The challenge of AI to criminal law, which will have to rethink the classic schemes

As demonstrated by the debate on the potential, limits and risks of ChatGPT, artificial intelligence, permeating ever more extensive sectors of everyday life, determines the emergence of delicate legal issues, which have also begun to affect criminal law, historically more resistant to innovations technological. Confirmation can be drawn from two recent provisions of a supranational nature: the European Parliament resolution of last October on the use of AI by the police and judicial authorities in the criminal field and the the so-called AI Act, a proposal for a European regulation which establishes harmonized rules on artificial intelligence.

One of the most controversial points concerns the possible involvement of an AI system as the perpetrator of a crime. There is no shortage of examples: from drones capable of killing on urban streets, to driverless cars involved in causing accidents to the detriment of things or people, up to software that performs increasingly advanced tasks in collaboration or even in place of humans. and which can sometimes interfere negatively with human conduct. Case studies stimulate a question, which in turn calls into question more general questions: in similar episodes, whenever a crime is found, except in cases in which the programmer, the producer or the user of the software (that is, when the system represents nothing more than a tool in the hands of man through which, due to the enormous potential that can be used for criminal purposes, the crime can be carried out), Can a latest generation system really take on the role of active subject of the crime? Although there are those who, relying on the fact that AI systems, at least the more advanced and sophisticated ones, are capable of acting autonomously, of assuming and possibly implementing their own decisions, which were not foreseeable by their programmers, offer an affirmative answer ( hypothesizing, in a parallelism with the discipline of the liability of entities, a planning fault), the main perplexities, at the moment, revolve around the difficulty of recognizing, on the part of a machine, the traditional requirement of guilt.

It is, as is known, the most intense expression of the perpetrator’s subjective involvement in the fact, the presence of which entails the possibility of making a reproach and which presupposes the recognition, on the part of the subject, of imputability, willful misconduct or fault, knowledge (or at least knowability) of the violated criminal law and absence of causes of exclusion of guilt. Well, one wonders: can these elements, originally conceived and traditionally referred only to humans, be ascribed to a machine? Can we hypothesize a robotic guilt, in the absence, among other things, of certain requisites such as self-awareness, free will, moral autonomy? Can we speak of “ability to understand and want” in relation to a software or configure a fault or even a fraud on the part of it? Again: what sanctions can be applied to these systems and above all, if necessary, will they be able to fulfill those orientation and re-educational functions that the Constitution recognizes for punishment? And at the end of what type of process?

When questions overcome certainties, the only certainty is that we need to move with caution. This, however, does not justify underestimating the issue: if criminal law wants to continue to fulfill its role, it must put itself in a position to consciously face the complex but unavoidable challenges of technological development, rethinking classic schemes and proposing, without distorting consolidated principles of guarantee, new models of indictment and paradigms of protection open to the thrust of a future that is knocking on the door.

Christian Cupelli, Professor of Criminal Law at the Tor Vergata University of Rome