Something like a teaser. Apps create portraits from selfies, algorithms write music and poetry, and all this is no longer even surprising. Artificial intelligence paints pictures without waiting for inspiration, releases albums, develops brand designs and even takes fame from top models. How appropriate is it to call such algorithms “creative”? Can the concept of “creativity” be applied at all to a soulless machine?
This post is based on the episode episode podcast “Let’s Get Out!” In it, we spoke with those who are directly involved in the creation of creative AI. These are the head of the experimental machine learning systems department in the SberDevices department Sergey Markov, art director of the Art. Lebedev Studio, designer and co-founder of timestripe.com Sergey Kulenkovich and co-founder and sound designer of the Endel project, composer and pianist Dmitry Evgrafov.
In the international competition Only robots participate in the Robot Art Competition.
An artificial intelligence called Sony Flow Machine writes music and has already released its own album IAMAI… The DeepBach program can turn a melody into composition similar to the works of Bach…
Artificial intelligence writes music
Even in creative tasks, no, no, yes, and I want to automate something. In the 19th century, mathematician Charles Babbage tried to create the first computing machine – the “analytical engine”. Ada Lovelace, who worked with Babbage, already assumed that someday machines would be able to compose meaningful and complex melodies. The first “classical” piece of music written with the aid of a machine was Illiac’s Suite for String Quartet. Lejean Hillier wrote it on the Illiac computer, which generated the music at random. In 1993, composer and scientist David Cope released Bach by Design, the melodies on which were recombinations of the great maestro’s tunes performed by a machine called Emmy.
Although machines have been generating music for a long time, news about the next joint project of people and machines comes out regularly. For example, Singer Grimes with the project Endel released an endless digital lullaby. Endel technology creates sound backgrounds for work, sleep or stress reduction. Now these songs are used on the planes of the Japanese airline ANA, and the Endel app was the app of the year for the Apple Watch, but it all started with the popularity of playlists for concentration or relaxation on streaming services. Dmitry Evgrafov thought it would be interesting to create such sound compositions automatically. The algorithm he created has released ten albums, becoming the first robotic artist to sign a deal with Warner Music. Dmitry decided not to limit himself to static tracks. We know that music can affect the psychological and even physical state of a person. However, machine learning allows you to create a personalized track that would take into account all available information about the listener: down to breathing rate and heart rate. This is exactly what Endel is working on right now – creating a personalized, adaptive melody that can immerse the listener in a flowing state.
It is possible that streaming services are already doing or are about to start making their own content using AI. Then, in the world of artificial musicians, there will be fierce competition for the listener’s attention. Moreover, it is likely that the listeners themselves will not even notice at first. For example, people listen to sleep playlists in the background, and not because they want to hear a specific artist. At some point, Spotify had a rather simple idea: “Why pay real musicians if we can just order music for sleep playlists on a turnkey basis, redeem the rights and not pay royalties?” No one suspects at all that they fall asleep under the tracks specially created for him by Spotify. From the current state of affairs to the introduction of generative artificial intelligence, in the literal sense, one step.
Artificial intelligence creates design
In June 2020, Art. Lebedev Studio reported that for more than a year, artificial intelligence, which was passed off as a living person, has been developing logos. This is Nikolai Ironov. For conspiracy, he was called a remote employee, a photo was generated based on the pictures of all the studio employees. The algorithm learns from already created images using keywords in descriptions, selects colors and fonts. Among his works (portfolio) – logo for the YouTube channel of Yuri Khovanskiy and a version of the logo for the show “It’s Time to Blame” Ruslan Usachev, who was really dissatisfied with the work of Ironov.
The work of a creative specialist can be broken down into stages. If you carefully consider these small parts, it turns out that the entire process of creating a work consists of a routine and a huge number of actions that can be described in code or text. It is with the routine that artificial intelligence copes well. Nikolay Ironov appeared because the studio has accumulated enough data, knowledge and ambitions to try to go beyond the boundaries of solving some routine tasks and move from automating routine processes to the territory of creativity. However, the human contribution to the work of a designer is still very great.
The fact is that context plays a very important role in the success or failure of a particular design solution. This is not only the context in which the final work will be placed, conditionally, whether the logo is suitable for a bakery or not, but also the context associated with who creates this work, whether the design customer has enough trust in the author, how then this work served as it is then placed in a new environment in order to generate new and new meanings. Just drawing a perfect sign is not enough, just solving some communication problem well is not enough. When artificial intelligence learns not only to solve applied visual problems, but also to create the right environment around it, when it understands how to further this work, how to present the result of this work, what data needs to be taken in order to use it in the process creating this creative solution – then the algorithm will be able to seriously compete with a professional designer or art director.
As for who is better at creating, a machine or a person, this in itself is not so important. If the problem is solved and solved well, then the machine has done it, a person or several specialists are already details. In any case, the ultimate expertise of professionals still plays a key role.
Artificial intelligence writes texts
Creating texts, working chat bots, answering questions and communicating with people – all this can be done by GPT-3. This is the third generation of Open AI’s language processing and text generation algorithm. It was introduced in 2020, and Sberbank released a Russian-language version… The model, presented in May 2020, was trained by October on a corpus of 600 GB of texts in Russian, from classic literature to news notes and materials from the Pikabu website and other portals. GPT-3, unlike the previous version, not only creates texts, but can also answer a question about the material read, solve a simple mathematical example, write a verse or decipher an anagram, translate a text from one language to another. The team also incorporated GitHub and StackOverFlow data into the tutorials to teach the model to handle and code.
The Russian GPT-3 was led by Sergey Markov. He became interested in artificial intelligence, when he became interested in algorithms for checkers, chess and card games, then he created models for bookmakers, and after that – algorithms for managing problem assets in banks, which included voice assistants for conversations with debtors, and recognition models and speech synthesis.
If we talk about the shortcomings of AI in writing texts, then all such models now have a limited context. Roughly speaking, the model cannot create “War and Peace”, write large-scale, logically connected text. If GPT-3 begins to write a novel and mentions a hero in the first chapter, then after 2 thousand words she will forget about him and will never remember. Therefore, this approach is, of course, suitable in order to generate funny stories, especially postmodern ones. In this sense, the situation has not changed much since 2016, when Alexei Tikhonov and I released “Neural Defense“. The algorithm can generate a short, expressive line, but it cannot tell a story.
Who Really Creates?
Another question is how much machine creativity is in the generated text? When we talk about a music album that an AI released, or a piece of literature that it wrote, we must always remember that machine learning is another tool that humans use to solve a problem. Algorithms are often used specifically for solving routine issues, optimization. It’s just that journalists willingly “forget” about it. For example, when Endel signed with Warner Music, the media reported that AI had taken over the world and the music industry was over. They simply ignore the year and a half that the Endel team had already worked on their technology at that time. Generating the final ten albums is the tip of the iceberg. There is already no creativity – just very heavy shifting of tens of gigabytes of audio stream. People live in a romantic notion that there is a creator, an artist with a capital letter. However, if you look deep enough, even the great artists took apprentices and did not prime the canvas themselves. It’s just that contemporary artists have digital apprentices.
Here it is worth returning to the very concept of creativity. Mozart once wrote the humorous “Guide to How to Compose Waltzes Using Two Dice Without Having the Slightest Knowledge of Music and Composition.” The idea was that any person can throw dice, find notes by the numbers in a special list and write them down on a sheet. The result is a new waltz. Is it possible to say that a dice, which, as a result of the rolls, produces a sequence of notes, solves a creative problem? Will it be creativity or not? And if we take some kind of model that will predict the parameters of this distribution, the dropout of notes based on the previous dropped notes, will it be creativity or not? The answers to these questions are not obvious, especially when you consider that in modern music there are directions of aleatory and serialism, which stand for randomness or for playing according to certain rules, as necessary elements of a piece of music.
So far, the machine is highly dependent on the person in the creation of music, and in the generation of texts, and in the issuance of ready-made design solutions. In addition, creative AI itself is the result of a programmer’s creativity. Creativity is not limited to writing music, painting or text. Coding and training a model is also a creative process.
Find out the detailshow to get a Level Up in skills and salary or an in-demand profession from scratch by taking SkillFactory online courses with a 40% discount and a promotional code HABR, which will give another + 10% discount on training:
- Data Scientist Profession
- Data Analyst profession
- Data Engineering Course
- Other professions and coursesPROFESSION
- Java developer profession
- JAVA QA engineer
- Frontend developer profession
- Profession Ethical hacker
- C ++ developer profession
- Profession Unity Game Developer
- Profession Web developer
- The profession of iOS developer from scratch
- Profession Android developer from scratch
- Machine Learning Course
- Course “Mathematics and Machine Learning for Data Science”
- Course “Machine Learning and Deep Learning”
- Python for Web Development Course
- Course “Algorithms and Data Structures”
- Data Analytics Course
- DevOps course