Artificial intelligence, ethics and conflict

Between peace and war there is no fine line, but a vast grey area, where states engage in what is called strategic competition, using in various combinations the four elements that make up the power of a state: diplomatic, military, economic and informational. It is precisely this last factor, thanks to the pervasiveness of digital technologies, that has taken on unprecedented relevance, for the simple reason that we have become so good at storing any data and thus potentially know everything about everyone.

According to EU forecasts, by 2025 the amount of data managed on a global scale could reach 175 zettabytes (175 followed by 21 zeros). An amount of information that is impossible for a human being to govern, but some promise that what we are denied will be possible for artificial intelligences. However, a question arises: are we willing to accept information or even a clear indication of behaviour without understanding the logic behind it?

In intelligence there are, among others, two fundamental rules and they concern the reliability of the source and the soundness of the information. Now, when I delegate such an analysis to an artificial intelligence, how is it possible to guarantee all this? The problem actually lies in the numbers. ChatGPT, for instance, was originally trained on the basis of 45 terabytes of data, including all of Google Books and all of Wikipedia, but not only. To find patterns, its algorithm boasts 175 billion connections. Thinking of backtracking the path that led an AI to a certain conclusion could take months of work or even be impossible. In this condition, there is what technicians call the opacity of the algorithm, precisely because in many cases one is unable to know on what basis it arrives at certain conclusions. Thus, the paradox arises whereby decision-makers will have to develop a greater tolerance for the risk of being wrong, and this in return for the instantaneous processing of large volumes of data.

To this must be added another not insignificant factor: not all artificial intelligences will think in the same way. I have often found myself describing this phenomenon as 'dual think'. The case of China is emblematic, where the basic requirements for a generative artificial intelligence to operate in China were published last year. Article 4 of these regulations states that it 'must adhere to socialist values and avoid any possibility of use to subvert or incite subversion of the established order'. This is a fact we will have to get used to: artificial intelligences will not and cannot be neutral.

We Europeans will behave no differently from Beijing, with the aggravating circumstance of being so arrogant as to think that our values are the universal ones. We too, therefore, will defend moral and political values and all those truths that, as the philosopher Carlo Sini said, are relatively absolute in a certain place and time. And like us the Arab countries, India and so on. Any artificial intelligence cannot be super partes because it is based on what those who created and trained it believe. This is a divisive technology and that world that one technology has united, i.e. the internet, another may come back to separate, because we will want 'our artificial intelligences'. So we have to start thinking that the saying 'when in Rome, do as the Romans do' will perhaps come terribly back into fashion in five or six years' time.

The question we are facing can be put in these terms: are we ready to base choices, which may condition the destiny of a nation or a continent, on knowledge generated by something whose deep logic we do not fully understand and which will easily be afflicted by cognitive biases so deep that they are invisible to our eyes? I think the time has come when we have to start thinking that the problem is not the ethics to be put into artificial intelligences, but the ethics with which we will use them. Although, I am reasonably certain that this will not make us happy.

(*) Cybersecurity expert .