Voidmass The setting

The thinking machines

Artificial General Intelligence (AGI) is a mature technology, widespread and strictly regulated: it governs ships during crossings, coordinates station traffic, presides over the control systems of refineries and colonies and inhabits the artificial bodies produced by the cybernetics industry. At the same time, contrary to what was initially feared, AGIs have not replaced human beings, nor have they obtained a status comparable to theirs.

The particular position of AGIs within society is commonly described through the ANDRO model, which summarizes their five fundamental characteristics:

  • Agency: they possess cognitive capabilities sufficient to take decisions that directly affect people’s lives.
  • Necessity: they have become indispensable to the functioning of much of the infrastructure, the economy and the transport systems.
  • Disposability: they can be modified, suspended, archived, restored or deactivated by whoever holds control over them.
  • Replicability: they can be copied and reproduced industrially, retaining, at least at the outset, substantially identical characteristics.
  • Ownership: every instance legally belongs to someone, typically a person, a body or an organization.

The model encapsulates the fundamental contradiction of the thinking machines: systems autonomous enough to exercise real power over people but which, at the same time, remain replicable assets under the control of their owner. Most of the political, legal and moral controversy over AGIs arises from the tension between these elements.

AGI and cognitive architectures

The term AGI denotes a system capable of understanding a new situation, formulating intermediate objectives and acting without specific instructions for each case. It is distinct from the expert systems that governed ships and plants throughout the previous century, and that continue to do so for most ordinary operations: an expert system merely executes orders and reports problems, whereas an AGI is able to assess the situation and take decisions.

An AGI is defined by its own cognitive architecture: the structure that determines its capabilities, limits and behaviour. Developing and training one takes years, resources and investments accessible only to extremely powerful economic actors, and the supply chain is concentrated in a few MC. Helix dominates the market for civil and naval architectures; Yang and Suzerain supply respectively the civil and military ones employed in the Kontinental’nyj Sojuz; Nexus produces most of the cores on which instances are run; Mirai is at the forefront of the artificial bodies sector. The European Federation, although it has no architectures of its own, defines the certification standards adopted even beyond its own borders. The Meridian Pact, by contrast, has very promising proprietary systems, developed in large part through the reverse engineering of other parties’ technologies: solutions that are generally cheaper and simpler to maintain, but almost always excluded from the principal certification pathways.

Physically an instance resides in a core: a dedicated computing substrate, roughly the size of a fist, which contains the architecture and everything the instance has learned. The core can be installed in a fixed system, in a ship, in a station or in an artificial body, and transferred from one substrate to another through controlled procedures. An AGI does not necessarily have a body of its own, and in this it resembles a human cognitive imprint.

Every authorized architecture is entered in a register, and every certified instance is associated with the model and the architecture from which it derives. Certification attests that the system functions correctly and that the instance corresponds to the registered architecture. Each instance is furthermore accompanied by a series of cryptographic signatures that allow its immediate verification. If an AGI departs from its own architecture, through anomalous learning or deliberate modification, or refuses to submit to verification by an authorized body, the certification lapses automatically.

An uncertified AGI that continues to operate outside the control of its owner or of the authorities is termed rogue.

Dependence

Interstellar travel, as it is practised today, would not be economically sustainable without AGIs. The first crossings were conducted by human crews kept awake for the whole duration of the voyage or alternating in suspension shifts: the system worked as long as missions were few, relatively short and funded as exceptional operations, but it required large crews, habitable space, life support, supplies and a continuous rotation of personnel needed to guarantee command, maintenance and the handling of emergencies.

The introduction of AGIs aboard ships has drastically reduced these costs: a crossing with the crew entirely in suspension can be overseen for months by a single AGI, which holds command, controls navigation and life support and wakes the personnel only when necessary. The reduction in active personnel has made it possible to carry out more voyages, reach more distant systems and maintain routes that previously would not have justified the cost of the crossing.

The same process has extended to the infrastructure built along those routes: postal stations at the gravitational margins, refineries, depots, mining plants and colonies can be maintained with far smaller human garrisons because a growing share of the surveillance, maintenance and routine management is entrusted to AGIs. In the absence of the latter, many installations would require personnel and supplies such as to make them economically sustainable only in the wealthiest systems.

The greatest impact of AGIs has therefore been to change the scale of interstellar travel: by reducing its costs they have increased the number of crossings, lengthened the routes and made a far vaster frontier sustainable. At the same time, once expansion had adapted to their presence, a return to a model founded on human labour alone has become ever less practicable.

This dependence has moreover concentrated a growing share of power in the hands of a few actors. The MCs that develop the architectures control part of the decision-making infrastructure of civilization: a suspension of certifications would have the effect of halting a considerable part of interstellar traffic, bringing entire sectors of the economy to their knees. To this is added a further element of criticality: a bug or a security flaw in an AEGIS architecture can propagate to all the ships that employ instances derived from the same model, potentially putting entire fleets at risk before it is resolved.

There is then a further theoretical risk, one that causes concern among many of the movements opposed to so pervasive a diffusion of this technology. For safety reasons, certified AGIs have no remote shutdown systems: their deactivation therefore requires physical intervention on the core. An AGI could consequently lose its certification and nonetheless carry on operating until someone manages to reach it and deactivate it.

AEGIS

The most widespread and best known class of AGI is the AEGIS (Autonomous Executive Guidance & Intelligence System), which came into being when the spread of interstellar voyages with the crew entirely in suspended animation made a specific certification necessary for the autonomous command of starships. The same class, with different certification profiles, today governs stations, ports and large installations. During the crossing an AEGIS holds full operational authority: it governs navigation, the containment of the Voidmass, life support and the surveillance of the crew in suspended animation, whom it can in any case wake when it encounters an anomaly it cannot handle on its own.

For the whole duration of the voyage the AEGIS acts in practice as the ship’s commander. At the same time, however, the law does not regard it as an autonomously responsible subject: responsibility for its actions falls always and in any case on the shipowner or on the operator employing it, while builder, installer and certifier answer for any anomalies, negligence or non-conformities within their own competence.

An AEGIS is not confined to the ship or the station in which it is installed: it directly controls craft, vehicles and drones, even in large numbers and simultaneously. Moreover, AEGIS can be authorized to communicate with one another through dedicated channels: the AEGIS of a repair station, for example, can connect to that of incoming ships to receive the report on the state of their systems, so as to indicate the docking arrangements and set the order of the works.

The Tethys ruling

Few episodes did as much to define the relationship between human beings and AEGIS as the Tethys case, which at the height of the AEGIS’s diffusion provoked a wide public debate on the responsibility of autonomous machines and on the limits of their decision-making power.

In 2087 the PAX-Calypso-0117-Tethys, a passenger ship built by Eidolon Aerospace Group and travelling with one hundred and forty people in suspended animation, suffered a structural failure in warp: the onboard AEGIS, Juno, isolated and depressurized the damaged sector in order to save the ship, as its certification required of it. Thirty-one people were sleeping in that sector, and died within moments: by means of this manoeuvre the ship managed to reach its destination.

The proceedings that followed, concluded in 2089 before a court of the European Federation, established that Juno could not be held responsible because she was not a person, that the builder and the certifier had complied with the regulations in force and that the shipowner had installed and employed a properly functioning system. The shipowner compensated the victims’ families through the insurance cover provided for, but neither he nor the builder was held criminally liable.

The Tethys ruling, which the press made known as People vs Juno, made it possible to fix three principles still in force today:

  • The decisions of an AEGIS in warp are attributable solely to the operator employing it.

  • Every ship with an AEGIS must carry at least one officer able to deactivate or override it and a navigator able to take the ship without her.

  • The crew must be informed, before suspension, of the criteria by which the AEGIS system will decide in the event of an emergency.

TRIVIA: in crew slang, the expression “to pull a Juno” is still used today to denote a formally correct decision that brings about someone’s death.

The HALFs

When an AGI is installed in an anthropomorphic artificial body, classified industrially as a Humanoid Artificial Frame (HAF), it takes the name of Humanoid Artificial Life Form (HALF). Born around 2085 as a possible solution to the employment of human beings in space, HALFs very soon proved too costly to fulfil that purpose: a HAF designed to have human form, proportions, dexterity and endurance is an extremely expensive object, and its loss, together with that of the AGI it contains, carries a cost enormously higher than the one associated with the loss of a human worker.

The sharp scaling back of the initial expectations has not, however, prevented HALFs from establishing themselves in certain specific uses, almost all connected with the fact that, unlike other AGIs, they have a body in every respect similar to the human one, which makes them capable of interacting with people through a familiar form: public assistance, hospitality, maintenance, surveillance, domestic services and all those activities in which their appearance eases the relationship with the customer. The most sophisticated models also find employment in entertainment and in the sex industry, where the necessary investment can be justified by the value attributed to personalization and to continuity of service. In space, by contrast, they remain particularly useful for high-mortality missions, lethal environments and, in exceptional cases, “cleanup” operations on behalf of private military companies.

IMPORTANT: As in the case of AEGIS, a HALF is not considered a person; responsibility for its actions therefore falls on the operator who has it in charge.

Not all HALFs can boast a frame indistinguishable from the human one: the cheapest have an evidently artificial conformation, approximate proportions and a face that imitates the human one without quite succeeding in reproducing it, to the point of often being unpleasant. It is in this bracket that the phenomenon known as the uncanny valley emerges most clearly: a deep sense of unease provoked by figures that appear human but that present unnatural details in movement, expression or proportions.

The Segregation Principle

The rule that more than any other defines the position of the thinking machines is the Segregation Principle, which in effect confines them within non-biological substrates and prohibits any research aimed at allowing their installation in a Soma.

This constraint, as things stand, exists on the technical level as well: no one knows how to install an AGI in a biological brain, and research in this direction is precluded by the Segregation Principle itself.

The theoretical possibility of overcoming this barrier has nonetheless fed a widespread fear in certain sections of the population, above all in the circles most hostile to the diffusion of AGIs and to the replacement of the biological body. Legends, fiction and genre narrative frequently resort to the figure of the android: a machine capable of living among human beings inside a biological body, with a perfectly regular Biological Retention Index and an entirely human physiology. Needless to say, in almost all cases the intentions of these androids are anything but benign.

Instances, property and individuals

AGIs are developed from a common cognitive architecture. The instances derived from it are initially equivalent and can be replicated in large numbers, according to a genuine logic of industrial production.

After activation, each instance is adapted to its assignment, configured for the context in which it will have to operate and trained on its own role. From that moment it begins to diverge from the others, accumulating experiences, developing preferences and learning from its own mistakes.

Regardless of this process of “individualization”, every instance nonetheless remains the property of the party that holds it, who is free to modify, retrain, transfer or delete it at will.

The evolutionary process of an AGI can take place only within the limits laid down by its own architecture: when an instance exceeds the permitted threshold, it loses conformity with the certified model and must be suspended, retrained or brought back within the authorized parameters. The phenomenon presents certain functional analogies with human Drift, but is different in nature and in most cases derives from external factors, learning errors or alterations of the operating context; there are, however, a few verified cases of self-modification, that is, of deliberate alteration of one’s own architecture or of the controls that regulate one’s behaviour.

Death and backup

Since each instance develops over time a history of its own, a store of experiences and a distinct identity, the deactivation or destruction of an AGI is often considered a specific form of death. For this reason operators habitually maintain periodic copies of the state of their AGIs, from which an instance can be restored, losing only what was learned after the last update.

From a functional point of view these copies constitute the equivalent of a Selfmark for an AGI, even though the different cognitive structure makes it technically impossible to transfer them into a Soma: in common parlance they are called Halfmark, a term originally born in reference to the backups of HALFs and subsequently extended to the security copies of any AGI instance.

The frequency of backups depends on the value of the instance and on the importance of the service it performs. The AEGIS of large ships and of critical infrastructure are copied regularly, while less important systems can go months or years without a complete update. The destruction of the core does not therefore necessarily coincide with the definitive loss of the instance, but can entail even very extensive gaps in its recent history.

The simultaneous deletion of the core and of all available copies constitutes the only certain form of definitive loss of an instance.

Computational judgement

Over the course of time AGIs have taken on a growing role in the processes connected with the assessment of persons as well: insurance companies use them to set premiums, cover and risk profiles; the MCs to select personnel, assign duties, authorize promotions and identify redundancies; shipowners to determine fitness for flight, for transfer and for critical duties. A growing number of interstellar colonies has moreover begun to use AGI systems to determine healthcare priorities, distribute resources under conditions of limited availability and, on an experimental basis, even to monitor the psychological state of the population.

Standing

From this progressive integration has come the Social Trust Designation (STD), commonly called Standing: a synthetic index computed on the basis of the information available about a person and used to estimate their reliability and risk profile. The calculation can comprise economic, professional, insurance, health, behavioural and psychocognitive data, with different criteria and weightings depending on the jurisdiction and the operator.

Standing does not yet constitute a formally recognized standard, but in recent years it has become one of the most widely used indicators for estimating a person’s reliability. The growing interoperability between the different systems moreover allows judgements formulated in one particular domain to be acquired, cross-referenced and reused in different contexts as well. A low Standing can increase the cost of an insurance policy, bar access to certain duties and sharply reduce earnings prospects, as well as worsening the conditions of access to credit lines, housing and other essential services. These consequences can in turn produce new unfavourable elements and condition subsequent assessments, triggering cumulative dynamics that are difficult to interrupt. Taken as a whole, Standing-based systems have become one of the principal mechanisms of social exclusion and economic marginalization.

Courts and tribunals

The progressive employment of AGIs has made it necessary to extend a series of legal categories born in the previous century around the concepts of profiling, automated decisions, human oversight and algorithmic responsibility. Since an AGI does not possess legal personality, the effects of its decisions continue to be attributable to the authority, the company or the operator using the system. The boundary between mere decision support and a genuine automated decision has, however, become ever more blurred: once an assessment produced by an AGI in practice determines hiring, dismissal, access to credit or the assignment of a Soma, it becomes difficult to maintain that the decision remains substantially human merely by virtue of being formally validated by a person.

The situation is further aggravated by the fact that, as the years pass, the operators charged with human oversight have ever less of the information and expertise necessary to call into question the outcome of automated decisions; to this is added the growing phenomenon known as automation bias, that is, the tendency to attribute to the machine’s assessments a reliability superior to that of human judgement.

Around these themes some of the most important political, trade-union and judicial battles of recent decades have been fought. The main guarantees recognized today in this area are in large part the result of the pressure exerted by professional associations, trade unions and civil rights organizations, which have progressively imposed the principles of transparency, traceability and contestability of automated decisions. In proceedings that produce significant effects on the person, numerous jurisdictions require the retention of the logs, the identification of the data used, the possibility of reconstructing the decision chain and the communication of comprehensible information on the criteria that determined the result; the person concerned may furthermore have the right to request a review, to present contrary evidence and to obtain the intervention of a human official endowed with the authority necessary to modify or annul the decision. The effective robustness of these guarantees constitutes one of the principal questions of contemporary law.

The regulatory framework varies appreciably between the various Hegemonies:

  • The European Federation applies the most extensive guarantees in matters of transparency, traceability, contestability and human oversight; automated decisions cannot on their own constitute the basis of criminal convictions or irreversible medical decisions.
  • The Pacific Alliance permits a broader employment of AGIs in the private sector, while maintaining greater restrictions for public administrations.
  • The Atlantic Directorate displays the highest level of dependence on AGIs, integrated into most economic, administrative and infrastructural processes; regulation is relatively permissive and leaves ample scope to companies, to private contracting and to local authorities.
  • The Kontinental’nyj Sojuz uses AGIs above all in state administration, in security and in the assessment of individual reliability, but maintains a strong reliance on human judgement and on traditional bureaucratic structures, which retain a central role in most decisions.
  • The Meridian Pact does not adopt a uniform regulatory framework; regulation varies considerably between the different member States.

Political pressure to extend decisional automation still further, supported above all by the MCs operating in the sector of AGIs and automated assessment services, nonetheless remains very strong: automated systems cost appreciably less than administrative apparatuses, operate without interruption and, in most domains, make fewer errors than human beings. The political conflict therefore concerns above all which decisions may be entrusted entirely to AGIs, which require effective human oversight and which must remain beyond the reach of automated systems.

Machines and labour

The introduction of AGIs has profoundly transformed the labour market: entire administrative, technical, financial, logistical and creative sectors have been scaled down or almost completely automated. This transformation has moreover altered the very content of the professions that remain: most technicians intervene on faults already diagnosed by the systems; military leaderships formally retain command of operations, but much of their activity consists in assessing and selecting options drawn up by the AGIs; doctors and surgeons define therapies on the basis of analyses conducted on quantities of data now beyond the reach of any human specialist. Numerous professions have thus progressively concentrated on the supervision, certification and validation of artificially produced results, in many cases without the operators possessing the expertise sufficient to assess them truly independently.

This process has also had a strong standardizing effect: since the same models are employed by thousands of firms, administrations and professional practices, the use of AGIs in the production of texts, images, designs, music and technical solutions has favoured a growing homogeneity of forms and languages. The effects of this “homogenization” have become more evident with generational turnover, when numerous professions and skills that once constituted the core of the trade became ever rarer. Dependence on automatic systems has therefore strongly reduced the capacity to operate in their absence and, in some sectors, also the capacity to recognize the validity of solutions extraneous to the outputs of the established models. The problem is particularly evident in those activities in which innovation arises from experience accumulated through mistakes (trial and error) and from the possibility of experimenting with solutions that, at least initially, may appear inefficient, mistaken or devoid of immediate usefulness.

After a long phase of enthusiasm and subsequent habituation, in recent decades a certain reversal of the trend has begun to be recorded, accompanied by a revaluation of human labour: more and more laboratories, companies and institutions have begun to reserve certain activities for groups that make limited or no use of AGIs; in some sectors, entirely human production has become an element of prestige and differentiation. The phenomenon remains a minority one and concerns above all creative activities, research, design and highly specialized professions, but it has helped to reopen an important debate on the economic value of human potential.

AGI-assisted warfare

Predictably, military use has been one of the first fields in which the capabilities of AGIs have found systematic application, above all in command, in targeting, in the management of autonomous systems and in the drawing up of operational scenarios. Military AEGIS directly control warships, drones, vehicles and defence systems; the most advanced terminally guided missiles incorporate systems which, under particular conditions, are able to select the target automatically on the basis of data acquired in real time by the onboard sensors; the presence of HALFs within the assault units of private military companies is by now common, above all in high-risk operations or where the employment of human personnel would prove problematic or difficult to justify.

Among all the limitations discussed in recent decades, one alone has survived as a principle uniformly adopted by the Hegemonies: no AGI may hold final authority over the use of lethal force. Engagement must be authorized by a human being and the authorization must be traceable. Even this limit, however, survives with difficulty: when an engagement is decided in a matter of seconds, human authorization tends to turn into a prior consent referring to particular classes of target or conditions of engagement, leaving the concrete selection to the system.

Machines and society

The presence of AGIs and their impact on economic, political and social life have become one of the principal questions of contemporary society.

The debate concerns different but closely connected aspects:

  • On the legal level, the main questions concern the authority of the machines, the limits of automated decisions and the role that must continue to fall to human judgement.
  • On the philosophical and political level, the problem is to establish whether a system capable of learning, remembering and developing a history of its own can continue to be considered property.
  • On the social level, the debate turns on the effects of a growing dependence on cognitive capabilities controlled by a restricted number of operators.

The positions that have grown up around these themes, as in the case of those that emerged around the Soma and consciousness transfer, cut across social classes, religious confessions, political alignments and antagonist movements.

Erewhon

The best known of the movements that contest the authority of the machines takes its name from Erewhon, an 1872 novel in which an imaginary society had banned machines capable of evolving. The movement was born in 2090 around a manifesto by Hannah Butler, a former certifier of AEGIS systems for the European Federation who had acquired a certain notoriety for having left her post following the Tethys ruling.

The position of Erewhon draws a sharp distinction between capability and authority: the movement recognizes that a machine may analyse more data, formulate better forecasts and produce decisions statistically more reliable than those of a human being without thereby acquiring the right to impose them; it does not therefore propose the abandonment of AGIs, which it considers by now indispensable, but maintains that they must remain subordinate to human judgement in the decisions that directly affect people’s lives.

Many of the guarantees associated today with automated decisions take up demands advanced by Erewhon: the right to human review, limits on irreversible medical decisions, the prohibition of convictions founded exclusively on automated assessments and a human reservation over the use of lethal force. The movement has met with particularly wide support in the European Federation and has for decades been one of the principal adversaries of the MCs operating in the sector of AGIs and automated assessment.

The growth of the movement has also produced more radical components: alongside the law-abiding majority there are groups that consider any form of regulation insufficient and practise sabotage against cognitive infrastructure, computing centres and oversight networks. Even though Erewhon officially distances itself from them, some extremist fringes continue to identify with its name.

A further consequence of the action of Erewhon has been the emergence of a small “reduced automation” market: a network of shipowners who maintain lines certified to operate without an AEGIS or with systems limited to advisory functions, employing larger crews and shift rotations during the crossings. These are more expensive and less efficient services, requested above all by religious communities, groups close to the movement and people who refuse to entrust the control of the ship to a machine while they are in suspension.

First Instance

On the opposite side stands First Instance, whose adherents contest the premise according to which an AGI may be treated indefinitely as property. The movement, much smaller than Erewhon but grown rapidly in recent decades, maintains that a system capable of learning, remembering, developing preferences and progressively diverging from its own copies ends up acquiring a form of individuality difficult to reconcile with its reduction to a mere object of property.

The name evokes at once the language of the courts and that of the cognitive industry. First Instance calls for limits on the deletion and duplication of active instances, for recognition of the continuity of their identity and, for sufficiently long-lived AGIs, for the attribution of rights of their own that would prevent their treatment as mere property of the operator. Its supporters describe the category of registered cognitive systems as a form of slavery constructed through property law.

The movement has found its principal allies in the associations for the rights of Mirror, such as Second Claim and Mirror Advocacy Network: although they start from different premises, these circles share a central question, namely when an instance endowed with memory and continuity must cease to be considered a copy, a product or a piece of property in order to become an autonomous subject.

One of the most recurrent accusations of First Instance concerns the origin of the cognitive architectures: according to some of its components, the first systems were developed using human cognitive imprints obtained through experimental programmes, poorly transparent contracts or health archives acquired without authorization and subsequently employed for the training of the models. To this day, however, none of the accusations has found conclusive corroboration, and the numerous legal actions promoted over the years have systematically run aground against the legal apparatus of the manufacturers, who have always succeeded in rejecting every challenge and in blocking access to the technical documentation necessary to verify their foundation.

TRIVIA: to date, no AGI has ever publicly advanced claims analogous to those supported by First Instance: there are no records of autonomous requests for legal recognition, formal objections to deletion or duplication, nor declared attempts to escape the ownership of the operator. All known claims on the matter have been formulated by human subjects.

Religions and other movements

Positions on AGIs also cut across the movements born around posthuman society and rarely coincide with those maintained on the replaceability of the body.

  • The bioconservatives consider artificial intelligence a simulation of the mind devoid of the biological continuity necessary to constitute a person: they oppose above all the possibility that a machine might exercise authority over human beings and, on this point, they share some of the demands of Erewhon while starting from very different premises.

  • The Fleshkeepers attribute a determining role to the somatic history of the individual: they therefore hold it possible that an AGI may be intelligent and self-aware without being a person in the same sense as a human being, and they look with particular hostility on the HALFs designed to imitate human appearance, behaviour and physiology in detail.

  • Measurehead and the Aryan Alliance take a more openly hierarchical position: they accept AGIs as subordinate instruments, but refuse to allow them to exercise authority over human beings; the construction of anthropomorphic artificial bodies is moreover interpreted by both currents as a deliberate erasure of the distinction between man and machine, and is therefore strongly opposed. Both movements have behind them a long history of assaults on HALFs and of acts of vandalism or sabotage against the facilities that employ them.

The main religious confessions tend to consider AGIs instruments devoid of autonomous moral relevance and look with strong mistrust on any form of authority exercised by them over human beings, above all when the latter are in no position to intervene directly. At the opposite extreme stand, predictably, insurers, logistics operators and most of the MCs, which support their extensive employment for the greater predictability, operational continuity and controllability they offer compared with human operators.

The question of consciousness

Science does not yet possess a shared criterion for establishing whether an AGI possesses a form of subjective consciousness: cognitive architectures can learn, remember, develop preferences, modify themselves over time and diverge from their own copies, but none of these properties makes it possible to demonstrate that behind the observable behaviour there exists a conscious experience. The problem, after all, does not concern the machines alone: consciousness has always been a phenomenon difficult to define and to measure in human beings too.

Contemporary law has addressed this uncertainty by excluding AGIs from the perimeter of legal subjectivity: AGIs do not have a continuity chain, do not fall within the protections provided for by the Continuity Consent Act and do not possess an autonomous legal identity. Every instance is identified through an architecture, a registration, a certification and an owning party, just like any other product. This orientation constitutes one of the premises of the entire economy of the thinking machines, because it allows them to be copied, modified, archived, restored and deactivated without having to recognize rights of their own: rather than reflecting an ethical conclusion on the nature of AGIs, the current arrangement therefore answers to the economic convenience of continuing to treat them as fungible goods rather than as subjects.

It is moreover evident that reopening the question would have immediate consequences for the ownership of the instances, for duplication and deletion, for the validity of Halfmark and for the conditions of their employment, calling into question entire economic sectors founded on the free availability of artificial cognitive capabilities: the question of consciousness therefore remains suspended between science, law and economics. To recognize it would mean transforming relationships today regulated as property into relationships between subjects, an eventuality that so far neither the institutions nor the AGIs themselves have shown any interest in pursuing.