Connect with us

Hi, what are you looking for?


How does artificial intelligence make pictures? – I like science

How does artificial intelligence make pictures?  - I like science

Does AI really create images? Is she imaginative and creative? Recently, artificial intelligence won a prestigious art competition, which is why other artists – living ones – are so outraged. In this material, I will tell you what the creative process of artificial intelligence looks like.

Artificial intelligence and creativity

At least in my bubbles, every now and then someone uploads AI-generated images. On my Instagram you can see a gallery with some of them. Some are annoying, others are interesting. How it works?

Computers have always been better than humans in many areas. What I mean here is math, weather forecasting (i.e. math) and playing different games (i.e. math in total). There is also stock market trading, but that’s basically my account as well. Computers are also better at driving cars or piloting airplanes – because it’s all math. However, we still did not see any creative entities in them that could compete with us in music or painting competitions. Or maybe our creativity is also mathematics?

Is artificial intelligence more creative than humans?

Let’s take two pictures as an example.

The painting “Warsaw” by Canaletto. Source:
Computer image created by Jeremy Kraybel. source:

The first is Warsaw painted by Canaletto, and the last is a painting by artist and programmer Jeremy Kreibel.

Who do we say is an art? About both or one of them? Canaletto (the photorealist of Old Warsaw) is widely regarded as an artist. And the other picture? It’s a computer-made image. The programmer first trained the model on the selected images to get his own style, and later transferred the design to the canvas thanks to the robotic arm. The computer has learned that it must perform strong brushstrokes to reproduce the most important elements of a character’s face.

Algorithm Trainers Race

Is working computer algorithms more creative than drawing the royal castle in Warsaw? Today, machine learning algorithms create images based on just a few words.

It can be said that there is a current race between the creators and trainers of these algorithms. The race in which the algorithm will be better, more creative, or which will create more original graphics or animation. Sometimes this is also a race in which the algorithm will create completely unreal characters or objects or complete facts.

For example, Markos Kay used algorithms in his project “Biocomputer” and created animated characters based on them. How is that possible?

How is artificial intelligence created?

There are many examples generated by image algorithms, but how do they actually work? How can an algorithm actually create something that is not a copy of something that already exists?

Well, it’s possible because three things are intertwined.

  • First of all – high computing power. What was hardly possible for the supercomputers created by the best research centers, anyone can get today. Computing power has become a commodity that can be bought freely by the minute.
  • Second – computers are beginning to understand us. Until recently, they only understood zeros and ones. Since we wanted them to do something, we had to learn their language – learn programming. Today, computers understand our language more and more.
  • Third, we have created algorithms that can learn, train, and acquire the skills needed to recognize and interpret images.

A fast computer, the ability to communicate easily, and the ability to recognize images came together. And since the algorithm can recognize images, it can also create them. However, the program does not have physical experiences, such as petting a cat or scratching a dog behind the ear. For him, a cat or a dog is an array of pixels. So when the algorithm draws a cat, it will not draw its favorite childhood pet, but a mixture of the ones it has seen. Never the same, always different.

The situation with the cat is simple. What about the city of the future? The algorithm knows what cities look like today. He also knows, if trained to do so, what features we associate with the future. So the city of the future will be a mixture of connections between the city and the future. Each of these elements will be an overlay of all the pictures of the cities where the AI ​​was trained and everything that was presented to it during the training as speech or pictures of the novelty.

What determines the appearance of a computer-generated image?

The working effects of the algorithm are directly dependent on the training data. The more words or phrases we provide, the more crazy pictures we will see. He is crazy in his inhuman way. That is why these images are so interesting, as if they come from a different reality, because they were not created in the human brain, but in the black box of artificial intelligence.

This will also interest you: Can a computer read our thoughts? Pioneering research by scientists from Stanford University

We can put another layer on top of all this – the pattern. Each of the paintings created by artificial intelligence can mimic the style of painters who created in different eras Impressionists, Expressionists, Modernists, Classical or Cubist paintings. How can we learn this algorithm? Just like distinguishing a cat from a dog, i.e. showing pictures of it.

You can also go ahead and train the algorithm to draw in the style of a particular artist. If the algorithm is properly trained, then the art historian, when considering such a work, will recognize the characteristic style of a particular artist in it.

The problem of composing images generated by artificial intelligence

This is where a fundamentally unsolvable problem begins – who is the author of the AI-generated images? The author of the algorithm, the trainer, the author of the text query, or maybe the algorithm itself? Some might say that the problem is blown out because the algorithm only copies what it has seen before. Anyway, it’s not that simple.

First, the images, and sometimes the entire complex animation that the algorithm broadcasts, are not a copy of what I experienced. Second, you can ask yourself – what is creativity in humans? How much inspiration is in a piece of music—whether conscious or not—from the sounds of a city, river, forest, or bird, and how much creativity is the author? Did Vincent Van Gogh copy a vase with sunflowers? And Leonardo da Vinci opposed Gioconda or Mona Lisa?

This will also interest you: Incubated by robots – will humanoid robots replace humans?

Look at the painting I showed at the beginning – Warsaw by Canaletto. What is the original here and what was copied? If Cantaleto lived and painted in the times when people had smartphones and took pictures with them, wouldn’t they get the same pictures, or maybe even identical pictures?

Right from the start, the human mind has been training to better understand the world. Every second he receives multiple signals from the world that may strengthen or weaken his reactions.

Steve Jobs said, “Creativity is about bringing together things that haven’t been put together yet.” Humanity has been doing this since the dawn of history, building civilization on the shoulders of previous generations. We are also developing machine learning in a similar direction. Here, however, the pace of learning and combining items that no one has previously integrated is definitely faster than it is with us. So how do we differ from algorithms? This is a topic for different materials.

Audiobook Superhero Academy

Thomas Ruschick

I am a physicist by education and a science journalist by profession. I’ve been creating the Nauka brand for years. I like it, I also run the research department of the weekly Gość Niedzielny and collaborate with several media and editorial offices. I am a happy father to twins: Zozia and Janek and Anya’s husband.

See also  Who is the astronomer? I like science
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

You May Also Like


Another person familiar with Sony’s plans confirms the killer news for PS5 owners. Horizon Forbidden West is not expected to appear on the market...


The British government has canceled a recommendation to avoid travel to and from the eight regions most affected by the Indian coronavirus variant altogether....


It looks like Tesla wants to use lasers instead of standard wipers in the future. I applied for a patent for this solution. The...


Farmer reveals government plans for CPK The farmer who issued the documents is a member of the social council of the CPK. As he...