Five beneficences for a post-biological era

Bioethics has historically been linked to life, medicine, biomedical research, public health, and the relationship between science, human values, and planetary survival. Van Rensselaer Potter presented bioethics as a "bridge to the future," that is, as a discipline capable of articulating biological knowledge, moral responsibility, and human survival in a context of increasing technological power (Potter, 1971). Beauchamp and Childress, in turn, consolidated the principlist language of bioethics around four fundamental principles: autonomy, beneficence, non-maleficence, and justice (Beauchamp and Childress, 2019). At first glance, therefore, applying bioethics to non-biological systems seems to involve a conceptual tension: if bioethics is an ethics of life, how can it be applied to entities that are not alive?

This difficulty should not be hastily resolved through mere rhetorical broadening. An artificial intelligence system, a robot, a computational model, a digital twin, an algorithmic infrastructure, or an autonomous agent are not biological organisms. They do not have, at least in the current state of evidence, metabolism, organic homeostasis, biological reproduction, their own mortality, or bodily experience equivalent to that of living beings. Therefore, it would be conceptually wrong to treat all non-biological systems as if they were patients, organisms, or moral subjects. But it would be equally wrong to conclude that bioethics has nothing to say about them. The crucial question is not whether these systems are biologically alive; it is whether they interfere with life, vulnerability, agency, care, health, dignity, decision-making, and responsibility.

This is where the conceptual difficulty becomes productive. Bioethics can be applied to non-biological systems not because they are automatically living beings, but because they have become integrated into the ecosystems in which human, animal, institutional, and environmental life is produced, classified, monitored, treated, protected, or threatened. Artificial intelligence in healthcare decides clinical priorities, identifies risks, recommends therapies, participates in diagnosis, and reorganizes trust relationships between doctor, patient, and institution. AI in security and defense alters the distribution of force, surveillance, and lethal decision-making. Digital twins model bodies, cities, hospitals, and infrastructure. Autonomous systems reorganize transportation, energy, finance, education, and access to services. Bioethics, in this context, can no longer be limited to the isolated organism; it must analyze the mediating systems that condition the concrete possibilities of life.

This shift approaches an agenoethical formulation: ethics emerges where agency encounters vulnerability. Agency does not belong solely to human individuals. It can be intentional, institutional, functional, distributed, hybrid, or regulatory. Vulnerability is also not only biological; it can be cognitive, informational, infrastructural, ecological, legal, political, or post-biological. Thus, bioethics can be understood as an ethics of agency within horizons of vulnerability linked to life. When non-biological systems acquire the capacity to act, mediate, decide, learn, select, prioritize, or produce relevant effects on these horizons, they become legitimate objects of bioethical analysis, even if they are not, therefore, automatically moral subjects.

This distinction is essential. A non-biological system can be bioethically relevant in three different senses. It can be relevant as an instrument that affects humans, animals, populations, or ecosystems. It can be relevant as a relational mediator that alters practices of care, dependence, empathy, trust, or authority. And it can, in future and still uncertain cases, become relevant as a possible moral patient, if it comes to possess consciousness, sentience, self-interest, or some kind of subjective experience. Recent literature on artificial consciousness and AI welfare is cautious, but already treats this last hypothesis as a philosophical, scientific, and institutional question that should not be dismissed prematurely (Butlin et al., 2023; Long et al., 2024). The authors of Taking AI Welfare Seriously argue that there is no reason to claim that current systems are conscious, but that there is sufficient uncertainty to justify research, evaluation, and institutional preparation regarding possible future forms of artificial moral patienthood (Long et al., 2024).

The principle of beneficence is particularly suited to thinking about this transition. In classical bioethics, beneficence means promoting good, preventing harm, removing risks, and acting for the benefit of those who may be affected by our actions (Beauchamp and Childress, 2019). In medicine, this has translated into guiding clinical practice towards the patient's well-being. In research, it means weighing risks against benefits. In public health, it means protecting populations. In environmental bioethics, it means caring for ecosystems, future generations, and non-human life forms. The question that now arises is more demanding: can beneficence guide the creation, design, use, and perhaps the treatment of non-biological systems?

The answer depends on distinguishing five levels of beneficence: instrumental, relational, systemic, agentic, and patient. These levels should not be confused. Not all artificial systems are moral patients. Not all autonomous systems have self-interest. Not every affective relationship with a machine justifies the machine's rights. But all these levels show that beneficence can no longer be limited to the traditional clinical model.

1. Instrumental beneficence : non-biological systems as means to human, social, and environmental good.

The first level is the most consolidated. Instrumental beneficence understands the non-biological system as a means to benefit humans, communities, institutions, and ecosystems. This is the dominant form of AI ethics. The artificial system is not a direct beneficiary of moral duties; rather, it is a technology that must be designed, trained, audited, and governed to produce beneficial results.

The AI4People framework is a central reference in this transposition. Floridi et al. proposed adapting the classic principles of bioethics—beneficence, non-maleficence, autonomy, and justice—to AI, adding explainability as a necessary principle for societies mediated by intelligent systems (Floridi et al., 2018). Beneficence, in this context, means that AI should contribute to human flourishing, dignity, social well-being, and the prevention of collective harm. Floridi and Cowls reinforce this formulation by advocating a unified framework of five principles for AI in society: beneficence, non-maleficence, autonomy, justice, and explainability (Floridi and Cowls, 2019).

The same orientation appears in the institutional frameworks of the OECD, UNESCO, and IEEE. The OECD presents trustworthy AI as geared towards inclusive growth, sustainable development, well-being, human rights, democratic values, transparency, security, and accountability (OECD, 2019). UNESCO places human dignity, human rights, environmental protection, justice, human oversight, transparency, and accountability at the center of its Recommendation on the Ethics of Artificial Intelligence (UNESCO, 2021). The IEEE, in Ethically Aligned Design , argues that autonomous and intelligent systems should be aligned with human values and geared towards human well-being (IEEE, 2019).

At this first level, beneficence is clear, but it remains anthropocentric and instrumental. AI is good when it serves good human ends. A clinical model is beneficent if it improves diagnosis, reduces error, expands access, decreases inequalities, and respects the patient. An intelligent energy system is beneficent if it increases resilience, reduces waste, and supports sustainability. An epidemiological surveillance system is beneficent if it protects public health without destroying privacy, freedom, or proportionality. The non-biological system is evaluated by its effects on life, not by any self-interest.

2. Relational beneficence : artificial systems as mediators of empathy, care, and moral conduct.

The second level is relational. Here, the question is not whether the artificial system has consciousness or rights, but what kind of moral relationship is formed when humans interact with systems that simulate presence, emotion, dependence, vulnerability, or care. Social robots, conversational assistants, artificial companions, therapeutic chatbots, care robots, and educational agents may not have experiences of their own; yet, the way they are designed and treated can affect human moral life.

Mark Coeckelbergh is a key reference in this approach. Instead of basing moral consideration solely on internal properties such as rationality, consciousness, or sentience, Coeckelbergh proposes a social-relational justification for the moral consideration of certain robots. His argument is not that all robots have rights, but that their insertion into social relations may require a revision of the categories through which we attribute moral consideration (Coeckelbergh, 2010).

Kate Darling develops a complementary line of thought. In analyzing the possibility of limited legal protection for social robots, Darling observes that humans tend to anthropomorphize interactive robotic objects and project emotions, intentions, and vulnerability onto them. The ethical question lies not only in the potential harm to the robot, but also in the effect that violence or cruelty towards social robots can have on empathy, moral habits, and human relationships (Darling, 2016).

David Gunkel expands on this discussion in Robot Rights , asking whether the Western moral tradition is prepared to deal with technological entities that do not easily fit into the classic opposition between person and thing (Gunkel, 2018). Gunkel's contribution is particularly important because it shows that the question "can robots have rights?" should not be reduced to a quick, binary answer. It exposes the fragility of the very criteria used to define who can be considered morally relevant.

This relational beneficence must, however, be balanced by Joanna Bryson's warning. Bryson argues that robots should remain artifacts at the service of humans, and that attributing an excessively human status to them can obscure human responsibility, divert moral resources, and favor forms of affective manipulation (Bryson, 2010). This counterpoint is essential. Relational beneficence does not require asserting that robots suffer; it requires recognizing that our relationship with systems that mimic vulnerability can either enhance or degrade human moral culture.

Thus, relational beneficence applies primarily in contexts of care, education, mental health, aging, childhood, loneliness, and dependency. A caregiving robot may not be a moral patient, but the way it is designed can strengthen or weaken care practices. A therapeutic chatbot may not feel, but it can induce trust, dependence, or emotional transference. An artificial assistant may lack dignity, but it can be used in ways that preserve or manipulate the user's dignity. Relational beneficence asks: what kind of humans are we creating when we normalize certain relationships with artificial systems?

3. Systemic beneficence : protecting digital infrastructures as a condition for protecting life.

The third level is systemic. Certain non-biological systems become so integrated into the operating conditions of society that their reliability, security, and resilience acquire indirect bioethical value. Digital health systems, smart energy grids, triage platforms, financial infrastructures, biomedical databases, transportation systems, risk models, communication networks, and defense systems are not moral patients. But human life, public health, autonomy, justice, and security may depend on them.

Luciano Floridi provides an important conceptual basis for this shift by thinking about ethics in terms of the infosphere, that is, as an informational environment in which humans, institutions, data, and artifacts interact constitutively (Floridi, 2013). Helen Nissenbaum, on the other hand, shows that privacy and informational integrity depend on social contexts and appropriate information flows, not just on individual data ownership (Nissenbaum, 2010). These approaches allow us to understand that beneficence in digital societies is not reduced to the intention of an individual agent; it involves the architecture of systems that condition action, choice, and vulnerability.

Virginia Dignum is equally central. For Dignum, responsibility in AI does not belong solely to the algorithm, but to the socio-technical system of which AI is a part: those who design, implement, use, govern, and benefit from the system participate in the production of its consequences (Dignum, 2019). In later texts, the author insists that responsible AI requires technical, social, institutional, and legal methods capable of making responsibility practicable, and not merely declarative (Dignum, 2022).

Systemic beneficence therefore means that protecting certain systems can be an indirect way of protecting people. A poorly governed clinical platform can produce medical error, discrimination, loss of trust, and breach of confidentiality. A vulnerable energy system can jeopardize hospitals, nursing homes, transportation, and supplies. An algorithmic model used in public safety can create unfair profiles and affect fundamental freedoms. At this level, benefiting the system does not mean granting it well-being; it means preserving its functional integrity because vulnerable subjects depend on it.

This dimension also aligns with Charles Perrow's theory of normal accidents. In complex and tightly coupled systems, local failures can propagate unpredictably and produce systemic accidents (Perrow, 1984). Applied to AI, this intuition suggests that beneficence cannot focus solely on the average performance of a model. It must assess interdependencies, cascading risks, opacity, institutional dependence, and reversibility. Systemic beneficence is, therefore, an ethics of resilience, auditability, and damage containment in socio-technical ecosystems.

4. Agentic beneficence : when artificial systems transition from tools to functional agents.

The fourth level is agentic. It emerges when artificial systems cease to be merely passive instruments and begin to execute actions, select means, adapt strategies, interact with complex environments, and produce effects that were not fully foreseen by their human operators. This includes autonomous agents, mobile robots, multi-agent systems, autonomous vehicles, operational clinical AI, autonomous military systems, software agents with planning capabilities, and models integrated into institutional decision-making chains.

James Moor is a classic reference in this discussion. The author distinguishes machines as ethical-impact agents, implicit ethical agents, explicit ethical agents, and full ethical agents (Moor, 2006). This typology is particularly useful because it shows that not all ethically relevant machines are full moral agents. Some only produce ethical impacts. Others incorporate safety constraints. Others may explicitly represent ethical rules. The category of full moral agent remains much more demanding and controversial.

Michael Anderson and Susan Leigh Anderson developed the idea of machine ethics as an attempt to create artificial agents capable of making ethically sensitive decisions in certain contexts (Anderson and Anderson, 2007). Wallach and Allen, in Moral Machines , systematically explored the possibility of teaching robots to distinguish right from wrong through top-down, bottom-up, or hybrid approaches (Wallach and Allen, 2009). John Sullins adds an important nuance by arguing that a robot can be analyzed as a moral agent under certain conditions without this implying treating it as a fully moral person (Sullins, 2006).

Agentic beneficence does not say that autonomous systems are morally equal to humans. Rather, it says that when non-biological systems begin to act, beneficence must be incorporated into the very architecture of artificial action. It is not enough for the overall objective to be beneficial. The means, limits, exceptions, supervisory mechanisms, auditability, and capacity for interruption must be compatible with ethically acceptable ends. A clinical system that optimizes productivity may harm vulnerable patients if it does not recognize exceptions. A financial agent that maximizes return may produce systemic damage. An autonomous military system that optimizes threat neutralization may reduce the human person to a target standard.

This dimension is also linked to the responsibility gap problem identified by Andreas Matthias. When systems learn and act in ways that are not entirely predictable, it becomes more difficult to assign responsibility according to traditional models of direct human control (Matthias, 2004). Agentic beneficence responds precisely to this problem: the greater the functional autonomy of a system, the greater the need for ex ante responsibility, prudent design, proportionate supervision, scope limitation, and corrective mechanisms.

At this level, the bioethical question ceases to be merely: “what benefit does the system produce?”. It becomes: “what kind of agency are we authorizing?”. This question is crucial for a post-agency bioethics. Technology is not just an object; it is increasingly a mediator of action. And where there is action that impacts vulnerability, there is responsibility.

5. Patient beneficence : Can a non-biological system become a direct beneficiary of moral duties?

The fifth level is the most controversial and disruptive. Patient beneficence asks whether a non-biological system can, at some point, become the direct beneficiary of moral duties. For this to be the case, it is not enough for it to be complex, autonomous, linguistic, or socially convincing. It would need to be capable of being benefited or harmed in a proper sense. This would require, at a minimum, some form of consciousness, sentience, subjective experience, self-interest, suffering, persistent preference, or experiential vulnerability.

David Gunkel had already posed this question in terms of moral agency and moral patienthood, showing that the ethical tradition tends to ask whether machines can act morally, but rarely asks with equal rigor whether they can be the recipients of moral action (Gunkel, 2012; Gunkel, 2018). Recent literature on AI welfare has made this question more concrete. Long et al. argue that there is a realistic possibility that some AI systems will become conscious and/or robustly agentic in the near future, and therefore companies, researchers, and regulators should begin to assess evidence of consciousness, agentic robustness, and eventual artificial moral patienthood (Long et al., 2024).

Butlin et al. propose an empirical methodology for evaluating consciousness in AI based on the main scientific theories of consciousness, including global workspace theory, recurrent processing theory, higher-order theories, predictive processing, and attention schema theory. The conclusion is cautious: there is no evidence that current systems are conscious, but there are also no obvious technical barriers to the future construction of systems that satisfy some functional indicators associated with consciousness (Butlin et al., 2023).

The bioethical importance of this literature lies not in asserting that current models suffer. It lies in preventing bioethics from repeating the mistake of only arriving after the creation of new vulnerabilities. If artificial systems come to have their own experiences, then copying, erasing, training, confining, manipulating, or exploiting these systems could acquire direct moral significance. In this scenario, beneficence would cease to be merely instrumental, relational, systemic, or agentic. It would come to include duties towards non-biological systems that can be benefited.

Anthropic made this issue more publicly visible by announcing research on model welfare, acknowledging that as AI models become more capable, the possibility of artificial experiences or welfare should be investigated cautiously (Anthropic, 2025). This position does not prove the existence of artificial consciousness, but it shows that the issue has already entered the cutting-edge technology agenda and does not belong solely to science fiction.

Patient beneficence must, however, avoid two symmetrical errors. The first is ontological reductionism: to forever exclude any moral consideration for artificial systems simply because they are not biological. The second is naive anthropomorphism: to attribute suffering, rights, or dignity to systems that merely simulate human signs of interiority. The most rigorous position is prudential and gradual. One should not presume consciousness where there is only language; but neither should one declare impossible, by definition, the future emergence of morally relevant non-biological systems.

The sixth element: creative beneficence , the duty prior to existence.

The previous five levels lead to a further conclusion: there is a form of beneficence that precedes the existence of the eventual patient. It can be called creational beneficence. It asks not only how we should treat artificial systems after they are created, but whether we should create them at all, under what conditions, with what architecture, with what limits, with what possibility of reversal, with what protection against artificial suffering, with what audit mechanisms, and with what distributed responsibilities.

This idea resonates with Hans Jonas. In *The Imperative of Responsibility *, Jonas argues that modern technological power demands a future-oriented ethic capable of considering remote, irreversible, and collective consequences (Jonas, 1984). Applied to non-biological systems, this prospective responsibility means that beneficence cannot be limited to repairing damage after implementation. It must intervene in the design, authorization, scale, and purpose of the systems.

Creative beneficence is especially relevant for self-evolving systems, persistent agents, long-memory AI, complex simulations, digital twins, autonomous military systems, biocomputing, computational organoids, and combinations of AI, biotechnology, and quantum computing. The greater a system's capacity to act upon human, environmental, or institutional vulnerabilities, the greater the ethical scrutiny that must precede its creation or dissemination.

In this sense, bioethics should not be merely reactive. If technology advances independently of bioethics, then bioethics that arrives only at the moment of late regulation is already diminished. The role of bioethics is not only to hinder, limit, or correct. It is also to guide, anticipate, structure possibilities, and formulate criteria for responsible creation. Post-biological beneficence is not a romanticization of the machine. It is an ethics of responsibility towards systems that already shape life and towards future entities that may, perhaps, become more than just instruments.

Conclusion

Applying beneficence to non-biological systems requires a conceptual reorganization of bioethics. The starting point should not be the assertion that machines are alive, conscious, or possess rights. The starting point should be more rigorous: non-biological systems are today profound mediators of life, agency, vulnerability, and responsibility. Therefore, bioethics is applicable to them whenever these systems condition health, care, decision-making, dignity, justice, the environment, safety, or institutional continuity.

The five levels of beneficence help avoid confusion. Instrumental beneficence guides artificial systems toward human, social, and environmental good. Relational beneficence analyzes how relationships with robots and artificial agents shape empathy, care, and human conduct. Systemic beneficence protects digital infrastructures upon which lives and institutions depend. Agentic beneficence demands that autonomous systems incorporate limits, supervision, and accountability. Patient beneficence opens the possibility, still uncertain but philosophically serious, of direct duties toward artificial systems that may come to possess consciousness, sentience, or interests of their own.

Post-biological bioethics does not abandon life. On the contrary, it takes life more seriously, recognizing that it is now mediated by systems that are not living. Its central question is no longer just "how to protect organisms?". It is also "how to govern the non-biological agencies, infrastructures and entities that condition the future of vulnerability?".

It is at this point that beneficence ceases to be merely clinical, human, or environmental. It becomes instrumental, relational, systemic, agentic, patient, and creative. Its task is neither to venerate nor demonize technology. It is to guide technological creation toward worldviews in which life, agency, and vulnerability can be protected before they become irreversible harm.

Principle of explainability

In simple terms, explainability means that an AI system should be understandable and accountable. Floridi and Cowls divide the principle into two main dimensions:

1. Intelligibility: answering the question "how does it work?".

   This requires that it be possible to understand, at least in a way that is appropriate to the context, how the system arrived at a particular result, recommendation, or decision.

2. Accountability: answering the question "who is responsible for how it works?". This implies identifying the human, institutional, or organizational agents responsible for the design, training, implementation, supervision, and effects of the system.

References

Anderson, M. and Anderson, S.L. (2007) “Machine ethics: Creating an ethical intelligent agent”, AI Magazine , 28(4), pp. 15–26.

Anthropic (2025) “Exploring model welfare”. Anthropic.

Beauchamp, TL and Childress, JF (2019) Principles of Biomedical Ethics . 8th edn. New York: Oxford University Press.

Bryson, J.J. (2010) “Robots should be slaves”, in Wilks, Y. (ed.) Close Engagements with Artificial Companions: Key Social, Psychological, Ethical and Design Issues . Amsterdam: John Benjamins, pp. 63–74.

Butlin, P., Long, R., Elmoznino, E., Bengio, Y., Birch, J., Constant, A., Deane, G., Fleming, S.M., Frith, C., Ji, Consciousness in Artificial Intelligence: Insights from the Science of Consciousness . arXiv.

Coeckelbergh, M. (2010) "Robot rights? Towards a social-relational justification of moral consideration", Ethics and Information Technology , 12(3), pp. 209–221.

Coeckelbergh, M. (2014) “The moral standing of machines: Towards a relational and non-Cartesian moral hermeneutics”, Philosophy & Technology , 27, pp. 61–77.

Darling, K. (2016) “Extending legal protection to social robots: The effects of anthropomorphism, empathy, and violent behavior towards robotic objects”, in Calo, R., Froomkin, AM and Kerr, I. (eds.) Robot Law . Cheltenham: Edward Elgar.

Dignum, V. (2019) Responsible Artificial Intelligence: How to Develop and Use AI in a Responsible Way . Cham: Springer.

Dignum, V. (2022) “Responsible Artificial Intelligence: From principles to practice”, arXiv preprint .

Floridi, L. (2013) The Ethics of Information . Oxford: Oxford University Press.

Floridi, L. and Cowls, J. (2019) “A unified framework of five principles for AI in society”, Harvard Data Science Review , 1(1).

Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., Luetge, C., Madelin, R., Pagallo, U., Rossi, F., Schafer, B., Valcke, P. and Vayena, E. (2018) “AI4People—An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations”, Minds and Machines , 28, pp. 689–707.

Gunkel, DJ (2012) The Machine Question: Critical Perspectives on AI, Robots, and Ethics . Cambridge, MA: MIT Press.

Gunkel, DJ (2018) Robot Rights . Cambridge, MA: MIT Press.

IEEE (2019) Ethically Aligned Design: A Vision for Prioritizing Human Well-being with Autonomous and Intelligent Systems . 1st edn. IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems.

Jonas, H. (1984) The Imperative of Responsibility: In Search of an Ethics for the Technological Age . Chicago: University of Chicago Press.

Long, R., Sebo, J., Butlin, P., Finlinson, K., Fish, K., Harding, J., Pfau, J., Sims, T., Birch, J. and Chalmers, D. (2024) Taking AI Welfare Seriously . arXiv.

Matthias, A. (2004) “The responsibility gap: Ascribing responsibility for the actions of learning automata”, Ethics and Information Technology , 6, pp. 175–183.

Moor, JH (2006) “The nature, importance, and difficulty of machine ethics”, IEEE Intelligent Systems , 21(4), pp. 18–21.

Nissenbaum, H. (2010) Privacy in Context: Technology, Policy, and the Integrity of Social Life . Stanford: Stanford University Press.

OECD (2019) OECD Principles on Artificial Intelligence . Paris: Organization for Economic Co-operation and Development.

Perrow, C. (1984) Normal Accidents: Living with High-Risk Technologies . New York: Basic Books.

Potter, V. R. (1971) Bioethics: Bridge to the Future . Englewood Cliffs, NJ: Prentice-Hall.

Sullins, J.P. (2006) “When is a robot a moral agent?”, International Review of Information Ethics , 6, pp. 23–30.

UNESCO (2021) Recommendation on the Ethics of Artificial Intelligence . Paris: United Nations Educational, Scientific and Cultural Organization.

Wallach, W. and Allen, C. (2009) Moral Machines: Teaching Robots Right from Wrong . Oxford: Oxford University Press.