Interview with the creators of the neural network Nikolai Ironov: how it works, what is the trick in general, and why everyone calls him Kolya
Recently we met with the guys from the Art. Lebedev Studio to ask them in detail about the design neural network, which they passed off to clients as a real living designer for a year - with a name, photo, mail, portfolio, page on the FB and all the things we asked along with @ de_arnst . And the answers were given by the art director Sergey Kulinkovich and technologist Roman Kosovichev . We talked for almost two hours and recorded a very long podcast - enough for the trip to and from the office. But we know: they love to read on Habré, so keep the longread transcript. See you in the comments in half an hour.
Spoiler: at the end of the article - a version of the logo from Ironov for our podcast.
What do you say to those who say that this is a fake? How can they prove that Ironov is real?
Sergey: I can say hello to everyone who does not believe. But I would not say that Nikolai Ironov is only a neural network, it is rather a complex of systems, among which there is a neural network, algorithms and all sorts of technical things that allow you to reproduce the work of a designer at a basic level. It's surprising to me that so many people questioned its existence, because in fact, now [people] have all the technology at their disposal to do this kind of thing.
Project Ironov is not about a technological achievement, but about a precedent - about how we packed and combined these technologies in such a way that they were able to create just such a story. This story is not about "Oh my God, the neural network creates logos." This story is about clearly demonstrating that technology has long been ripe for relatively autonomous solutions to tasks that many people consider to be creative.
Roman: It's also interesting that we have broken the system into understandable steps. The designer works the same way: separately with the ability to draw something, separately with the ability to understand something.
Do I understand correctly that the main feature of Ironov is that he gives the result faster than a person?
S: I would not say that this is the only advantage. Speed is understandable, this is all good, but besides that, we solve many other related tasks and problems. The fact is that, unlike a living person, Nikolai is not afraid. We have created a system that filters the results within the bounds that we set. But unlike a living person, you can expect from Nikolai that a living designer is afraid to admit it.
A living person is afraid to even draw this. He can physically reproduce such forms, but his history, his formation as a designer with all the communication at specialized conferences, with mentors - on the one hand, it developed him as a designer, and on the other hand, it formatted his brains in a certain way. Therefore, Nikolai generates a lot of things, and he generates fundamentally new things.
It's not a problem for us at all to make small shifts by teaching Nikolai to make normal logos, but it's not interesting, we already know how to draw normal logos. It's cool to draw logos that don't fit in people's heads. Logos that require additional space in the brain to put knowledge about logos there. Nikolai is also needed for this, as a source of new pieces.
Is it even possible for a designer to have such a" look "in a neural network? the strongest abstractionism, and it is unclear how to apply it and how to choose from the options that he did. Maybe the designer cannot draw what Nikolai is doing, because he is afraid that he will be fired for this. Neural network cannot be fired. do you think that living designers don't paint so freshly?
S: Design craft looks very complicated until you start unwinding this ball piece by piece. When we started to do Ironov, we did not understand what we were doing. We had an idea in our head that we need to automate this topic. We tried to go from different angles: somewhere we found a cool visualization technology, somewhere we connected a neural network that allows you to create unique images that are nowhere else, somewhere we tweaked a system that converts a text brief into a set of units of meanings. Then, at some point, Nikolai clicked and was born. Everyone sees the work of a designer as a black box, you come to him - he creates. If you look closely, what does create mean? You can unwind it into small steps that are quite automated. In many cases, even a neural network is not needed.
Nikolay Ironov was born here not because we are the smartest, but because we had experience in streaming logo production. We have generated a lot of logos over the years with the help of different designers for different clients. From the inside, we saw the patterns behind this. Accordingly, a very large part of the designer's work is spent to put logos on media, or to parse them, make patterns, graph design, or look at the client's brief, get units of meaning from there. We have seen that in many places these things are quite automated.
Comes to you, for example, a bakery. Founder sends a description of the company, says that it is a family company: "we make buns and croissants, we need a logo." A living designer will start drawing croissants, applying different text options. If we feed the same text to Nikolai, he can see the word “family” in the description, and deduce the heart symbol from the word “family” and suggest such a logo for the family bakery. In some contexts, it will be more appropriate and contrasting than those logos drawn by real people.This is the idea. There is absolutely no problem to draw a logo, there is a problem to draw a logo that will be contrasting. Nowadays, an infinite number of logos are drawn every day, they are all clean, but no one cares about them. Our goal is to find growth points at the intersection of technology and design, so that we get the advantage not only in speeding up work, but also the advantage of wide creative range.
R: It's enough just to generate a logo, it's enough just to generate a clear, clean form, but it will be tasteful. The trail of the author will be visible there. One of the most difficult tasks was to make sure that this brush of the author was not visible in the work. So that every time a new logo is generated, you don't understand what it is made of, how it works, in principle, because if you look at simple shape generators, they all work according to clear, logical algorithms.
I think that being overlooked becomes a little blinkered. If you are an experienced dude who is well versed in how to draw logos, you do it for a long time, you have a lot of expertise, then you can not generate something new, but you will rest on your experience and do a beautiful thing. You cannot generate something new every day. When you have not seen it, this is also a plus, because you can do a new, outstanding thing that will surprise everyone. This task was one of the most difficult, we worked on it the most. If we made a regular logo generator, then it would be immediately read, it would not be interesting. It wouldn't work the way Kolya does.
But you trained the neural network on an array of studio data. It turns out that the neural network has seen enough of the work of people, and this is the result based on what has already happened?
R: Our system consists not only of neural networks or algorithms. The complexity of the system is that it includes a lot of knowledge accumulated in the Studio, experience on how to how to build a logo. As a technician, I myself had to go through all the stages of how a logo is drawn.
First, the designer reads the brief from the client and understands what topics can be associated with the logo. Then he looks for unexpected solutions so that they are related to the context of the brief. After that, he looks at analogs, looks at how he can draw it himself, what new can he offer. Going through all these stages, in the end he gets a logo. But we have a complex system of rules that the designer already knows himself, they, in principle, lend themselves to algorithms, so when they start arguing among themselves, an unexpected result is obtained. The logo goes through all these steps and systems, at the end you no longer understand how you built it, what part of the logo you changed, what rules the logo succumbed to at a certain step.
S: In other words, there is the genre of the logo. And we describe this genre with a code: there is a sign, there is a text part. For the text part, such a set of algorithms is used, for the characters another. Many people on the outside think that a neural network is a huge array of data, everything just gets in the way and the output is something averaged. This is not the case. We use the neural network only in those places where it is necessary, otherwise it is a set of different rules, where we also take into account our experience in building logos. For example, we understand what compositional schemes are, we understand the theory of complementary colors, we understand in what range it is possible to transform the font part of the logo so that it does not look ugly. Accordingly, these are the restrictions that we put in Ironov on the basis of our expertise. They are, on the one hand, limitations for him, because he “lives and thinks” within this framework, but on the other hand, we try to “let him go” as much as possible so that absolutely crazy logos can exist.
In the case of fonts, we have been drawing them for a very long time and accumulating this data. As a result, we got a huge array of different options that Nikolai can operate in order to solve certain problems. In fact, what we have announced now is the beginning. Nothing prevents, based on these principles, building more complex systems in other areas: illustrations, management. The fact that this has now appeared with us is not some kind of insight, it is just a combination of circumstances, including economic ones.In the sense that at this point we had the opportunity to use the market in which we operate, the configuration of the team, our needs, ideas, ambitions, and so on - in order for this breakthrough to happen in design.
But absolutely nothing prevents from doing the same things in other industries, even creative ones. This business can be promoted into specific steps, and these steps can be automated. Our project is about this, we have highlighted what is already there. Due to the fact that we hid the personality of Ironov from real clients, we thereby proved that an absolutely synthetic design can not only exist as an experiment, but can live on signs and labels. And not just live, but be loved by end users.
How long does it take for Ironov to complete the task? And in what form do these tasks come in?
R: Now the task comes in the form of a text description and company name. And it generates the result in seconds. Further, only the same set of algorithms, the results of neural networks, which we spoke about earlier, changes. A client came, entered text, title, description. Further, the system understands the context with the help of the image and metaphor search system.
We gave an example where the text describing a banal bakery comes in: "we make buns, pies, and so on." Kolya can find unexpected associations, and when he finds them, he further transforms into images. For example, he will have an association with the sun, because the sun is rye. He will draw the sun for the bakery, and this, most likely, can go to many bakery owners, because this is an unusual move. But, in principle, it is included in the context. Then he chooses several different images of the sun - these are not the same shapes, different in the composition of points. And different algorithms that help enrich these options. For example, he can combine images that are completely different in context. He can add an unexpected object to the sun, such as a pipe. And someone may really like it, he will choose this logo for his future company.
Next, Kolya can apply unexpected color palettes, compositional solutions - all this happens quickly enough, because some of the changes are made with the help of neural networks, and some with the help of algorithms that work with graphics. Thus, it makes an output from several logos in a few seconds.
Do you edit the texts that come to the entrance? How do you explain what the company does?
S: In our experience, that doesn't make sense. If you look at the most popular logos, in rare cases they reflect the nature of the company. Sports brands don't feature runners. It's a matter of framework. Nothing prevents us from spending a lot of effort, a lot of time to realize this part, to make him [Ironov] more like a human, but this does not help solving the problem. Our task is to search for fundamentally new solutions in graphic design. And in some contexts his imperfection, the fact that he does not fully take into account and understand the context, is an advantage that gives him freedom of creativity. It can be done if necessary, but we don't go there.
Now we are focused precisely on the search for new visual opportunities that we endow him with, accelerating his work in this sense. Our basic idea is that it doesn't make sense for tens of thousands of hours to pull curves and try to get the perfect shape. In many contexts, it doesn't make sense to do this, because the success of the logo does not lie here. If you open up Pinterest, you will see that thousands of designers generate neat looking logos every day. Nobody cares. There is no particular interest in spending your life replenishing this design plankton. Better to create systems that allow you to quickly do routine tasks.Shoot the entire range of solutions using automatic algorithms, and spend your brains yourself looking for the right meanings in order to understand this particular logo is right for this client or not, what kind of story you can tell about it. This is the designer's job - to find the one that will fit the context from a huge number of different options.
Why did Nikolay Ironov cause so many discussions? Because many have realized deep down in their hearts that they are taking the time to hone their skills where they can already be automated. In other words, if you sit and draw, draw curves, try to persuade a client to love your style, and then an absolutely synthetic logo gets on a cafe sign or on a bottle label (something that you won't even see in your career), this raises questions. Do you need to keep tinkering with curves, or should you think about other logo components? About the context in which it will fall, how it will live, why it is needed, what is your role as a designer.
That is, you need a designer to come up with a context for the logo generated by the neuron? It looks not like a design, but like the fact that we have a picture on which we have to put context in order to sell to the client.
S: Not exactly. Since Nikolay, as in the example with the bakery, can offer a more adequate solution in terms of composition and, in general, the composition of the logo itself. He may find some images that the designer would not think of. And the designer will see a large list of images that Nikolai generated and say: “Wow! The bakery has a heart logo, I'll use that. "
If a living designer sat down to draw, he would be faced with the fact that there is a threshold. You still need to open the tool and start visualizing using your skills. You have learned to work with vectors, you work with them in the style that you are used to. You have to spend time looking for meanings using the tools that you know. And here you just sort through, and not random forms, but forms that are tied to the context by an algorithm.
Nikolay is not a complete replacement for a designer with all the advantages and disadvantages, he is an accelerator, a tool of a new order. This is not Photoshop, where you spend hours to bend curves. It is a tool that generates new solutions at the click of your fingers. You can transform them in some ranges, ask them to suggest other colors, compositional solutions or "play with fonts", which everyone loves very much. Nikolay will do everything himself, instead of trying to simultaneously think about meanings and somehow visualize.
How many variations does Nikolay generate? How is the choice of what to give to the client?
S: It generates an infinite number of options, the choice is made by the designers and art directors, who are in fact Nikolai's operators. That is, based on their experience, they choose solutions that are better suited to the context, are contrasting in relation to everything that is in this area.
That is, the art director previously “stung” at living people and spent hours waiting for the logo, but now he presses on Ironov and spends seconds?
S: That's the catch. Having looked at the process of creating logos, we see that in many cases it is just a dumb brute force. Friends, we [art directors] are just looking at a lot of different options.
Are you waiting for "the same"?
S: Quite right. You just sit, and some designers send their versions. At the same time, if you choose for too long, then they start to have questions about how much more to draw, "I offered you a good option." Nikolai does not think, this is acceleration. Sometimes, you can push yourself so hard with a living designer and it will be such a difficult process that you yourself will sit down to write your own Nikolai Ironov.
Does this mean that in the future there will be no designers in Artemy Lebedev's studio, but there will be neural networks and art directors?
R: I think that designers will just change their work profile a little. There will be a little less routine. Designers will go in my direction [technologist]. Ideally, they should know how to code the routine, because they have the expertise on how to build a logo, all that remains is to automate this process.I think that a future designer is something between a technologist and a creative person, because it gives a lot more opportunities when you understand how it works in terms of design, and how it can be implemented in terms of technology. I think that the future belongs to such people. I think we need to learn Python and neural networks.
Will you automate the art direction? To "watch" the neuron so that it understands the "click" or not?
S: We already have some kind of art director inside - a system that filters out very bad options from acceptable ones. But it is precisely the context and understanding of what may and may not come in, in my opinion - it's just a matter of time. If you look closely at how the art director works, what context he takes into account, then it is quite possible to assume that this is an unwinding thing. The question is, will there be in any company in the near future so many free creative and financial resources to implement such a thing. Here demand should shape an event that will occur in the future. If there's nothing behind it, what's the point of doing it?
Having created Nikolay, we have solved many different problems. The task of being the first in this topic, the task of finding new solutions in the field of graphic design, and so on. These are tasks that fit well into the business we are doing, so we have the energy to solve them. But it is likely that there will be areas in which such a combination of circumstances will not occur. We were just lucky.
How many man-hours did it take to develop Ironov?
S: It took a very long time, it is very difficult to calculate. Firstly, the project itself, and secondly, we used those systems and ideas that were laid down a long time ago. In particular, “ matrix of ideas worked in the Studio - this product was implemented a very, very long time ago. We can say that the ideas that arose in this product, over time, appeared in Ironov. This is a very long process, stretched out over the years. I'm afraid we won't be able to count.
R: It all started with research, I spent a lot of time to understand if this was possible in principle. At the initial stage, we did not know where we were going and why. That is, it was a research work: we tried different technologies, it happened more often in our free time, so it is impossible to count. But a lot of time [spent].
A very common question. Often admin panels look like complete shit. Everyone has a cool showcase, but you go into the admin panel, and there is just hell. Have you made Ironov's dashboard fun or not?
S: Of course, we did it, because we understood that time passes, people change. If we left this technology in the hands of one holder of the knowledge of how to use this tool, there would be a great risk that all our efforts would simply be wasted. Ironov has a user interface - we did our best to make it as clear as possible to a person who comes there from the street for the first time and does not know how it works. Everything is built on chat.
Is this a bot like that?
S: Not that a bot. One way or another, we simulate real communication between the art director and the designer. Very often in our speech, Kolya skips, because communication happens as with a real person. We say try again - it brings up even more options. He has a lot of human traits. And the most interesting thing is that this is not just a marketing shell, it is also a part of the product, because through the implementation of this kind of relationship with Nikolai, we understand how Nikolay should work. This is the big problem with many experiments in the field of neural networks - they all end in experiments. Some kind of technology has appeared, we are able to generate pictures - take one more step, come up with the right context, pack it into a clear script, make this solution replace some part of [work]. In our case, this is the complete opposite. There is a good interface to which you can connect any designer inside us, which is what we are doing now. We just give out such an accelerator.
R: Moreover, this interface helped enrich Kolya, because when we connected a designer who could poke at this interface, he immediately began to understand how Kolya works, what he can do, what can be improved.There is also a click in the head of the designer that you can teach him a more correct composition. So I will set such settings there, see what he does with such settings. Yeah, here he is not very good - I can still push this Kolya. Thanks to the cool interface (the word “cool” is important here), the designers helped a lot in the finalization process. We would not be able to generate all this on the command line using some numbers that the designer does not understand. It wouldn't be possible.
S: We have created this system, further it is not difficult for us to teach him new ways of visualization. Further, everything rests simply on the fact that we will add new styles and approaches, expand the boundaries or narrow them, thereby opening up another great advantage over a living person. If you try to tell the designer that you want him to suddenly start drawing "like this", then, firstly, you will spend a lot of nerves, and secondly, it is difficult. People's brains are really oak, especially if you learned to draw in a certain style, believed in yourself as a designer, collected a bunch of likes, and so on, then you know that this is the horse you should bet on. Then the art director comes to you and says: draw me something fundamentally different. That's it, you're paralyzed! Therefore, in this context, it is better to go to the technologist and say: can you write a plugin for Kolya?
In fact, it doesn't matter to an art director who who does the work. It doesn't matter if it is drawn by a person or a system. The main thing is that the work is done so that a new solution is found. Accordingly, answering the question about the future, about how the role of the designer is changing, I dare to assume that it will change towards the creation of systems that allow you to quickly, interestingly and effectively solve routine tasks. And the main thing is to find new things.
What's all this written on?
R: Since this is a system, a lot of things are included here. I'll start with the history of the study. When we were at the start of the project, we had to test many different technologies. At the time when I just started taking the first steps, neural networks such as, for example, Pix to pix, when a picture is generated based on another picture, were very popular. I rushed to try this kind of system, for example, GAN systems and their related networks.
At first, we generated common geometric shapes. It was a kind of test that the neural network can generate how it can diversify simple shapes. The first version of Kolya is sharpened, pixelated images, which, as I thought then, can be transformed into SVG format and further work with this form. This is what it was in the first stage. It was interesting because I used various tools for this kind of transformations from a sharpened incomprehensible image to SVG. Even then, blots of a very cool and unusual shape were obtained. This was something that the designer would not generate just like that, per click.
Then I realized that we need to switch to packing all the input data not into pixel images, but into SVG points. I started using recurrent networks - they are used to recognize and generate handwritten text, for example. This turned out to be the right decision, because we were able to teach how to generate various forms, and reached full-fledged images that could be further used for the logo.
First, we used Python and tested different neural networks. Then they came to several specific neural networks. We call one of them "designer" - it generates an image. There is a network that classifies images - this is an algorithm that understands: good shapes are generated or not very good, but bad ones are cut off. Further, various libraries are used that enrich the images generated by the neural network. This is how it works.
How did you derive the parameters of "crap"?
R: It was a visual process. First, we selected the unsuccessful forms that the first neural network generates, taught the "art director" that these forms are not very good, so you won't miss them. But these forms are okay. At this stage, the "art director" could already cut off good results from bad ones. It does not work at the level of getting into the context, but, in principle, of pattern recognition, as far as readability is what the first neural network recognized.The "art director" does not check the context, does not check how similar it is to the client's brief, to what extent it can be applied in the current situation, but this is also a matter of time and resources. I believe that this can also be taught to a neural network - to understand how an apple falls into the context of a logo for Microsoft.
S: Many people look at the logos that Nikolai generated, and wonder how? Not how they were drawn, but how they were received. It's funny that people who have not gone through hundreds of different logos - even, uneven, red, white, colored - have a very ossified understanding of what a logo should be. Some people believe that the logo must necessarily reflect the company's activities. Some still believe that the logo should be light enough to be printed on a business card. People have a lot of stereotypes about this, but in fact it is a group hallucination. Nothing can prohibit us from using in logos any absolutely crazy thing, even synthetically generated.
Why can we allow a coffee shop in a computer game to have a completely futuristic logo that glows, spins, emits particles? But when we leave the apartment and see something like this, then we have questions? In fact, there is no border, the border is inside the head. Our task is to swing this range so that people do not tremble over what they are used to, what they grew up around, what surrounded them. There is really no objective reason to require a logo or other visual communication to look like this. If you managed to reach an agreement with a client, a consumer, your sales are made, you have a cool new logo that others do not have, then no one will tell you anything. This is the main idea - to destroy stereotypes by hook or by crook, synthetic designers and living. We see that with living designers, these stereotypes are difficult to destroy, even in a format where the client cannot tell you anything. Even in this format it is very difficult to get people to draw something fundamentally new, something flying off, something crazy, but this is the interest. Do you want to be remembered as the designer who drew a pair of neat logos? It's boring!
You humanized Ironov for a reason, but in order to relieve some of the strain from what the neural network did in two seconds, and not a person worked all day? How did your customers react after you released it?
S: Now it is difficult to separate the effect that occurred in the media from the effect of the logos themselves. But, of course, brands and customers are experiencing great reach, and therefore many are satisfied. Moreover, some of the guys wrote that now they see even more value in these logos. They became participants in this point in history, when the first, completely synthetic logos solve real business problems, a new context appeared.
The whole story associated with Ironov is an integral part of the logos that he drew. It is important to understand that just like that, a set of lengths and shapes makes no more sense than any other set of lines and shapes. Context is of paramount importance. There is no absolutely universal figure. Maybe there is a circle, but without context it has no meaning. Any designer can draw a circle, but not every designer can make it become a historical theme. This is the role of the designer - to package your ideas in such a way that they acquire new meanings, so that you can tell about them. The logos they [clients] have at their disposal, their value and the story behind them are already far more expensive than any freelancer or studio can paint them. This is the main idea.
Of course, after we discovered Ironov, the clients were happy. But the most interesting thing is that there were clients who were happy before when they found out that Nikolai was not a human being.
To what extent did you humanize him? Was it a "special forces" project even within the Studio, about which not everyone knew? Did you write letters for him? Anyone answered?
S: It was very interesting. First, the question arose how it should look. We parsed the portraits of all the Studio employees, fed them to the automatic system and got a certain photo of a robot that looked rather sinister - it is on announcement page , you can see it.
Next, we created a page for him on the FB and started mail. Moreover, we created an intranet card to make everything believable. Then he began to work and, along with other designers, send his options for consideration. The most interesting thing is that they began to write to him in a personal on FB: Nikolay, I am a fan of your works, I have an offer for you. Letters began to pour into the mail with various questions: you have a unique style, how do you do it?
Did you want to hunt him?
S: I think it was about freelancing.
Did Nikolay "levitate" or not?
S: We didn't have time to deal with this tightly, so we somehow formally responded [to messages]. Nikolay was busy and closed for new projects.
Did you only use Nikolai in express design?
S: Not ready to comment.
Did the art director know when the logos came to him that it was a neural network? For example express designs, where Lebedev is listed as art director.
S: Of course.
Then I wonder why the second version of "It's Time to Blame" was not made by Nikolai?
S: Just another designer made a different logo.
That is, there was no problem for Nikolay did another approach?
S: No, [second version] is just another logo. Moreover, we have already achieved great results. Ironov made a result that came in. In fact, the first result is no worse than the second - it is cool, sane. It was especially interesting to look at the reaction when phrases were heard about the fact that this cannot be a coincidence, and so on. The whole team enjoyed it.
How does the Studio treat Ironov? Maybe you have a tote inside: Ironov/NeIronov?
S: I probably won't talk about the inner kitchen. I can say that we now have the technology at our disposal, and we are trying to use it and look for new growth points. Now the main achievement of Ironov is that we have seen for ourselves that this is not an experiment, that it can go beyond the bounds of a joke, that we can use it to solve real problems. This is the main thing that we got from the launch of the project.
When the news about Nikolai came out, for some time we monitored social networks, watched what people wrote. It's amazing how much people don't want to go into details, understand what happened. Someone goes to comment on the style of the logos or say that Khovansky never liked him.The signal is so strong that people forget to take it apart and see what actually happened - look at the work, how it was made and how these ideas can be applied to themselves. In the announcement, we published a video with the guys, we gave answers to possible questions - take and reproduce the same effect in your business, try to find where your data streams are in your business, how you can pass them through automatic systems, but everyone is sitting and discuss Nicolas style.
It seems that if people were arranged in such a way as to go to disassemble and get into the essence, you would not get such an effect.
S: That's fine too, because our announcement is a statement that you can do this. Everything is there, all the tools are there. It's great if they see and start doing something for themselves. This will be a great result for me.
People very often do not understand what design is. They look at the final image on a white background: "I don't like" or "I like". And design is a different genre. It is clear that there are routine tasks, but we are playing a completely different game, we are not interested in visualizing some neat things. The announcement that was around Nikolai, the way we presented it - this is the design in fact. The statement we posted is design.
Not the shape of Khovansky's logo is the ultimate answer to what Nikolai Ironov is. We did not just manage to generate some new, beautiful, interesting, original forms, but gave them to living customers, and the customers produced this and applied it to their products. Thus, we proved that a completely synthetic design can stand on the same step as a man-made one.
The precedent has happened so far only on logos, but the design is wider. Can Nikolay, say, design a helicopter?
S: Why not? Maybe of course.
It seems that there are spheres, besides visual images, where not only an endless stream of images is important. it is important to know the strength of materials so that everything does not collapse. Can all this be taken into account and also loaded into Ironov?
R: Strength mainly consists of very strict requirements and rules that are subject to the same automation. It is clear that there are many different situations: where a house can stand, where it can be built. But this is also a matter of time - with the help of simple algorithms and the same neural networks that are not very complex in terms of filling, you can teach the system to build houses.
Architecture is a very big science, which includes many steps: designing a building, what kind of foundation pit should be, what materials are needed, what kind of facade should be - all these are steps. The same thing that we described now, everything consists of steps.
S: We are lucky that logos are easy to draw in terms of labor intensity. There are not many systems that need to be written that will force the results of work to fall into the genre. There are also more complex areas where genre requires taking into account a larger number of contexts: the same material, materials, cost, and so on.
Now, having gained experience in building Nikolai, I see no objective reason to assert that complex creative professions cannot be replaced or supplemented by artificial systems that are built on algorithms and neural networks.It is clear that a specific technology has limitations, but for this people are needed - to use the technology not just head-on, but use it where it is appropriate.
In the case of Ironov, we use it to get original images. There is no pre-drawn logo template because it would be a boring constructor. We use a neural network to make the logos copyright. Perhaps, in automating the work of an architect, a neural network is not needed at all, but some other technology is needed. Now there is no reason to spend such a huge amount of money, time and effort to try to somehow reproduce this thing [in architecture]. It's a matter of time.
You both said that a future designer is a technologist/art director, but not everyone can become an art director, otherwise there would be more of them.
R: When I advised everyone to learn Python and neural networks, I did not mention that a person's imagination is much more important - without it you cannot come up with a beautiful picture. A dry programmer does not understand the sense of beauty. In my opinion, you need to have a rich imagination, learn everything around, accumulate images, read a lot of books and so on. There are already tools that allow you to do a lot on one fantasy, not even knowing how to program.
S: Not everyone can be a designer either. You are born, you begin to feel the urge to visual activity, you begin to design. After tens of thousands of hours, you become a designer. You taught your head to solve design problems with predictable results. Previously, you had to absorb this craft, draw with your hands a large number of some primitive forms - in order to see the patterns that lie behind successful or unsuccessful forms, what the client comes in and what does not. Now this is gradually changing: when machines can generate content on their own, you need to learn your internal neuron from other things. If you want to be a designer or art director, you do not need to learn how to reproduce something, but to develop an understanding of in which contexts, which tools work best. This is the difference.
I believe all designers. It's just that someone calls himself that, and someone does not. One way or another, we design our relationships in the family, at work, on our website, on Instagram, and so on. It's just that the tools are different. Accordingly, interesting transformational times are coming for the classic average designer. This does not mean that people are not needed, good brains are always needed. This means that routine skills are devalued as faster, brighter, sometimes more talented cars enter their territory.
If we perceive design as a way of solving problems, where can Ironov be applied in the future? Will he be able to become a manager?
S: We're big fans of automating everything. It's hard for me to imagine such a position that cannot be automated, from an office manager to a programmer. If you really decompose a person's work into steps, you can see many points of growth. But from an economic point of view, all these technologies are very expensive. To make Nikolai Ironov, a large number of qualified people worked for a long time, it is expensive. Everything else can also be automated even now, but there are no economic circumstances. Sometimes, it is cheaper to hire a person, because the automation of his work will not give a large number of new growth points.
Ironov is not about the fact that we can make logos automatically. It has many consequences of a different order, including the relationship of the Studio with the public, expanding the boundaries of the possible in our own heads. This is also the thing that Ironov decides. Perhaps in other areas there is no combination of benefits that automation can bring, so these processes do not occur.
In fact, nothing prevents us from entering 3D territory. Nothing prevents you from entering the territory of industrial design and talking about real objects. It is a matter of what data streams we will connect, and in what business contexts we will embed it.
The neural network and other algorithms enter the territory of things that were previously difficult to automate, and people from this territory run away "higher into the mountains", where they take advantage of the complexity of their brains. They will live there for some time, and then the neural networks will become more complex, and they will climb to a higher height.People will use their imperfections to create something that this machine cannot simulate. Such is the race.
R: Automation opens up new professions, this is inevitable. If you go back into history, look at how the industrial revolution took place, then there were the same problems. I'm not sure if it is appropriate to compare with the current processes, because now everything is different, but nonetheless. In the past, everyone believed that factories would replace people. This process dragged on for a very long time. And now the same thing is happening - new professions will inevitably appear. This is very cool in the sense that routine tasks do not have to be done by people who will have more opportunities to bring something new to humanity. An accountant will not have to sit and count something - he will have tools that can reveal his hidden possibilities and ideas. He can turn out to be a very cool art director who clicks a button and selects things that are very relevant to the context.
What was your name for the neuron inside the team? How did the name Nikolay Ironov come about?
S: There are several meanings behind his name.
Ironov is iron, iron, a reference to iron brains.
If you write down N. Ironov briefly and say quickly, you get n [e] ironov. The solution was very close all the time, right in front of the nose.
It's also funny that Ironov is consonant with the word irony, and this also falls into the context. This is just an example of working with contexts. We could have called him Petya, but it would have been different.
Has Ironov ever produced frightening or, on the contrary, funny results?
R: The first results of the neural network were not that frightening, they inspired hopelessness. These were the first experiments, then I felt that it was impossible to generate an understandable image.
S: For the first time, I felt the movement of hair on my back when I saw that Nikolai painted a cactus. He sent a green rectangle, at the entrance there was a need to draw a cactus, I looked and thought: I need to look for a living designer who will draw a cactus in such a laconic manner. Very often, due to the fact that a designer has a lot of context in his head, some skills, an understanding of how the tools work, it is difficult to get a "clean signal" from him. In logo design, this is a big problem - how to show the image as succinctly as possible.
Once Nikolai sent a flamingo of four pink rectangles of varying degrees of color saturation. But they were instructed in such a way that I would sweat a lot to get the same recognizable image. After that I started calling him Nikolay. After he sent the cactus, I realized that we were entering the territory of something new.
Once we did a demonstration of Nikolai. We took a description of [the company] from a person, passed it to Nikolai, and in the first issue he drew an indecent symbol. And it was not an accident - he somehow analyzed the text and found in this description such signals. A living designer would never do that, he would be afraid. Nobody would ever visualize a business in a way that Nikolay can. This is Nikolay's strength - he doesn't care. Art director sometimes may look and think: maybe boldly, but why not.
This is a big growth point in the sense that you separate the selection stage from the logo creation stage. Very often, consciousness dictates the image that you propose. If you have learned to draw something well, then with a high degree of probability you will offer what you can. This is also a very strong acceleration of the creative process, the search for new images, because the generation of meanings is separated from the choice of which meanings to use.
After such a story, I immediately want to see what Ironov will give to the brief about our podcast.
S: We can.
The brief was like this:
Habr Speshal is a podcast where developers, designers, scientists, businessmen and other smart guys talk about difficult things in simple language.
Here's the result:
And the last big question: someday, even in Will you open Ironov for others in some way?
R: We made a statement [by releasing Ironov]. I think so, there will be some kind of foundation, an exoskeleton for developers to further develop the industry....