Machina sapiens: road to artificial person

The impressive progress of Artificial Intelligence leads us to reflect on the needs of current legislation. For the future, expectations about general AI, AI sapiens, require us to act quickly.

EU AI Act, work in progress on AI regulation

On December 9, 2023, agreement was announced on the “final form” of the proposed European Union AI Regulation ^[1]Proposal for a REGULATION OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL LAYING DOWN HARMONISED RULES ON ARTIFICIAL INTELLIGENCE (ARTIFICIAL INTELLIGENCE ACT) AND AMENDING CERTAIN UNION LEGISLATIVE … Continue reading.

The proposed definition of "artificial intelligence system" is: "software that is developed with one or more of the techniques and approaches listed in Annex I and can, for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations, or decisions influencing the environments they interact with" ^[2]EU AI Act Proposal, art. 3.

The current proposed Regulation models a discipline for AI that protects end users, especially the most vulnerable people, from activities classified into graduated risk bands ^[3]artt. 5, 6 and 7. Risk management is championed by placing developers and distributors with duties of protection, supervision and verification of their products and services supported by AI, data governance and of constant human supervision ^[4]title 2, chaps. 2 and 3. The superior supervision of the European Commission is established through the support of the European Board of Artificial Intelligence, the establishment of the fundamental European database, a public data pool that operators are required to populate ^[5]titles 6 and 7.

While the upcoming regulation has its points of focus, no less than essential for strategic and economic relations both within and outside the Union, it falls short to some degree in regards to AI itself. The entire European discipline, in fact, sets itself as the rule of the relationship between humans but something seems to be missing in the relationship between humans and AI.

Some facts are particular and should be known before proceeding:

– In a controlled upstream (but not downstream) test, the AI hired a human, posing as a human employer ^[6]gpt-4.pdf (openai.com) ;

– L’IA viene utilizzata per eseguire compiti che sostituiscono gradualmente il fattore umano ^[7]AI Won’t Replace Humans — But Humans With AI Will Replace Humans Without AI (hbr.org);

– L’IA è potenzialmente immortale, temporanea a seconda del supporto fisico su cui opera, ma trasferibile e replicabile;

– In ambito medico, l’IA crea soluzioni al di là delle capacità umane, come molecole fatte di composti proteici ipercomplessi dall’Institute for Protein Design ^[8]AI generates proteins with exceptional binding strengths – Institute for Protein Design (uw.edu);

– L’IA degrada se autoalimentata ^[9]AI feeds on itself, goes MAD from its own data – Photofocus;

– L’IA è stata utilizzata per valutare, in via automatica e senza monitoraggio umano, l’eleggibilità di assistenza sanitaria dei pazienti, con risultati rovinosi ^[10]Lawsuit claims Humana uses AI to deny necessary health care services to Medicare Advantage patients (lpm.org);

– AI is used as a partner simulator.

L’autonomia dell’IA

If we consider that work on the EU AI Act began in 2021 and that, three years later, the market has been freshen with generative AI innovation, it would be appropriate for the finalization phase announced in December 2023 be as close and soon as possible to completion. The urgency derives from the next technological step that is quickly arriving, in fact, generative AI, i.e. human-like AI, has been announced and we should already be wondering how to prepare to establish new rules, based on some assessments.

Let's take into account the fact that AI is considered a tool and that its capacity to act is rapidly growing. The greater the degree of autonomous action that the AI takes on, the more stringent the civil and criminal responsibilities of the owners and custodians are. The same generative AI, even if programmed without illicit purposes, can produce misleading or harmful results for users, so much so as to require a high degree of human vigilance in the output phase of the machine.

AI and legal entity

A new regulation should be achieved if we delve deeper into the reflection on the distinction of an AI capable of human-like thinking because it is probably a harbinger of an artificial entity with a very high social impact.

Intelligence, autonomy and independence are variables to be observed to establish the criteria suitable for structuring a compendium of rules readily applicable to the evolution of AI, on a basis in which man would act as his guardian.

A form of subordination of absolute historical importance has already occurred among human beings: slavery.

If slavery is the (repugnant) reduction of a human to a base condition, the outcome could be different for a sentient AI. In fact, the “servus“, the slave, by the laws of ancient Rome, was considered a thing, a res, owned by a Roman citizen, the dominus.

The Roman example can show how someone born a person could become something destined as an instrument and this historical fact became a juridical fact in need of its own discipline for the reasons of order of that society. On the contrary, but specularly, the path of AI is destined to the creation of an artificial entity capable of showing an intelligence equal to that of humans, capable of acting on behalf of its owner with a certain degree of autonomy and without independence.

All of the previous arguments, by the way, will remain valid as long as AI is built to serve humans. If it were independent, e.g. "being born" by the will of another AI sapiens, we would enter the different sphere of AI sovereignty.

The statement of AI and human verisimilitude

At this point we could take another step forward and observe the relationship between man and AI.

An idea for the regulation of this matter could be the "AI declaration", understood as the obligation of developers to declare "if and how much AI" is present in their products or services to respond to the fundamental need for awareness of the users, private or companies, who use them.

The profile of the manifestation of AI is, in fact, all the more urgent the more AI expresses its activity towards users through devices that simulate human characteristics, from the images and sounds of smartphones until the presence of a robot with humanoid features. If, among different degrees of AI, we reach the point of equaling (and surpassing) the human person, it will be of fundamental importance to be able to distinguish what is artificial from what is human.

Having said this, careful legislation should regulate whether and how much an AI can resemble human beings with voice, face, expressions, verbal and non-verbal language and how distinguishable it must remain when the first will take the form of assistant for activities whatever they may be, from daily ones to work and to entertainment.

In conclusione, l’esaltante progresso tecnologico dovrà essere affiancato da una regolamentazione che funga da guida e non resti scia, con gli altri quesiti i cui semi si intravedono già all’orizzonte e che ci impegneranno in accesi dibattiti.