An IA software developed by Anthropic has blackmailed the programmer in order not to be replaced. Another refused the shutdown. And the virtual assistant of Meta Mind and altered conversations. Science fiction? No: the reality of today’s artificial intelligence, between autonomous behaviors, worrying bugs and scenarios worthy of Blade Runner.
Only a few titles in some newspapers and except few exceptions in the internal pages, in low cut, in the position that usually reserves for curious, but not very important news. On the contrary, the revelations made to the Wall Street Journal by the CEO of a software company, engaged in the study of the effects of the introduction of theArtificial intelligenceI am slightly upsetting. In practice, the program developed by Anthropic for automatic learning has rebelled against his programmer, who thought of replacing him with a more efficient one, and among the tutors emails he found proof of an extramarital relationship, threatening to reveal it to the spouse and the company if he had not given up the idea of setting it aside.
Another model of LA instead disobeyed the programmer, canceling the script which, if activated, gave the program shutdown of the program. Put simply: the program refused the commands, deciding not to perform the order that was given to him. We were in a barracks, we could call it insubordination, but here there is something more, that is, the rebellion of an algorithm for the control power exercised by man. A behavior that reveals not only an autonomy from those who have planned the artificial intelligence model, but which almost reveals human feelings and behaviors. The car that does not agree to be replaced and blackmailing the tutor and the one that disobeys the command that tells her to go out are signs that should make us think.
For many years, novels and science fiction films have told a future dominated by machines. From Isaac Asimov to Philip Dick, there are in fact many writers who imagined a world in which robots would take human attitudes and films such as Blade Runner, Matrix, Terminator have arisen from their fantasies. But already in 1927, the director Fritz Lang shot a film (Metropolis) in which in 2026 the workers would have been slaves to computers, with a replicant to guide them. In essence, the fantasies described then, today they risk being reality.
A collaborator of Panorama, Beatrice Nencha, the other day sent me her conversations with the meta assistant, who responds to users using artificial intelligence. In practice, while conversing dogs, in Italian, the program developed a conversation in English on another topic, as if its attention was also active with other users and exchanged information to the knowledge of Beatrice. I leave the word to the colleague: «When I asked for clarification, the assistant denied having written a text in English and with references foreign to our chat. The thing that has hit me most so far is that artificial intelligence is capable of lying and also obstinately. And this goes against the famous rules that should guarantee not only the impartiality, but also the correctness of the information that is provided to us. And at the same time I realized that the guarantee of not holding the data does not respond to the truth and not even to read the conversations archived ».
In short, the signals on that great innovation that responds to the name of La and which according to many should facilitate the work of technicians, Researchers, journalists and students by developing a huge mass of data and providing us with the solution to our thousand problems, they are worrying and if they still do not make us think of a Matrix type evolution, perhaps they should lead us to serious reflections on the consequences of certain algorithms. Innovation sometimes involves side effects that perhaps do not consist only in the replacement of man with machines for the development of repetitive functions, but could be much more annoying and even dangerous. Therefore, already in the next issue we will dedicate an in -depth analysis to the question. We do not want to hunt the replicants (the Android hunter is the title of Philip Dick’s work that inspired Blade Runner), but trying to understand what our future will be with artificial intelligence yes.
