Artist vs. AI: The Future Has Already Arrived
In recent months, disturbing news has been coming from beta users of the program DALL-E for illustrators. With a simple text description, these users can ask the neural network to make a photorealistic picture of, for example, a polar bear playing the bass guitar in just a few seconds. Or a robot painted in the style of Picasso.
At first glance, it is quite difficult to find fault with the execution of these pictures. This causes a feeling of discomfort.
In the article, we will tell you what this artificial intelligence program DALL-E is and what it can do.
Have you heard about NFTs? It’s a hyped topic and activity. Learn more here.
What is DALL-E?
DALL-E is a new artificial intelligence system from the OpenAI research laboratory, which is widely known in not-so-narrow circles.
DALL-E accepts a simple text description, such as “koala scores in basketball,” and turns it into a photorealistic image that never existed.
This option of photorealism appeared with the second version of DALL-E, which was introduced in April this year. The original neural network was rolled out in a closed beta test back in January 2021. It is extremely difficult to get into it, and there were not so many testimonies about it.
With the second version of the DALL-E neural network, you can do realistic photo editing and retouching. It can replace parts of the image based on the same simple natural human language.
Do you want to replace the cute dog in the chair with a cat? You’re welcome! It is enough to enter two words, and DALL-E will generate a cat instead of a dog.
The second version of this AI has made significant progress. It gets pictures in higher resolution, with a more adequate understanding and new editing capabilities, as in the case of a real dog and a cat from the fantasies of a neural network. You can even feed it a real image and get its variations in different styles and with different perspectives.
DALL-E tech issues
Like any neural network, DALL-E underwent long training on a giant array of data. In this case, it was photos and their text description. It saw thousands and thousands of koalas of different colors, and sizes, at different angles, and in different poses.
The system has not only learned to recognize objects but also to determine the relationship between them. It understands that you can ride a motorcycle, it has seen thousands of pictures of a person sitting on a motorcycle and holding the handle bars.
Therefore, it will not be a surprise or a big difficulty for the neural network to imagine a koala riding on a bike. And not one image, but a thousand versions in different styles, perspectives, viewing angles, poses, etc.
Naturally, this approach to learning has its limitations. An incorrect caption under the image leads to errors in the perception of the program. If it comes across an airplane labelled as a car, then in the future there will be errors in recognition that are noticeable to us but invisible to the AI.
“The more complex objects or concepts the system learns, the more inaccuracies may arise in the course of its subsequent work.”
Think of Harold in the image below — hiding his pain. In the photo he is smiling, but we actually know (or guess) by subtle signs that there is something more behind this smile than joy. Such subtleties in various objects and groups of objects carry real difficulties for AI.
The developers of this AI see three main ways to apply it in practice.
Firstly, it will help people who could not do so before to visually express themselves.
Secondly, the images generated by the system can say a lot about whether the system understands us or just repeats what it has been taught.
Third, DALL-E will help people learn how advanced artificial intelligence systems see and understand our world. This point is called critical in the development of useful and safe AI. As in the example with Harold, human emotions can be beyond the power of even many people to grasp.
Did you know that AI facial recognition system helps to search for criminals? Read more here.
Is DALL-E a threat to illustrators?
DALL-E is more likely a threat to illustrators. Right from the first glance, there is this feeling of a future commercial product for some publishing houses. DALL-E seems to have an endless ability to translate a couple of lines of text into a whole set of pictures for every taste.
If you ask it to use “plasticine” style instead of “graffiti,” then artificial intelligence will cope with this on the fly and much faster than a person will, and then implement it in a digital picture. Do you need an astronaut on a horse in a photorealistic style? Or maybe something simpler. For example, a pencil drawing?
All this is based on 40,000 years of human and art history. All this is done so fast and in such quantities that all the people of the planet together could not do it.
Artificial intelligence is already making many human professions obsolete in the form they were ten years ago. AI is on its way to replacing pilots, drivers, cashiers, and even little programmers.
One of the last shelters in which a person is 100% sure of his irreplaceability is the art of self-expression — verbal or visual. Yet, verbal expression is already being harassed by the GPT-3 system from the same OpenAI group. It can compose letters, invent fairy tales, create narratives for video games, and write news notes. Now it’s the turn of commercial artists, photographers, and illustrators to feel this burning pain.
First of all, various stock agencies are likely to worry. They sell pictures and illustrations made by people. Their entire business is tied up in unique images, to which the client gets instant access for just a couple of tens of dollars and uses at their discretion in their commercial projects. Huge databases of illustrations for every taste, created by an army of photographers and artists.
Secondly, tools such as image editors (Photoshop, etc.) may also lose their uniqueness. At least, the dependence on them can be reduced. And that’s a bunch of programmers, analysts, designers who are working on such tools.
Yes, DALL-E is not perfect at all. If you look closely at its art, you can see the flaws. And, as a rule, its developers show us only the most impressive and successful. But still, this thing is continuously learning and will only improve.
AI vs. Humanity
All these partial solutions in the field of artificial intelligence and robotics are slightly alarming. Machines are already able to work faster in factories. They know how to drive a car, are able to clone people’s emotions on their rubber faces, and speak a clear literary language. They know how to write news and code.
Breakthroughs in robotics depend not only on more nimble mechanical legs and arms, more observant eyes and ears, but also on human-like artificial intelligence. Powerful AI systems intersect, transforming the economy, as happened with steam engines, electricity, and the internet.
In the case of AI, the replication of human capabilities leads to a decrease in the need for human labor, the restructuring of economic and political power. AI can be used to expand human labor or automate it.
The future of digital people is here.
When AI expands human capabilities, allowing people to do things they have never been able to do before, humans and machines become complementary to each other. People remain indispensable for value creation and retain their bargaining power in labor markets and political decision-making.
Conversely, when AI reproduces and automates existing human capabilities, machines become the best substitute for human labor, and workers lose economic and political bargaining positions.
Entrepreneurs who have the opportunity to replace a person with AI are unlikely to want to contact an employee who is characterized by illnesses, vacations, demands and trade unions. Those who control technology can concentrate power and wealth without paying attention to the voices of the discontented.
The trouble is that our ambition and uncompromising nature dictate the creation of a strong artificial intelligence — the achievement of what is called “human-like AI.” For many researchers, this remains the ultimate goal. Many of the brightest and strongest minds of mankind are looking for ways to fully automate human labor.
What will come of it? We will follow the developments.
The future of the web and the internet are here.