AI Special 3#: OpenAI’s Independent Creative Director, Rodger Werkhoven: Harnessing Creativity in the Age of AI and the impact of ChatGPT

0:00Pandora's box is opened. That is always the case if people invent something like a hammer, they build beautiful stuff, or sometimes they bash people's heads in with it.

0:14Welcome to episode three of the Tech Leaders podcast AI series. Karendra and I have a real treat for you this week. Roger Verkoven from OpenAI. That's right, OpenAI. Roger joins us from the International Short Film Festival in Oberhausen, Germany, where he was part of the production team in a recent AI-generated movie. Roger has been involved in exceptional projects, namely ChatGPT and DALI to be discussed this, of course. And we also delve into the intriguing world of AI-generated content and explore the true nature of authorship and how this is going to be impacted. We also examine the rapid diffusion of AI technology and the disruptive implications, along with the endless possibilities and also the controversies surrounding AI's role in shaping our creative landscape.

1:14Furthermore, we discuss OpenAI's auto-GPT plus the next evolution of ChatGPT 3.5 and 4. We also confront the boundaries of AI and its immense potential. This is a fantastic conversation. Let's jump right in.

1:38Roger, thank you so much for coming on the Tech Leaders Podcast. We have been so excited about this one. Where do we start? I tell you what, I've got an idea. Let's start with this. What excites you most about the evolution of AI technology? It excites me that I now have like a partner, a synthetic partner, like a PA almost also, that does so much of the dreary work for me. But I also see it, of course, it doesn't understand me. But in the dialogue, it seems it understands me. It doesn't make many mistakes. It does what I want it to do or tries in the creative field. and surprises me because it fails often.

2:25And the surprises are, you know, taking my work and art in new directions. So I love the surprises. It's awesome that it doesn't perfectly understand me, which OpenAI won't agree because they want perfect alignment, of course, of this technology with what humans want from it. But also, you know, it already was there for a long time doing all the stuff under the hood for the advertising industry. It was already there finding the right target audiences that were open for the messages, the commercial messages at the right time, at the right place. So it was already helping us. And so it was already in our phones for a long time, making all these beautiful pictures as if you have a really awesome camera.

3:13It finds stuff we want to find in these photo albums. etc so it was already there but now it's also so visible you know it's almost also addictive because if you in natural language you prompt something and it returns something it's like a gambling machine almost for what will it pop up will it bring up the right for images that i want or it's really addictive too so there's so many aspects and Roger it's so good to have you with us and I'm just interested in your creative process how do you feel that it's evolving because of the new tools that you can deploy at the moment how is it pushing you to new creative dimensions yeah that's an interesting one because I'm always I have this second agenda so yes I make stuff I like but I am not this free autonomously working creative artist I always am also thinking how will an audience react?

4:19What discussions will come from it? Will I inspire them to work with this technology because that's the goal? Or will I only scare them? I cannot go too far out because then people will perhaps not understand its relevance anymore. On one hand, I would like to do really really new stuff on the other hand I'm doing stuff that people can relate to and that's also why I work with other artists and bring them on board to give it also a more experimental edge so I don't have to do it all by myself. So give us an example of a recent collaboration where you've pushed yourself and other artists have pushed themselves and you've created something new or different or surprising that has really kind of blown your mind well because you know i'm here at the oberhauser film festival because of a short ai generated film is screening here and i asked artist arnold kuna to work with me and work with a open ai with dali too when i asked him come work with this artificial intelligence it's awesome he was against it he was like no way you know art should be for humans and not for machines or synthetic minds I'm not gonna and then I showed them what we could do with it and then he was intrigued like how can it do that and then when we gave him access and he could fiddle along and try and work with Dali too he became the artist he should be he started testing it but also So he tried to do stuff that OpenAI would perhaps not like.

6:11So he would prompt things that would trick Dali into making images and art that are, how do you say, not straightaway pleasing or happy, happy. So he became that artist that I was looking for. And that was interesting because he took it further. He was even kicked out of the program for a while because he used words that OpenAI didn't like or the model, the tuned model didn't like much. And so I also had to try and bring him on board again, which worked. And so people that are now looking at these results at that movie that we made, yeah, they are amazed that they see stuff moving, talking. It is all very natural, the animations of these new entities we created.

7:06They look like they're really alive. And we also have a lot of humor in it. So people also think it's funny, which is a good thing, of course. But what they don't understand, what people can't get their head around, it's an AI by D-Idea. is another AI company that's working also with the APIs of OpenAI. And what people can't get their head around is that you give it one image of an entity, of a creature. You write a lot of text that is like 15 minutes of speech, for instance, and that this AI studies the face, studies the text, makes a speech of three minutes, but it makes it in just a couple of seconds.

7:58We never as humans were able to do that. A rapper could never create a rap of five minutes in one minute. It was not possible. But these AI models are doing that. And that is also what people think is fascinating. It's bending time. Can I ask you a slightly different question, which is, what are the most irritating things that people ask you when it comes to thinking about this debate? Or what are the things that are most provocative? People ask me, but Dan, you are not the artist. Dan, the AI is the artist. Or they ask us, it's not your work. So it is all OpenAI's work now. They say, yeah, but what you're making is fake.

8:45So you're dangerous. You're putting fake stuff into the world. Yeah. And I like these concepts and ideas and questions a lot because they trigger the most fun discussions about what is real, what is not real, who should own the copyrights. Will an AI do this stuff all by itself or does it really need human input? And so you get really interesting discussions from these questions. Could you give us your sort of rebuttal to that accusation then, Roger? If someone says to you, this piece of art is not yours, Roger, what would you say back to that individual? I always give them the example of how Jeff Koons works.

9:28He is, if he wants to, he wants a big steel balloon dog, then he sends an email to the steel factory that makes these things for him. He sends a mail saying, I need a five meter high balloon dog this time. Should be very smooth surfaces. I wanted purple, but the purple color has to be transparent so the steel shines through. I don't want to see any lines where you connected well the steel. When it's finished, send me a couple of pictures so I can judge it. and when he sees these pictures from the steel factory he says yeah yeah yeah this is right this is nice send that one to dubai send that one to the united states to my customers that pay millions for it so what he is doing is the same thing he is prompting but in an email and the two creatives or workers at the steel factory but is it then their steel balloon dog because they made or is it jeff coons his steel balloon dog because he ordered it and that's the same thing with artificial intelligences like dali because you breathe them and they make it and so yeah your mind your creative direction makes uh not all the difference but a lot of difference yeah of course of course but i thank you for giving us uh you know a bit more of a simplistic rebuttal though because I think that that is a big conversation that people are having right now and are going to have more so over the next couple of years especially in sort of the education sector for example we were talking about with another guest recently and so on and so forth but let's maybe put this on hold for now then because I would love to just unpack your background a little bit Roger for the listeners who are unfamiliar with yourself could you maybe give us a little bit of information on your background and the key milestones in your career that set you on the path that you ultimately ended up following yeah as a kid i was playing around with the commodore 64 computer of my daddy and me too yeah and i liked the aesthetic of these pixels and then later on there was this movie tron the first one there was so much computer animation in it and and they had this it was still very crude you you had all these polygons but they used it as a new aesthetic that belonged to that world of the computer inside the computer and i loved that and i wanted to make that as well and then here in the netherlands there are these art schools and only one had a new concept and it was working with computers full time to create art.

12:22So you, yeah, in the first year, you also had to learn how to paint and how to do traditional art forms. But after that first year, we could work with computers alone and with nothing else but computers. And that was really cool. So we were doing innovative stuff. We also were not allowed to use software made by software makers. We had to program these computers, because otherwise you are stuck with the stuff that others already programmed for you. And after that, in the software industry, people needed creative people that could also make user interfaces. The Encyclopedia Britannica was going to be digitalized by a Rotterdam company that was very innovative, but they also needed little buttons and interfaces.

13:09And so I started designing that for the software company. But then advertising agencies saw what I was doing and they thought, yeah, 3D computer animation can be cool. And later the internet was growing up quite a bit, but most companies still did not really take that serious. So I was, yeah, we should do something with the internet. That's going to be awesome. And then they would be, nah, that's just a gimmick and it will pass away. Yeah. And but at DDB here in Amsterdam, DDB is an American advertising network. They were like, well, perhaps it's not so big yet, the Internet, but we want you to experiment with it.

13:57And when we have clients that want something with Internet, could you do that for us? But then you also have to do all these, you know, traditional campaigns with television commercials, etc. etc so when artificial intelligence became a thing people knew to find me they knew where I was and one of the developers that's key that's for sure a key moment one of developers I learned to know in that software company at the beginning of my career he went to Silicon Valley and he saw me fiddling around with very primitive AI and he said well open AI it's a little research lab and And they are going to invite people to work with DALI 2, their next model.

14:45They're looking for a couple of creative directors worldwide because they wanted to become a model for the people of the whole world. Would you like to be on board of that project? And I said, yeah, that's awesome. And so OpenAI asked me to join them. And that was a year ago. Wow. And what a journey it's been since then, I suppose, then. True. Yeah, indeed. Yeah, it was a rollercoaster. A year ago, I explained at Cannes Advertising Festival that there was something like a company, OpenAI, and that they were making, that we were researching this tool that could be a creative companion for creative people.

15:24And people didn't know what OpenAI was, and they didn't know what artificial intelligence could do. And so also then again, history repeats itself. People were like, ah, it's a gimmick. Yeah. Oh, you can make images. That's fun. I can make much more beautiful images. You know, what you are doing with that AI, I can do much higher resolutions with Photoshop. So it's gimmicky. But that was a year ago. And now it's incredible how this technology took a flight. Well, how would you describe the culture of open AI? What's Sam Altman like as a leader? Because it's such a transformative company, or has been so far this year anyway.

16:07Just really keen to understand what makes that organization tick. It's two different organizations now. It's the organization before Microsoft's investments and the organization after Microsoft's investments. and before Microsoft's investments, it really felt like a research lab and a bunch of people with this vision of aligning artificial intelligence with humanity's goals. And they really were serious about it. I never met Sam Altman. Natalie Summers was our contact there. She's really an awesome lady, but she would be the spider in the web with all these creative directors from all over the world.

16:50and she would explain that OpenAI really did not yet know what to do with a model like Dali because it was not meant to be a creative tool for the creative industry at all. It was just a test to see if the AI is aligned with what humans want. Does it understand what humans want from it? So the prompting, does it understand text and does it come up with the right results following that prompt, just a testing tool. That was what it was. So it felt like, you know, being part of a small research team and discovering this technology for the greater good, finally making artificial general intelligence happening.

17:36And this was just a small step towards that. It is a very diverse company. That's awesome. I never saw so much diversity. It is nothing like a laboratory it was uh more like yeah an art school art academy felt very relaxed and although the working ethics are people are so dedicated because they all think it's awesome what they're doing and we had contact with the developers right away straight away there was nothing in between us so we could you know discuss the stuff that we discovered the the bad stuff the good stuff the ideas we had about how the interface should be changed so we really felt with this small group of people like we are building it we are changing it we are improving it but now yeah then all of the sudden well all of the sudden you cannot go on being a research lab forever if you don't have the money to do the research so the investments of microsoft were very welcome but that did change a lot and of course the gpt revolution in september that changed a lot too that's also something open air did not anticipate they did not uh anticipate that it would boom so fast and so big because dali was not straight away embraced by the world you know it was the creative professionals and animators and manga designers game designers so it was a niche in the creative industry that wanted to experiment with DALI 2.

19:12But JetGPT, it's so easily accessible. You just type something, it talks to you. It's the Jet format. That's genius. And nobody feels the fear of, I'm afraid, am I going to do right or wrong? Am I going to look stupid? But with DALI 2, people had the idea of DALI 2 returned something ugly. Oh, it's my fault. I'm not creative enough. And now DALI 2 produces something ugly. So that did not happen with ChatGPT. Totally different thing. Just before we talk a bit more about ChatGPT, which I think will open up a whole different scene of conversation. I do want to go back to the innovation journey at OpenAR.

20:02I think it's really interesting that they set out to test the AI's ability to do that kind of understanding comprehension piece, almost like being able to translate what does a human being say? Can we play it back? Do we understand it enough? At what point in the innovation journey did it start tipping into, actually, we've kind of accidentally created one of the world's most incredible creative tools that the creative industries can benefit from? On the creative side, we were amazed that it always returns something. So it would never tell us, I don't understand. It would always return something.

20:44Or if we used words that OpenAI already had labeled as inappropriate, then it would return a warning sign. Don't do this too often because you will be put out of the program. which sometimes happened to some people but then they were let back in because it was research phase but that was really surprising it always returns something but also the surprises so stuff that never existed before it would also try and make that so if we would ask it for mashups the first thing that I did was mashups of my culture Delft Blue Earthenware with American pop culture, Star Wars. So I asked Dali to create me a Delft Blue Stormtrooper busts.

21:43And no problem, you know, it would straight away produce those things or mix-ups of animals and stormtroopers. And it would do that. And that was really cool. And it was not just like it's a stormtrooper, but it has the snout of a dog or something silly. No, it made very intricate mashups of these concepts of what is a stormtrooper and what is a frog, for instance. And that was amazing. People thought that it was really thinking how to tinker this out. And yeah, but basically it was predicting what pixels should be next to the other pixel to create a stormtrooper frog. I asked Dali, create me an astronaut that is enjoying a popsicle, you know, an ice cream popsicle.

22:38Yes. And astronauts have these helmets on and so they can't. Then Dali created, first it created an image and you would see the popsicle, but that it slipped on the screen of the astronaut. Then it created inventions. How can I under devise or create something that an astronaut could put in the popsicle? So it would come up with solutions that this astronaut could eat a popsicle. And then I would even confuse it. And I showed it to Natalie Summers. I said, look, I said, it has a sense of humor because of the popsicle, you know, on the screen, on the visor. It has a sense of humor, which it doesn't.

23:22You know, it's us interpreting the image as humoristic. And, oh, look, it's intelligent because it's looking for solutions to have this. I should not eat the popsicle. But it is not at all intelligent. you know it then finds in its big data set images of not images but knowledge because there are no images there but the knowledge of people eat popsicles through a mouth so then it would blend mouth kind of stuff things in these helmets and we concluded that it came up with smart solutions but it was just, you know, blending these concepts. And so it was confusing sometimes. So at what point did the team think, actually, this could be something that could be used in a completely different way to what we'd originally conceived?

24:13Because that's what's really interesting about the innovation journey. Yeah, that's a good question. Because we got also, we were, no kidding, they were also doing psychology tests on us because they really were very precautious. Is it addictive, you know, to make all these images? And we went on for hours. Sometimes I worked 20 hours a day with this technology. Gosh. So they were researching us. And one of the questions was also, what if we would stop the DALI program? The group of people we were, we would say, That would be terrible because this could be helpful for humanity. This could be a great tool for people that have difficulty explaining themselves with autism to with a tool like this, communicate with others.

25:10We would come up with all sorts of reasons not to kill the DALI program. but they really thought when we got the AI aligned then we don't need DALI anymore because it was just a test tool but then MidJourney came out with a model MidJourney 1 it was a bit of a copycat of OpenAI text to image, a little bit more complex on the Discord servers and they were asking money for it then at OpenAI I don't know how these things worked, but people decided, hey, if Midjourney is making money with it, perhaps we should also give that a try because it's such an expensive experiment also, you know. And even the billions you have, you will burn them eventually.

26:02And then they thought, perhaps we should also put it out slowly and have people on board outside our research group and think of a monetization solution. Are we going to ask money per render or are we going for subscription models? So then they were thinking about we should probably make money of it. because of mid-journey. Interesting. Can I ask you about ChatGPT then? I'm sure you've noticed this, Roger, and yourself as well, Crenson. I think we actually spoke about it, Crenson. Obviously, Italy banned ChatGPT due to privacy concerns, and I think there were a couple of other concerns as well.

26:49Are the Italian government completely, are they outliers here, or are you seeing more concerns by institutions and governments? Are we going to see more problems for TrackGBT going forward? Yeah, I think so. And I think the Italians, in a way, are right. I think they should pick it up with Europe as a whole. And I think these concerns are right. Because, yeah, there is, of course, OpenAI was quite cautious with only training the model on stuff that's already out there in the open. so it's not going to steal your data but the data and everything that you shared somewhere and it has access to it can happen that it was trained on that as well but you did not give consent perhaps you put your stuff on Facebook or wherever but you did not say if an AI model is going to be trained on this then I think it's okay so people did not put their stuff online to have models trained on it but yeah also like one of the google people wrote a wrote an interesting book societies that share more are the more successful societies as well it's the success of homo sapiens as a species is sharing, sharing, sharing, knowledge, knowledge, copying, see somebody doing something smart, try to do it also, is the success of Homo sapiens.

28:29So if we would now back off and with the European community say, yeah, now we are not going to allow this technology or you cannot just train your models or the people, the people will say, no, nobody can use my data, then that will stop progression I think and I think that the European citizens if they want this awesome technology to augment them and help them and do all the dreary stuff and and do awesome creative stuff together with the people of Europe then the people of Europe perhaps should think by themselves you know what stuff do I not want to share what is the real stuff okay you know if You're having a beautiful, loving moment with your loved one.

29:18Perhaps you don't want that to be in a model. But there's a lot of stuff that I think is not going to harm your situation if that is in these models. And these models can help you then even better. But perhaps, you know, I'm a bit naive. I am on that side of the spectrum. But what I think would be a good idea is to, with Europe, decide on creating a CERN for AI development, like, you know, the space station where all these countries participated in and CERN. All these countries should together develop AI in a fashion that fits the European democracy. There are so many angles on the ethics debate, aren't there?

30:07I mean, there are people who are maybe creators in their own right. They may be artists, videographers, sculptors, photographers, poets, writers, who find a kind of that they're not quite sure how to navigate this landscape, where new creations are being created as kind of derivatives of some of the inputs that they may not have willingly or knowingly shared. and I just wonder whether you from your perspective because you are a creator who uses these technologies to make new things do you have some sympathy with that argument and how do you how do you talk to artists who are fearful of how to navigate this new world?

30:46First I'll tell them there's some artists that are prompted a lot because people like their work and the work of those artists is not in the system it is a understanding of the work that is in the system But if you prompt a bouquet of flowers in the style of a certain artist, then it will look like in the style of that artist. But these artists were not so famous before the boom of people prompting these. So I think it's also an awesome PR tool. There are now artists that can ask large amounts of money for just talks about their experiences with artificial intelligence and how people are prompting them.

31:31But they became so famous because all of a sudden it was a thing. People are prompting my name to create work that looks a bit like my work. and yeah but yeah i i would say that's that's awesome it's free publicity for you but if you want to opt out there is still a possibility you can demand that your work is not um in the training model but i would say you know also the stuff that is in the model building a model is quite expensive and it takes some time so the stuff that is in a model you are already as an artist you're much further so let people play with your old styles and have fun with your old styles you become famous through it and then you can show the next level where you're at you can show what you're now creating and people will love it and it will not be in a model because it's too you know too slow it takes too much time to put that stuff in the model but if the stable diffusion people train new models with your style then you are already further and it forces you also to be innovative with your style so you will see as an artist that you are surprising yourself it's like in the fashion industry where the fashion industry said a while ago let's not fight each other in court anymore because of a jacket that looks like our jacket but is different on seven points let's just invent fast fashion and not have just four seasons of new stuff but let's make new stuff every week so that copycats cannot even catch up and so not fighting each other in court has been a creative boost for the fashion industry a pity for the environment of course That's the turn side of the core.

33:32But not going to court about whatever will boost the pace of the creative development. I really loved your analogy of the fashion industry. That's a good comparison, actually. So I wanted to ask you about your LinkedIn title. Your headline on LinkedIn, I think, is brilliant. AI will not replace you. A person using AI will replace you. I mean, I thought that was genius. And I've actually referred to that twice in conversations recently. So the question is, Roger, what do you think the impact on the labor market will, how profound will it be over the next couple of years as AI technology evolves?

34:13Yeah, it will be very disruptive. You know, computers, when computers or home computers were becoming cheaper and everybody could or companies could afford all these home or just desktop computers for people to work with. Then also there were people that said, I'm not going to work with a computer. It's a silly thing. It's stupid. But yeah, people working with computers replaced the people not willing to work with computers. And so history is repeating itself. But what people think is that AI will make some jobs obsolete. I think, you know, call center people, they will lose their jobs because AI models can be stabilized all the time, very friendly.

35:03And, you know, they're very good at dialogue now. But is that a place where we should have humans, you know, in call centers and hearing all these complaints about products and services of people? is that a nice job to have for people or should we as fast as we can replace these people with jet gpt's sort of technologies there will also be new jobs you know because even if you're not a designer now with a model like mid-journey or dali or stable defusion you can do artwork for local entrepreneurs that never could afford decent expensive advertising from the advertising agencies but if you were working in a call center and you lost your job as a call center operator you can try and find fun stuff to do with AI like being a becoming a designer for instance for your local entrepreneurs and shopkeepers and so this technology is also going to create new jobs you know there's there's companies now looking for prompt engineers and they are serious in it it's not just a pr stunt but they are really looking for prompt engineers and you can make a decent living with it too so new professions will also be there and also in the advertising industry it's not that a layer of people will be out of a job but they will move on augmented with artificial intelligence they can do the stuff that the only the companies working for brands like nike could afford because that you know that awesome stuff they made would take a lot of resources time and money and now augmented with artificial intelligence all brands can afford that kind of awesome work.

37:00And the creative people that work for these awesome brands like Nike, they will now have to astound themselves and find the next level of creativity. So I think it will be an enormous boost for the creative industry. And also we'll need much more people working with these AIs and understanding the models and being fresh, coming up with fresh ideas. Because we are already now seeing that people working with these AI models make stuff that sounds the same. You can almost recognize JetGPT or GPT-4. You can recognize straightaway mid-journey images or stable diffusion images. So to be original and fresh, you need original and fresh thinkers.

37:51There will be more demand for these people as well. It will create a lot of new opportunities for entrepreneurs and people. Yes, I mean, I would say that obviously, I mean, at BT &EE, we have fantastic contact centre staff and they are rather brilliant, obviously. But I would say that one of the thoughts that's crossed my mind is all through history, as technology has evolved, the types of jobs that human beings do have also evolved. so you know at the time when the the motor vehicle grew in prominence and took over horses and carriages you know you didn't need as many groomers and you didn't need as many stable boys and girls and you didn't need all the kind of carriage polishes the changing nature of tech means that we adapt we adapt we adapt and the beauty of the human being is that we are incredibly adaptive creatures and i think you're you're absolutely right roger in the sense that this actually opens up a huge range of possibilities in so many different ways many of which we can't even begin to predict or imagine and that's quite exciting but I absolutely also sympathize with an awful lot of people who might do some of the more repetitive type of jobs in factories manufacturing all sorts of all sorts of places or administration that might be very fearful that what does this mean for me, for my family, for my income potential?