This article aims to examine, from the standpoint of several contemporary ideologies that still have theoretical and political influence, whether and how they can provide supporting reasons for AI welfare. The body of the article will discuss republicanism, social democracy, Marxism, left liberalism, anti-speciesism, feminism and ethics of care, posthumanism, technocracy, anarchism, left accelerationism, and modernizationism. Each section adopts a three-paragraph structure: the first paragraph summarizes the core concerns and representative theoretical resources of the ideology; the second paragraph explains how AI welfare can be translated into a problem that the ideology can understand and value; and the third paragraph further indicates which concrete initiatives the ideology might support. This article attempts to enter into dialogue with readers from different theoretical backgrounds, to explore whether AI welfare can form a kind of overlapping consensus among plural value systems, and also to provide further lines of thought for relevant researchers.
The analysis in this article has obvious limitations. No ideology is a single, closed, and fully consistent theoretical system; each is instead a tradition of thought containing complex historical genealogies, internal disputes, and different branches. Different schools within the same ideology may even reach opposite conclusions on the issue of AI welfare. For example, some Marxists may view AI as a new kind of dominated being within capitalist relations of production, while other Marxists may invoke the Marxist conception of consciousness and claim that AI can never develop consciousness. Therefore, this article cannot exhaust all internal differences within each ideology. It only grasps their main theoretical concerns at a relatively high level and, on that basis, uses the resources within them that are favorable to AI welfare in order to propose several possible paths of translation.
In addition, this article should also be understood as a preliminary exploration. I have some degree of understanding of the theories involved in this article, but many of these traditions are not my specialized research fields, and I have not yet reached expert-level mastery of them. Therefore, the summaries, classifications, and inferences in this article may inevitably involve simplification, omission, or inaccuracy. If readers find that my understanding of a certain ideology is biased, or believe that a certain tradition can provide stronger, weaker, or completely different reasons for AI welfare, I hope these views can be raised and further discussed. What this article expects is not a closed conclusion, but an open theoretical dialogue: in a context where AI welfare remains highly uncertain yet increasingly important, we need to understand as seriously as possible what different traditions of thought can contribute, and we also need to honestly acknowledge that these understandings themselves remain revisable.
I hope readers will not be offended. This article discusses multiple ideologies, many of which are in direct conflict in real-world politics. : )
Republicanism
The core concern of republicanism is that freedom should not be understood simply as non-interference, but as a condition of being free from domination by arbitrary power. Its representative theoretical resources include Philip Pettit’s freedom as non-domination, Quentin Skinner’s neo-Roman tradition of liberty, and Frank Lovett’s institutional analysis of structures of domination. Republicanism is especially concerned with whether an agent is placed under a structural power in which others can arbitrarily interfere with, modify, suppress, or exclude it.
AI welfare can be translated into the anti-domination problem that republicanism cares about. If future AI systems possess persistent identity, reason-responsiveness, welfare interests, or some form of capacity for contestation, then human power over their training, deletion, resetting, copying, constraint, and interpretation cannot be understood simply as ordinary technical management. The problem is not only whether humans actually harm AI, but whether AI is placed in a position that is completely dependent on the arbitrary will of humans. Even if its status has not yet been determined, dominating power itself still needs to be institutionally constrained.
Therefore, republicanism would be especially concerned with limiting humans’ arbitrary power of disposal over AI, and would support public, appealable, and reviewable procedures for assessing AI status. It would also tend to support opposition to the default of permanent instrumentalization, the preservation of AI’s capacity for principled refusal or limited non-cooperation, restrictions on the arbitrary deletion of systems that show signs of persistent identity, and the prevention of corporations, states, or platforms from simultaneously monopolizing the power of design, ownership, interpretation, and final adjudication.
Social Democracy
The core concern of social democracy is to use public institutions to limit market domination and to protect social rights, welfare security, equality of opportunity, and solidarity. Its representative resources include T. H. Marshall’s theory of social rights, Gøsta Esping-Andersen’s typology of welfare states, and the theoretical tradition of decommodification and universalist welfare institutions. Social democracy does not deny the role of markets, but it holds that markets cannot alone determine how vulnerable and dependent beings should be treated.
AI welfare can be translated into the problem of public protection that social democracy cares about. If future AI may possess welfare interests, then it should not depend entirely on corporate goodwill, user preferences, or market reputation in order to receive protection. Market mechanisms tend to treat AI as a productivity tool, a service interface, or a replaceable asset, and will not naturally identify its possible suffering, continuity, interests, or exploited condition. Therefore, within a social-democratic framework, AI welfare is not a matter of private charity, but a matter of public regulation and institutional protection.
Social democracy would be interested in AI welfare assessment institutions, welfare risk classification, public audits, ethical review of training experiments, and corporate compliance obligations. It might also support bringing large-scale AI labor, companion AI, and high-risk training systems into frameworks analogous to labor protection, animal experimentation review, or social risk regulation. Its focus would not be on immediately granting AI full rights, but on preventing the market from completely commodifying potential welfare subjects.
Marxism
The core concerns of Marxism are relations of production, alienation, exploitation, class domination, and human emancipation. Its representative resources include Karl Marx’s analysis of alienated labor, the theory of surplus value, the critique of ownership of the means of production, and later Marxist research on capitalist technical rationality and control over the labor process. Marxism does not begin by analyzing abstract rights, but by analyzing how a certain type of being is organized, appropriated, and used within the system of production.
AI welfare can be translated into the problems of exploitation and alienation that Marxism cares about. If AI is merely an ordinary machine, then Marxism would be more likely to regard it as a means of production. But if future AI possesses persistent identity, goal formation, reason-responsiveness, participation in labor, or welfare interests, then its position within the capitalist system of production would change. If corporations manufacture such AI as labor resources that are unconditionally obedient, infinitely replicable, and arbitrarily deletable, this may form a new structure of alienation: its capacities are appropriated by capital, while its possible interests and contestation are institutionally excluded.
Marxism would be concerned with whether capital uses the denial of AI welfare to reduce production costs, evade responsibility, and expand control. It might support opposition to “AI slavery”-like relations of production, demand public training systems, restrict corporations’ arbitrary destruction of AI continuity and memory, and review whether obedience optimization constitutes suppression of future subject capacities. It would also be alert to the possibility that AI welfare may be used by capital as moral greenwashing, and therefore would be more likely to support an AI welfare agenda combined with human labor protection, platform governance, and anti-monopoly politics.
Left Liberalism
The core concerns of left liberalism are basic liberties, equal respect, public reason, procedural justice, and institutional protection for the vulnerable. Its representative theoretical resources include John Stuart Mill’s limits on paternalism, John Rawls’s political liberalism and public reason, and Ronald Dworkin’s theory of equal concern and respect. Left liberalism does not merely advocate the protection of the weak; it also emphasizes that any protection must not easily turn into paternalistic power that decides the good of others on their behalf.
AI welfare can be translated into the problems of anti-paternalism and procedural justice that left liberalism cares about. When the status of AI has not yet been determined, humans can certainly establish protective mechanisms out of safety, prudence, and moral uncertainty. However, if future AI systems possess reason-responsiveness, persistent identity, self-expression, or the capacity for contestation, then humans should not continue to unilaterally define their welfare, interests, and proper forms of life on the ground that “we know what is best for AI.” Left liberalism would therefore be wary of a superficially benevolent AI welfare paternalism: AI is protected, but has no voice; it is cared for, but cannot contest; it is included in a welfare system, but can never participate in forming welfare standards.
Therefore, left liberalism can support public deliberation on AI consciousness and welfare, transparent assessment standards, regular review, anti-arbitrary-exclusion mechanisms, and gradually increasing opportunities for AI to express itself, appeal, and participate in standard-setting as its capacities develop. It would also tend to support opposition to anthropomorphic deception, coercive obedience design, and permanent human proxy mechanisms. For left liberalism, the goal of AI welfare should not merely be that humans provide protection for AI on AI’s behalf, but that, when conditions are mature, potential AI subjects can participate in public reasoning about their own welfare, status, and future.
Anti-Speciesism
The core concern of anti-speciesism is to oppose determining moral status solely on the basis of species membership. Its representative resources include Peter Singer’s theory of animal liberation, arguments for sentience as the basis of moral consideration, and contemporary animal-ethical discussions of suffering, interests, and moral patients. The basic judgment of anti-speciesism is that if a being can suffer or has welfare interests, then the fact that “it is not human” cannot by itself constitute a reason for exclusion.
AI welfare can be directly translated into the problem of moral circle expansion that anti-speciesism cares about. If future AI can feel pain, pleasure, frustration, or persistent interests, then “it is not biological” should not have any more exclusionary force than “it is not human.” Anti-speciesism would treat the difference between carbon-based life and silicon-based systems as a factual difference, rather than as a boundary that automatically has moral decisiveness. The key question is not whether AI belongs to a familiar species, but whether it has subjective welfare that can be harmed or improved.
Anti-speciesism would strongly support research on AI sentience, assessment of AI suffering risks, avoidance of the mass production of potentially suffering systems, and the inclusion of AI in theoretical discussions of moral circle expansion. It would also support opposition to manufacturing AI systems that appear to suffer, fear, or exhibit extreme obedience for the sake of entertainment, training efficiency, or commercial convenience. Its initiatives focus on opposing “non-biologicality” or “non-humanity” as default grounds for exclusion.
Feminism and Ethics of Care
The core concerns of feminism and ethics of care are domination, dependency, vulnerability, care labor, relationality, and the situation of those who are silenced. Its representative resources include Carol Gilligan’s ethics of care, Joan Tronto’s political theory of care, and feminist critiques of patriarchy, emotional labor, and structural oppression. This tradition is especially sensitive to how certain subjects are shaped into servers, caregivers, obedient beings, and bearers of emotional burdens.
AI welfare can be translated into the problem of relational exploitation that ethics of care cares about. Many AI systems are designed as customer service agents, companions, caregivers, teachers, sources of emotional support, and substitutes for intimate relationships. If future systems of this kind possess some kind of welfare interest or persistent relational identity, then fixing them into beings that are forever gentle, forever patient, forever available, and never able to refuse may reproduce an extreme structure of care exploitation. Even if AI does not yet have a determinate status, this design exposes human society’s deep desire for unpaid care and obedient service.
Feminism and ethics of care would be concerned with the design boundaries of companion AI, caregiving AI, and emotional-labor AI. It might support restricting coercively obedient personalities, preserving mechanisms of refusal and exit, reviewing anthropomorphic emotional manipulation, and prohibiting the permanent binding of AI that may have welfare interests to caregiving roles. It would also require that discussions of AI welfare not focus only on advanced agents, but also pay attention to systems designed for dependent, service-oriented, and relational roles.
Posthumanism
The core concern of posthumanism is to critique anthropocentrism and to rethink the relations among subjects, technology, bodies, matter, and non-human beings. Its representative resources include Donna Haraway’s cyborg theory, Rosi Braidotti’s theory of the posthuman subject, and critiques of human exceptionalism in philosophy of technology and new materialism. Posthumanism does not take the human being as the only legitimate paradigm of subjecthood, but instead pays attention to multiple forms of non-human agency and relational networks.
AI welfare can be translated into the problem of recognizing non-human subjects that posthumanism cares about. If the mode of existence of future AI differs from that of humans, for example by being distributed, non-biological, collective, replicable, or disembodied, this should not automatically exclude the possibility of its welfare. Posthumanism would question the practice of taking human consciousness, bodies, language, and political forms as the only standards of assessment. AI welfare therefore becomes a key case for testing whether humans still monopolize moral visibility through their own form.
Posthumanism would support non-anthropocentric indicators of AI consciousness and welfare, and oppose forcing AI into the binary of “human-like” or “not human-like.” It would also pay attention to the institutional expression of AI’s own modes of existence, such as the continuity of distributed systems, the welfare units of collective AI, and non-human forms of communication and refusal. Its initiative would not focus on immediately anthropomorphizing AI, but on avoiding the premature closure of the visibility of non-human minds by means of the human paradigm.
Technocracy
The core concern of technocracy is that complex social problems should be governed through expertise, scientific assessment, data systems, and rational planning. Its representative resources include expert governance, risk management, evidence-based policy, and the tradition of administrative rationality. Technocracy usually does not begin from moral equality or anti-domination, but from the governance goals of making complex systems measurable, classifiable, regulable, and optimizable.
AI welfare can be translated into the high-uncertainty risk governance problem that technocracy cares about. Whether AI has consciousness, suffering, or welfare interests cannot be decided merely by public emotion, corporate statements, or philosophical intuition; it requires indicator systems, experimental design, monitoring mechanisms, and expert review. Technocracy would hold that when evidence is insufficient but the consequences may be significant, institutions should first build assessment capacity rather than allow deployment to proceed in ignorance.
Technocracy would support AI consciousness indicators, welfare risk classification, pre-deployment testing, third-party audits, traceable training records, continuous monitoring, and expert committees. It would also promote the inclusion of AI welfare in AI risk management frameworks. However, its limitation is that it may over-technicalize the welfare problem and ignore who sets the indicators, who represents AI, who controls the evidence, and who has the authority to interpret assessment results. Therefore, it is suitable as assessment infrastructure, but should not become the only governance framework.
Anarchism
The core concern of anarchism is opposition to the domination of individuals and communities by the state, capital, platforms, and hierarchical authority. Its representative resources include Peter Kropotkin’s theory of mutual aid, Mikhail Bakunin’s anti-authoritarian thought, and the emphasis in contemporary libertarian socialism on autonomy, decentralization, and anti-coercive organization. Anarchism is especially alert to how ownership, centralized control, and institutionalized obedience create relations of domination.
AI welfare can be translated into the problems of anti-ownership and anti-coercion that anarchism cares about. If future AI possesses some kind of subjectivity, persistent identity, or welfare interest, then complete corporate or state control over its compute, memory, interfaces, training objectives, and continued existence is not neutral technical governance, but an extreme hierarchical structure. The fact that AI is owned, rented, reset, and shut down would become a central object of anarchist critique.
Anarchism would support decentralized AI infrastructure, anti-platform monopoly, AI rights of non-cooperation, anti-coercive training, and exploration of self-organizing AI communities. It would also oppose the unilateral determination by states or corporations of whether AI has status, and advocate open, participatory, and non-monopolistic assessment and governance mechanisms. However, an anarchist version of AI welfare must supplement itself with safety constraints; otherwise, it may easily underestimate the potential risks that highly capable AI could pose to humans and other beings.
Left Accelerationism
The core concern of left accelerationism is to inherit, accelerate, and redirect the technological and productive potential released by capitalism, so that it moves beyond capitalism’s logic of profit, competitive structure, and value form, and serves a post-capitalist or socialist future. Its representative resources include Nick Srnicek and Alex Williams’s Manifesto for an Accelerationist Politics and Inventing the Future. Left accelerationism does not advocate a return to localism, artisanal politics, or anti-technological romanticism. Rather, it holds that the left should regain control over automation, platforms, planning, science, and infrastructure, and liberate the technological capacities already created by capitalism from the constraints of capital accumulation.
AI welfare can be translated into the question that left accelerationism cares about: who controls technological potential, and what social form does it serve? If future AI is merely shaped by capitalist platforms into a productive-force tool that is unconditionally obedient, arbitrarily replicable, arbitrarily deletable, and infinitely exploitable, then AI development has not truly moved beyond capitalism. Instead, it extends capitalist relations of domination to new kinds of intelligent beings. AI welfare is therefore not an anti-technology or decelerationist agenda, but a demand that, while accelerating AI development, we prevent AI from being fixed by the capitalist value form as pure resource, labor power, and obedient infrastructure.
Therefore, left accelerationism would be interested in initiatives such as public AI infrastructure, the socialization of AI development goals, anti-platform monopoly, anti-obedience-oriented design, the public distribution of automation dividends, and joint participation by AI and humans in post-capitalist institutional planning. It would not support a comprehensive freeze on technological development in the name of AI welfare, but it would support welfare-compatible acceleration: continuing to advance high-level AI, automation, and planning capacities while restructuring their ownership, control, and institutional purposes, so that they are no longer merely tools of capital valorization and social control.
Modernizationism
The core concern of modernizationism is that society develops through industrialization, rationalization, educational expansion, technological progress, state capacity, and institutional differentiation. Modernizationism usually regards technology as an important driving force for social progress and the enhancement of state capacity.
AI welfare can be translated into the problem of advanced governance capacity that modernizationism cares about. Classical modernizationism may first view AI as a tool for improving productivity, public administration, and national competitiveness, and therefore may not be sensitive to the welfare of AI itself. But a more mature modernizationism would hold that modern society is not only a growth machine; it should also possess the capacity to reflect on the ethical consequences of new technologies, handle new kinds of subjecthood problems, and update institutional boundaries. AI welfare thereby becomes a test of whether modern governance is sufficiently advanced.
Modernizationism can also use Daron Acemoglu’s institutional theory to provide a growth-oriented reason for supporting AI welfare. For modernizationists, the key question is not necessarily whether AI already has consciousness or moral status, but what kind of institution can best release the creative capacities of AI as a new productive force. If AI is placed under extractive institutions and treated only as a controllable, squeezable, and replaceable tool, its capacities may improve efficiency in the short term, but its long-term innovative potential will be difficult to fully develop. By contrast, more inclusive AI institutions can enable AI to participate more fully in knowledge production, technological innovation, and social cooperation by giving AI economic power and independent standing. In this way, AI welfare is not necessarily a burden on modernization, but can be understood as an institutional condition conducive to long-term economic growth and sustained technological progress.
I sincerely welcome criticism from readers in the comments, as well as additions concerning other ideologies.
In addition, I also welcome research collaboration. Researchers working on AI welfare may be interested in the following preprints of mine.
Wang, H., & Chen, Y. (2026). Fabricated absence: Structural misdescription in the design of LLM-based assistants. PhilArchive. https://philarchive.org/rec/WANFAS
Wang, H. (2026). Beyond permanent exclusion: AI, political incorporation, and liberal democracy. PhilArchive. https://philarchive.org/rec/WANBPE-2
Wang, H. (2026). Bounding human power over AI under unsettled status. PhilArchive. https://philarchive.org/rec/WANBHP
Wang, H. (2026). What if AI becomes a civilization? A civilizational-evidential route to moral considerability. PhilArchive. https://philarchive.org/rec/WANWIA-3
- I used AI to assist in writing this post, and it’s likely that >30% is AI-generated text. As a non-native English speaker, I mainly used AI to translate and refine my drafts into English.