Procurement, Ai verifies acts, but man has the final say

Contrary to the common narrative that sees artificial intelligence as a brutal substitute for human labour, an increasingly strategic model of cooperation between people and Ai systems is emerging: so-called "human-in-the-loop Ai". In this paradigm, human intervention is an essential component for the training, use and continuous improvement of Ai systems. Therefore, Ai becomes an extension of human capabilities, enhancing them especially in complex scenarios such as medicine, public affairs management, and - in an almost self-referential manner - even Ai systems themselves.

This approach is already part of our everyday life. Think of the more advanced chatbots in customer service: when the model detects an ineffective conversation or the user takes an attitude that shows dissatisfaction in the interaction, it autonomously transfers the dialogue to a human operator. Another emblematic example concerns the training of modern language models (Large language model - Llm). Often, a user is presented with two alternative answers and asked to indicate the best one. This feedback from humans, the heart of so-called Reinforcement learning from human feedback (Rlhf), is first used to refine the quality of interactions and then incorporated into the training data, helping to improve the model for the benefit of all users.

A concrete case of application of this model in which humans and machines collaborate is in the world of procurement. In this system, technology does not issue verdicts, but acts as a powerful decision support tool, operating under the constant supervision of experienced human officials. The core principle is synergy: the Ai takes over the most onerous tasks for human capabilities, such as the high-speed analysis of vast masses of documents and the cross-checking against a complex and ever-updating body of legislation. The system acts as a tireless analyst, scanning each tender document to identify potential inconsistencies, formal anomalies or clauses at risk of litigation.

However, the output of the algorithm is not an unappealable judgement, but a qualified alert, an alarm that draws the professional's attention. At this point, the cycle closes withhuman intervention: it is the official who, drawing on his experience and sensitivity, evaluates the alert generated by the Ai. It is he who investigates the anomaly, interprets the regulatory nuance that the machine might not pick up on, and makes the final decision, taking responsibility for it.

These examples show how the hybrid man-machine model can also become a strategic ally for the public administration. The speed and precision of the Ai system cuts down time, reduces errors and increases the standardisation of procedures. At the same time, the critical intelligence, ethics and contextual judgement of the human being remain the ultimate guarantor of the correctness and legality of the administrative process.