slikanje. Dakle njezin cijeli svijet sastoji se od podataka koje joj dam. Kao i s okusima sladoleda, ne zna ni za što drugo.
So it is through the data that we often accidentally tell AI to do the wrong thing. This is a fish called a tench. And there was a group of researchers who trained an AI to identify this tench in pictures. But then when they asked it what part of the picture it was actually using to identify the fish, here's what it highlighted. Yes, those are human fingers. Why would it be looking for human fingers if it's trying to identify a fish? Well, it turns out that the tench is a trophy fish, and so in a lot of pictures that the AI had seen of this fish during training, the fish looked like this.
Tako da zapravo kroz podatke često UI slučajno govorimo da napravi krivu stvar. Ovo je riba linjak. Bila je grupa istraživača koja je trenirala UI da pronađe linjaka na slikama. Ali kada su je upitali koji je dio slike zapravo koristila da pronađe ribu, evo što je pokazala. Da, to su ljudski prsti. Zašto bi tražila ljudske prste ako nastoji pronaći ribu? Pa, ispada kako je linjak trofejna riba, tako da je na većini slika riba koje je UI vidjela tijekom treninga, 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 that actually can understand what it's looking at. And this is why designing the image recognition in self-driving cars is so hard, and why so many self-driving car failures are because the AI got confused. I want to talk about an example from 2016. There was a fatal accident when somebody was using Tesla's autopilot AI, but instead of using it on the highway like it was designed for, they used it on city streets. And what happened was, a truck drove out in front of the car and the car failed to brake. Now, the AI definitely was trained to recognize trucks in pictures. But what it looks like happened is the AI was trained to recognize trucks on highway driving, where you would expect to see trucks from behind. Trucks on the side is not supposed to happen on a highway, and so when the AI saw this truck, it looks like the AI recognized it as most likely to be a road sign and therefore, safe to drive underneath.
Tako da vidite zašto je toliko teško dizajnirati UI koja zapravo razumije u što gleda. I zato je dizajniranje prepoznavanja slike u samovozećim autima toliko teško, i zašto je toliko pogrešaka u samovozećim autima zato što se UI zbunila. Želim vam pričati o primjeru iz 2016. Dogodila se smrtna nesreća kada je netko koristio Teslin autopilot, ali umjesto da su ga koristili na autocesti za što je i bio napravljen, koristili su ga na gradskim ulicama. I ono što se dogodilo je da je kamion izletio pred auto i auto nije zakočio. UI definitivno je bila trenirana da prepozna kamion na slikama. Ali izgleda kako je ono što se dogodilo bilo da je UI trenirana da prepozna kamione u vožnji autocestom gdje biste očekivali vidjeti kamion sa stražnje strane. Stranice kamiona nisu ono što bi se trebalo vidjeti na autocesti, tako da kad je UI vidjela ovaj kamion, izgleda kako ga je vjerojatno prepoznala kao znak na cesti i zbog toga, kao sigurno za proći ispod.
Here's an AI misstep from a different field. Amazon recently had to give up on a résumé-sorting algorithm that they were working on when they discovered that the algorithm had learned to discriminate against women. What happened is they had trained it on example résumés of people who they had hired in the past. And from these examples, the AI learned to avoid the résumés of people who had gone to women's colleges or who had the word "women" somewhere in their resume, as in, "women's soccer team" or "Society of Women Engineers." The AI didn't know that it wasn't supposed to copy this particular thing that it had seen the humans do. And technically, it did what they asked it to do. They just accidentally asked it to do the wrong thing.
Evo pogreška UI na drugom polju. Amazon je nedavno morao odustati od algoritma za razvrstavanje životopisa na kojem su radili, kada su otkrili kako je algoritam naučio diskriminirati žene. Ono što se dogodilo je da su ga trenirali na primjerima životopisa ljudi koje su ranije zaposlili. A iz tih je primjera UI naučila izbjegavati životopise ljudi koji su išli na ženske fakultete ili koji su imali riječ "žena" negdje unutar životopisa, kao u "ženska nogometna momčad" ili "Društvo žena inženjera". UI nije znala kako nije trebala kopirati ovu osobitu stvar koju je vidjela da ljudi rade. I tehnički, učinila je ono što su je tražili. Samo su je slučajno tražili da napravi krivu stvar.
And this happens all the time with AI. AI can be really destructive and not know it. So the AIs that recommend new content in Facebook, in YouTube, they're optimized to increase the number of clicks and views. And unfortunately, one way that they have found of doing this is to recommend the content of conspiracy theories or bigotry. The AIs themselves don't have any concept of what this content actually is, and they don't have any concept of what the consequences might be of recommending this content.
A ovo se s UI stalno događa. Može biti destruktivna a da i ne zna. Tako da UI koje preporučuju nove sadržaje na Facebooku, YouTubeu, optimizirane su da povećaju broj klikova i pregleda. A nažalost, jedan način na koji se ovo može raditi je preporučiti sadržaj teorija urote ili netrpeljivosti. UI same po sebi nemaju predodžbu što taj sadržaj zapravo je i nemaju predodžbu o tome koje bi posljedice mogle biti kada se preporučuje ovaj sadržaj.
So, when we're working with AI, it's up to us to avoid problems. And avoiding things going wrong, that may come down to the age-old problem of communication, where we as humans have to learn how to communicate with AI. We have to learn what AI is capable of doing and what it's not, and to understand that, with its tiny little worm brain, AI doesn't really understand what we're trying to ask it to do. So in other words, we have to be prepared to work with AI that's not the super-competent, all-knowing AI of science fiction. We have to be prepared to work with an AI that's the one that we actually have in the present day. And present-day AI is plenty weird enough.
Tako da kada radimo s UI, na nama je da izbjegavamo probleme. A izbjegavanjem toga da stvari krenu u krivom smjeru može doći do drevnog problema komunikacije gdje mi kao ljudi moramo naučiti kako komunicirati s UI. Moramo naučiti za što je UI sposobna, a za što nije, i razumjeti kako, sa svojim minijaturnim mozgom gliste, UI zapravo ne razumije što želimo od nje da napravi. Dakle, drugim riječima, moramo se pripremiti na rad s UI koja nije svemoguća i sveznajuća UI iz znanstvene fantastike. Moramo se pripremiti na rad s UI koju zapravo imamo u sadašnjosti. A sadašnja UI je već dovoljno čudna.
Thank you.
Hvala.
(Applause)
(Pljesak)