AI-Crimes in the Silicon Age

It is foreseeable that AI-crimes may create blurred boundaries and legal uncertainties. The following areas display the highest bent to exposure to AI-crimes: cartel offenses, insider dealing, market manipulation, price fixing, collusion, trafficking, selling, buying and possessing banned drugs, harassment, stalking, torture and information extraction, theft, robbery, forgery, and personation. Furthermore, “data deletion” has surfaced in countless juridical discussions. While it may appear to be a straightforward process, data deletion poses many practical problems in actual machine-learning environments and can border on the edge of impossibility.

At first glance, this may all seem quite farfetched. However, in recent days one AI product faced acrimonious and vituperative requests to be withdrawn. The AI tool on a specific device invented false claims and misrepresented the facts about the private lives of several notable people. The technology was described as being “out of control.” This has raised concerns that generative AI services are still too immature to produce reliable public information.

It could be the case that artificial agents or otherwise, engage in criminal acts or omissions without sufficient concurrence of liability for the offense to constitute a specific criminal offense. The conduct proscribed by a certain crime must be done “voluntarily” to establish the “actus reus.” But voluntariness turns on notions of conscience, will, and control which may never be met.

The second condition or a “guilty mind” (mens rea) must meet different thresholds of mental states for different crimes. But as AI artefacts become more autonomous and delegated authority becomes more pervasive, the chance of a criminal act or omission being decoupled from a mental state becomes more acute. This splintering is likely when a bot or any artificial agent commits the “actus reus” and a human manifests the “mens rea.”

This highlights troublesome liability concerns. AI-crimes may undermine existing liability models, thereby threatening the dissuasive and redressing power of existing legislation. Present liability models may not be adequate to deal effectively with the future role of AI in criminal conduct. Ultimately, the limitations of existing liability models can undermine the certainty of the law.

Certain acts or omissions in the Age of AI constituting an offense punishable under law would probably not happen but for the use and support of AI. This gives rise to AI-crimes as a separate phenomenon, one that may be related to but different from cybercrime, and one that is uniquely problematic, and a distinct criminal phenomenon compared with crimes of the past because of AI’s specific “affordances.” Affordances are shaped by what individuals perceive about an artefact, what they can do with it, and its cultural and institutional legitimacy.

In the infosphere revolution, information systems must be viewed as social tools that, when appropriated in the cultural milieu, function to mediate action, making certain kinds of actions more possible, and others less likely. In other words, AI assemblages bring different affordances, that is, potential or opportunities to serve as tools to execute, truncate, or complete certain actions.

From this point of view, affordances are understood as relational rather than as properties of things. Affordances are functions in the sense that they can both allow and confine action. In Latin America and the Caribbean, we should not ask only what AI assemblages afford, but how, for whom, and in what circumstances.

Interrogating the “affordances” of all digital technologies, especially AI, offers jurists a way to unravel how AI assemblages substantially constrain and enable acts while under construction and situated in social settings. AI does not make people do things but instead, push, pull, enable, and constrain.  AI is purely a new form of human agency in the infosphere revolution. It is also a political and social ideology rather than a suite of technological tools or a bundle of algorithms. AI is humans. AI is a system of different components, not just the algorithms at its core. It is an assemblage of multiple technologies and is not a public utility.

Current laws are ill-suited to handle the complexities and challenges of AI. Another area in which current law is unsatisfactory is privacy regulation. AI changes our current understanding of privacy. The “Right to be Forgotten” and its inapplicability to AI is rooted in our misunderstanding of privacy in the machine learning era. How can we guarantee a balance between the machine learning model’s “Need to Remember” and the human “Right to be Forgotten?”

The legal history of the “Right to be Forgotten” began in 2010. On May 13, 2014, the European Court of Justice legally solidified that the “Right to be Forgotten” is a human right in the Costeja case. But there must be a balancing of liberties. The “right to know” must balance the freedom to remove or increase the difficulty of accessing truthful information.

The Right to be Forgotten, as outlined in Article 17 of the EU General Data Protection Regulation, poses challenges for AI-Assemblages. This right empowers the digital citizen to request the deletion of personal information and to stop disseminating private information that can serendipitously appear on the dark web during a DOS attack. Our sense of justice is quickly becoming out of date.