Do you want to know what the sports and weapons of the future will be? Read a report as it would have been written by Oriana Fallaci, Ernest Hemingway, George Orwell with their own hand? An alternative history in which Aldo Moro survives being kidnapped by the Red Brigades? Know what Napoleon Bonaparte would advise Vladimir Putin? See a robot’s answers to Tinder approaches? These are just some of the proposals you’ll find inside our new magazine made with Artificial Intelligence, the first of its kind. The advent of systems for automatic text production is not, in our opinion, the death of journalism, but a tool capable of offering new potentialities to the information sector, provided that we get to know its strengths and limits. That’s why the Zeta editorial staff decided to create a special issue of its magazine, written, illustrated and titled entirely by ChatGPT, the AI tool launched by OpenAI institute, a start-up founded in 2015 by Elon Musk and Sam Altman. AI can produce texts, answer questions and even write codes. The key word of the publication could not be anything other than “experiment”, where experimenting means facing the novelty with an open and critical attitude. A bet that also included the possibility of failure, of offering a banal magazine, not up to the standards of those conceived by human minds. By studying, tinkering and even playing with the machine, we managed to create a magazine full of original, funny, even surreal pieces, being ourselves surprised by how the chat knows how to translate our ideas into practice.
ChatGPT is a deep learning-based language model, a chatbot capable of producing different types of texts based on the inputs received from the user. To write the articles in the magazine, we proceeded with questions and requests entered in the chat, each following their own ideas and sensitivity. If someone thinks that to write an article with Artificial Intelligence it is enough to say to the system “Write a piece about…”, they will be disappointed. With a request that is too generic, the text will be flat, trivial, and it will be difficult to get anything worth reading out of it. Sometimes it is necessary to ask the chat more questions and then ask it to synthesize the information generated in the meantime. Without impulses from the journalist, the machine remains inert, so finding an original idea remains crucial to fully exploit its potential. Each article carries the double signature of the chat and the curator. The images accompanying the articles were also created through Artificial Intelligence tools, including Dall-E, produced by OpenAI itself, and Midjourney.
Like all language models, ChatGPT is trained on a huge amount of texts available on the web. The machine “understands” Italian and other languages without problem, but being developed in the United States its generative capacities are far greater in English. To obtain the most complete result, we generated the pieces in English and then asked the chat to translate them in Italian, so that not even the translation would be “tainted” by human hands.
The other limit of ChatGPT is not having access to the internet. Since its knowledge stops at the end of 2021, it is therefore very effective on analysis articles, in-depth research and parallel realities, while it cannot be used for current news pieces unless the journalist explicitly provides the information, such as in the article on Russian strategies for the siege of Bakhmut in eastern Ukraine.
After exploring the potential of the chat and encountering its limits, our experiment can be considered successful. But as always in laboratories, every success is just the basis for new questions. Soon the automatic language GPT-4 will be released, hundreds of times more powerful than the current version, GPT-3, and capable of generating photos and videos, as well as realistic texts. As for Zeta, we will soon also produce video and radio services.
Authoritative experts fear that AI will replace journalists. We, after our experience, disagree. Of its own saying, at least so far, the machine’s articles may «lack the creativity, nuances, and perspective that human writers bring to the table», in addition to the risk that some passages might be imprecise or incorrect. There is no choice between accepting progress or rejecting it, there is only a crossroads between learning to use new technologies or being overwhelmed by them. The synergy between Man and Machine remains central, but the campaign to make sure that Artificial Intelligence is always a tool in our hands and not an – invincible – enemy is yet to be fought.
Article translated in English by ChatGPT
Read our ChatGPT magazine here: Esperimento • Zeta Numero 0 | Gennaio 2023
