ChatGPT, the artificial intelligence that copy and paste. And to reveal it and the enemy Google-Corriere.it

“There is nothing more unpublished than what has already been published.” ChatGPT, we suspect, must have read Prague cemetery by Umberto Eco and his famous provocative passage. When in doubt, we asked him: do you always respect copyright when you generate texts? Answer of the chat that allows you to dialogue with a neural network that has just received 10 billion dollars of investment from Microsoft: «As a language model trained by OpenAI, I generate texts based on the processing of large amounts of textual data present on the web». Let’s call it sincerity. ChatGPT generate, invent, copy and paste. But it does not mention (moreover the copyright we Italians invented it, in late fifteenth-century Venice, with the privilege of printing, so we should defend it).

ChatGPT’s competitor, the “obsolete” Google spider, helps to find out: just put the “generated” texts to the test (by copying them to Google, the original documents often emerge) to discover that we are not dealing with the miracle of ” I think therefore I am”. The question is important, not so much, as we have already had the opportunity to write, for the development of artificial intelligence. The discovery of these childish flaws does not detract from the incredible evolution of a system capable of understanding the natural language of man. But why — while everyone was worrying about how young people would use ChatGPT to no longer do their math homework or Latin versions on the Of beautiful Gallic — in some papers, as the scientific journal par excellence has pointed out natures, ChatGPT has already appeared as co-author alongside flesh-and-blood researchers. First lesson: as usual we are prejudiced against our children having never learned the lesson that the father of a certain Charles Darwin wanted to give us when he said: “My son will never do anything good in life”. Keep silent about the children, always.

(Here the episode of Massimo Sideri’s «Geni Invisibili» podcast dedicated to artificial intelligence with Maria Chiara Carrozza of the Cnr and Giorgio Metta of the Iit, here the complete series)

But going back to the papers and who knows how many other professional uses, the economic problem is who certifies the originality of what the chatbot with natural language developed by OpenAI adds? When it “simulates” a style (answering the question “write like Shakespeare or Dante would have done”) ChatGPT seems to give its best. Which is not writing like Dante, of course. And not even come close. But it’s aping the great bard like we all did at school, maybe after a beer. That’s what the human mind does: copying, replicating, with generally and drastically lower quality than the original.

With some neurologically exceptional exceptions such as the construction of metaphors, the use of irony and, again, the art of creating parody: «L’inferno di Topolino», by Guido Martina with drawings by Angelo Bioletto («Io son nomato Pippo and I am a poet”) is not Dante, but he is brilliant. Here, too, intelligent cinematography, not Terminator-style cinematography, has managed to describe certain developments well and in advance: in «Interstellar», the film by Christopher Nolan in whose screenplay the Nobel prize-winning physicist Kip Thorne participated, the robots have “irony” emulators that can be set by man. ChatGPT filters tend to work exactly like this: they don’t understand the jokes, but they look for the traces and emulate the posturing in the reply.

As with the term “artificial intelligence” — introduced in the 1950s by Alan Turing laureate John McCarthy — here too there is a misunderstanding with the term “generative.” It should not be understood in a Cartesian way. Ai may be the best worker bee of our life. She underlines this herself: «In the event that a copyrighted text is accidentally used, we would be happy to remove it if reported. As a language model, I can generate texts in different formats, including scientific articles. However, being a machine I lack the ability to understand and interpret scientific data and do original research. Therefore, I would not be able to be a primary author of a scientific paper. However, they can be used to generate draft articles, assist in the writing and formatting of papers, or to generate parts of the text such as an abstract or introduction.

As in all things, however, also for ChatGPT the devil is in the details, which risk escaping behind the humble, almost greasy approach of our little electronic monkey. Time, for example, highlighted how OpenAI was able to moderate the deviant excesses of version 3 of its program by calling on Sama, another Silicon Valley firm, which employs workers in Kenya, Uganda and India to clean up data for big tech from Google to Meta and Microsoft. Using tens of thousands of dark web documents about sexual assaults, torture, murders, and having them tagged by humans paid $1-2 an hour — a terrible experience — these documents were made readable by ChatGPT and used to prevent intervened on these issues.

Man, always man, with his miseries and his abilities, is the basis of the surprising images that the faceted mirror of the most powerful Chatbot of the moment sends us back. And that’s exactly the point. This is a technology created by programmers, archivists, cleaners, trainers and, above all, by an enormous number of people who have left traces of their intellectual, emotional, creative and experiential activity on the web.

All of this, placed inside a sophisticated electronic container, takes on seemingly new forms and allows each of us to come into contact with traces that come from all the others, in a process of continuous, rapid evolution. It is more a language than an artificial intelligence, the human realization of the anthill effect, an area in which decisions are made in common, exchanging pheromones and information in a way impossible to trace in detail. In the case of the anthill, Mother Nature and Darwin’s selection supervise the meaning of common action: the survival of the ant species. In the case of Chatbots, however, the superintendent is also human, even if he tries not to be seen too much by hiding behind the needs of the market: the goddess of war Durga had 10 arms, but only one head. Who controls the controller?

And above all, who will control it once our little monkey has spread pervasively in our societies? Chat denies having a conflict of interest and shifts this responsibility onto its programmers. This time he’s right.

Subscribe to the newsletter of The Economy

Whatever it Takes by Federico Fubini

The challenges for the economy and markets in an unstable world

Europe Matters by James Fontanella-Khan

Italy and Europe seen from America

And don’t forget the newsletters
The Economy Opinions and the Economy 6 pm