The launch of a book written by GPT-3, What Makes Us Human: An Artificial Intelligence Answers Life’s Biggest Questions, became a surprise at the spring non-fiction fair. And yet, why a surprise? If neural networks can paint, write press releases or theses, why can’t they go as far as writing a novel?
How much interest can these books spark? How are literature and science changing in the era of artificial intelligence? Can further development of neural networks drop the curtain on human development? We addressed these and other questions to Valentin Klimov, Director of the Institute of Intellectual Cybernetic Systems at the National Research Nuclear University MEPhI (Moscow Engineering Physics Institute), and will share with you most curious observations and crucial conclusions.
Neural network can write like Leo Tolstoy, but cannot create War and Peace
If you keep trying to find the meaning of life, you will never find it because it comes from within, from inside us, – this is what artificial intelligence teaches us. It also says that the ideal person is not paralyzed by social fears, greed or lust for power, does not fear changes and accepts things, people and the world as they are. But first things first…
The book What Makes Us Human: An Artificial Intelligence Answers Life’s Biggest Questions was written by GPT-3, one of the most advanced language prediction models released by OpenAI, headed by Elon Musk. The algorithm of this literary work is very simple. The main texts of world culture, from the Bible and Tao Te Ching to the poetry of Rumi and Sappho, as well as works by philosophers and writers were downloaded into the neural network. Then, the author Iain S. Thomas and innovative scientist Jasmine Wang asked GPT-3 the eternal questions of happiness and unhappiness, good and evil, hatred and love. The neural network gave ambivalent answers, which, according to the authors, can change our perception of ourselves.
In fact, the GPT-3’s book will unlikely become a revelation even for inexperienced readers. The artificial intelligence’s contemplations are vague, and it seems that of all gems of the world literature it picked and loved only Paolo Coelho and completely disregarded Nietzsche, Schopenhauer and Heidegger.
Our entire life has been already written, even our smallest deeds. All of our actions have a meaning. Why do we waste so much time wondering why there is evil in the world instead of simply doing good? – says the book. These quotes are not the pinnacle of philosophical thought. But the presentation of the book gathered such a big audience that even Yevgeny Vodolazkin would be jealous. And this is frightening.
Neural networks develop fast. What if the texts written by GPT-4 or GPT-5 are more interesting than Agatha Christie’s detective stories, scarier than Stephen King’s stories and deeper than Leo Tolstoy’s novels?
“Neural networks are perfectly capable of summarizing information and highlighting hidden patterns. They use words correctly and create quality content. They can ‘read’ as many books in one day as a person cannot read in their entire life. But artificial intelligence cannot create anything new. It can easily create a painting in the impressionist, modernist or realist style, but will never create its own style. AI can write like Leo Tolstoy, but is not able to conceive a new story, a new plot. Machine cannot make any literature, artistic or scientific discoveries,” Valentin Klimov says.
However, according to the expert, artificial intelligence can give witty answers, write good texts on given topics, and generate quality illustrations. And it would be impossible to tell who created what, a human or machine. It was proven by the Turing Test and life itself. There are numerous examples.
AI to compete with journalists and help physicists
German photographer Boris Eldagsen has recently won the prestigious Sony World Photography Awards and refused it for ideological reasons. Eldagsen admitted that his work was generated by artificial intelligence. By submitting the image for the competition, the photographer wanted to draw the attention of the public to the integration of AI into art, and at the same time see whether the jury would be able to see the fake, and if it is ready to tell a real photo artist from a charlatan with good knowledge of advanced technology. It turned out that the jury was not ready. And this is not surprising, the AI-generated black and white photo looks very realistic. It features two women. One of them looks tired and pensive, the other is apprehensive and energetic. The image is appealing and interesting to look at.
Eldagsen’s concern is understandable. If a neural network defeated professional photographers at a major international competition so easily, what if artificial intelligence takes all our jobs?
Valentin Klimov believes that this prospect is real and in the nearest future, the labor market might change significantly.
“The machine picks the necessary content very quickly; it can process and analyze huge amount of data. It is pointless to compete with it in terms of speed, we cannot beat artificial intelligence. In some Western media, their daily newswires are updated by neural networks. And they are very good at it: they write short news stories in a fast and quality way. The algorithms of the stock markets have also changed: an entire army of brokers has been replaced with a neural network, which performs all short transactions. Artificial intelligence can replace low-skilled lawyers, designers and journalists. It can create good images, texts, music and interface. In addition, unlike humans, AI cannot get tired, get sick and never gets old. In the nearest future, we will see a global transformation of some 30% professions,” Valentin Klimov believes.
We can already see AI models performing functions of a personal assistant, a secretary, or an assistant to the director; they find and process information, structure texts, draw up documents, and create presentations. Yet, the expert claims, we should not fear the competition: AI will take over routine, mundane tasks, saving time for us to engage in creative activities and work. Changes in the labor market will be obvious not only to designers, journalists, photographers and secretaries, but also to physicists.
“Modern physics deals with a vast amount of complex units and devices as well as huge data arrays. Sometimes, finding a tiny rational grain requires filtering out 99.9% of redundant, unnecessary information. This cannot be done manually; neural networks are required to help us,” the expert says.
Help is a good thing. But will AI become a source of disinformation, infiltrating the world with fake news, false conclusions and absurd ideas? Could it outsmart humans, rendering them useless and redundant?
Neural networks should be regulated, not banned
A possible human-AI confrontation is being discussed by both general public and prominent representatives of the technology industry. When Tesla and SpaceX CEO Elon Musk, Apple co-founder Steve Wozniak and other high-profile entrepreneurs wrote an open letter calling for a ban on creating powerful AI, many supported this idea. We, as humans, tend to get offended by a ‘senseless piece of machine’ beating a grandmaster in a chess game. Musk warned that creating more powerful and advanced neural networks could lead to job losses and even has the potential of civilization destruction. Microsoft founder Bill Gates spoke out to advocate AI, claiming that the future belongs to neural networks are we cannot restrain their development.
Programmers compare GPT-3 to a five-year-old who has read 100,000 books. Meanwhile, GPT-4 and GPT-5 will be more advanced. Also, there are plenty of other neural networks apart from GPT whose list will take as many as several pages – and all of them will become smarter and more advanced as well.
“Banning neural networks makes no sense; we should learn working with them. We have already entered the era of information explosion, with all smart devices that surround us, such as telephones, kettles, watches and refrigerators. We need a smart assistant able to handle vast amounts of information. Of course, there is a risk of AI turning into a sort of ‘crutch’ that a person always needs to walk; neural networks handing necessary information to us on a platter may diminish our analytical skills. It would be proper to develop a document to regulate the use of AI – just similar to traffic regulations that were necessary to develop following the invention of cars,” Valentin Klimov explains.
The expert believes that such regulations are particularly required in the field of education, where AI will have a significant impact – and this is not about the use of neural networks for writing essays, theses and end-of-year papers. Such functions as fast reproduction of any type of information, its compilation, combination, and logical processing will very soon become ubiquitous. The educational sphere should not be limited only to school and university students acquiring a certain amount of knowledge as all necessary information can be easily obtained and even presented as a coherent text through the use of AI models. What is becoming really essential is the ability to identify and solve new issues in your professional field, meaning that educational programs should be dominated by project and research activities, with students learning to set goals, interact with other people, discuss results, and correct mistakes.
Valentin Klimov mentions three major trends in the development of AI models:
- the number of neural networks is rapidly increasing, with the so-called large language models (LLMs) as the most numerous group;
- neural networks are becoming multimodal; they can handle not only texts, but also drawings, audio recordings and videos;
- low-resource models are also becoming available that can work autonomously based on simple processors, with reduced technical requirements for the use of AI.
According to the expert, built-in AI elements will be applied in all more or less complex technical devices in the foreseeable future, such as industrial gear, transport vehicles, unmanned vehicles, and household appliances. We must create, develop and utilize neural networks – otherwise, Russia will fail to achieve its technological sovereignty.
Neural networks will be advancing; they will continue to surprise people. It’s enough to say that they have already surprised us. What is next? Responding to the question posed by the compilers of the first AI-written book, GPT-3 suddenly abandoned its verbose good-natured mask and spoke out in a gloom, laconic and decisive manner: “Where to next? The end of meanness. The end of hopelessness. The end of loneliness. The end of scarcity. The end of fear. The end of hatred. The end of guilt. The end.”
By Natalya Sysoeva