The expression Generative AI, often abbreviated as GenAI, defines a category of systems designed not only to analyze pre-existing data, but also to produce new digital artifacts such as texts, images, or audio tracks. Unlike traditional AI, which has historically been distinguished by its ability to classify information or formulate predictions, GenAI systems have the ability to create content that, although reflecting the patterns learned during the training phase, is not a copy of elements present in the original dataset; rather, it is a probabilistic and, above all, original reworking of those elements.

Generative artificial intelligence is capable of performing these tasks thanks to advanced neural networks that map deep relationships within the training dataset. In this way, the machine does not memorize information, but rather “understands” its structures and statistical correlations. Thus, when it receives an input, the system is able to construct an original sequence of elements (whether words or image pixels) that have the highest probability of being consistent with the request received.

The most common example of a current application of GenAI is represented by Large Language Models (LLMs). It is significant to consider how, as early as 1956, during the Dartmouth College workshop that historically marks the birth of artificial intelligence, the manipulation of natural language had already been identified as one of the fundamental pillars for emulating human faculties; today, millions of users use generative chatbots on a daily basis, confirming language understanding and generation as one of the most productive areas of the entire field of AI. Another easily understood example is the generation of multimedia content, images, and videos: users can verbally describe a complex visual idea and see it translated into a photorealistic image, an illustration, or a short film.

It is clear that GenAI brings with it quite a few implications on several fronts. First of all, since these systems generate outputs using data drawn from pre-existing human works, a central question arises concerning the nature of originality: artificial intelligence does not create out of nothing, but rather acts as a mirror of collective culture, raising questions about where the software’s own work begins and where the derivation and reproduction of other people’s work ends.

GenAI also makes it possible to adopt working paradigms such as that of co-intelligence (Ethan Mollick, Co-Intelligence: Living and Working with AI, 2024). The term co-intelligence is used to indicate a paradigm of interaction between users and LLMs that involves dynamic dialogue and considerably lowers the barrier to accessing skills that were once specialized. For example, a user without any particular specialized skills can query an LLM to translate a complex financial analysis report into a more understandable and manageable summary, or use generative models to visualize design prototypes that would require weeks of manual modeling. This democratization of technical capabilities shifts the value of human work from execution to critical expertise, thus making the ability to formulate the right questions (prompting) and the ethical validation of the result the new pillars of professional competence.

The advent of GenAI calls for a new form of digital literacy. In an ecosystem in which the production of high-quality content has zero marginal cost, and in which it is becoming increasingly difficult to distinguish whether content is the work of a human being or a machine, technology, although not in itself constituting a threat to human ingenuity, makes it necessary to evolve towards roles in which intuition, ethics, and strategic vision remain irreplaceable (human) prerogatives.

Benedetta Bonomi

(translated by MIRAI)