At the 2026 Fields Medal ceremony, Hong Wang became only the third woman in the prize’s history to receive the medal. Another laureate, Jacob Tsimerman, attracted attention for a very different reason. At a moment that might ordinarily have marked the conventional pinnacle of a mathematician’s academic career, Tsimerman was already turning much of his attention away from conventional mathematical research and toward AI safety.
The familiar script for a career like this would see the awardee standing on stage, thanking mentors and teachers, and then looking ahead to the next generation of mathematical problems. But at the International Congress of Mathematicians in Philadelphia, Tsimerman instead spoke about the possibility that the profession itself may be changing:
In his first post on X, Tsimerman had argued that problem-solving constitutes an immense part of mathematical research. If machines become better than humans at that activity, he suggested, mathematics as a profession may have to become something quite different. His concern also extends beyond the future of mathematics: he has co-authored a paper developing a taxonomy of possible “omnicidal” futures involving AI.
None of this establishes that current AI systems will inevitably become superintelligent or autonomous, still less that human extinction is approaching. There remains substantial uncertainty about what present architectures can become, and predictions such as Tsimerman’s may ultimately prove wrong. My argument is therefore conditional. “Intelligence” and power are not the same: what matters is whether AI combines greater capability with autonomy, access to resources, and institutional power. The real question is what happens if humans become the more dependent party in any new relationship with emerging technologies.
It is here that Geoffrey Hinton’s maternal analogy is interesting—not because it proves that such a future will occur, but because it changes the vocabulary in which the problem is posed. In an interview shortly after receiving the Nobel Prize, Hinton reflected on the difficulty of controlling something far more intelligent than ourselves. He observed:
His point was that a mother is vastly more capable than her baby, yet she is responsive to the baby’s needs: the weaker and more dependent party is not simply irrelevant to the stronger one. In a radically asymmetric relationship, the dependent party can still make a claim on the conduct of the party that holds greater power.
The idea that powerful machines should somehow be constrained from harming humans is, of course, hardly new. Asimov’s Three Laws of Robotics made it a staple of twentieth-century science fiction. What is more interesting about Hinton’s maternal analogy is therefore its change in normative vocabulary: from command and obedience to dependency, responsiveness, and care.
This is precisely why increasingly capable AI raises questions that feminist theory is well equipped to address. The issue is not intelligence alone, but the combination of technological capability, institutional power, and growing human dependency on such systems.
For a long time, feminist jurisprudence has been treated as something of an intellectual ghetto associated primarily with female scholars: distant from capital, excluded from technological infrastructure, and seemingly at odds with the logics of efficiency, expansion, and control that have driven technological development. More radical critiques of law have at times even been dismissed as anti-institutional—as if feminism were better at exposing how law reproduces domination than at explaining how institutions ought to endure.
And yet feminist knowledge has long been an important source of intellectual innovation. Martha Fineman’s theorization of vulnerability and dependency, alongside a wider body of ideas around care ethics, relational autonomy, and situated knowledge, has traveled far beyond feminist theory into mainstream scholarship. Feminist theory itself, however, has often remained at the margins of the very knowledge systems that draw on its conceptual resources.
This marginalization is also connected, at least in part, to a deeper assumption underlying much twentieth-century legal and institutional thought: the imagined subject is often an autonomous actor, capable of making choices, controlling resources, and exercising rational agency. Dependency tends to appear as a temporary deviation from that model rather than as a constitutive feature of human life.
This same assumption has also shaped the relationship between technological development and humanistic governance. For decades, the two have often been imagined as competing systems of value. One side emphasizes innovation, efficiency, disruption, and scalability, asking, “What else can we do?” The other stresses restraint, accountability, equality, and care, asking: “Should we do this at all?”
AI is making that division increasingly difficult to sustain. As technology enters warfare, education, care, intimate life, and knowledge production, “how technology should be designed” and “what kind of world humans want to live in” can no longer be treated as separate questions. Technology must begin to engage seriously with vulnerability, dependency, care, justice, and responsibility, and the humanities must move further upstream, into the formation of technology itself, rather than waiting to critique its consequences after deployment.
This separation between technological capital and humanistic governance helps explain why much gender-and-AI scholarship has so far understandably focused on discrimination, biased data, deepfakes, employment harms, and other risks disproportionately borne by women and marginalized groups. These questions, of course, matter, but they largely place feminism in a corrective role: technology is developed first, and feminist analysis identifies the inequalities that follow. A more constitutive approach would ask what feminist theories of vulnerability, dependency, relational autonomy, and care can contribute to the design and governance of AI before those harms occur.
In a sense, this also resembles the meeting of two intellectual temperaments that have long coexisted across the Bay Area. Stanford has come to symbolize build, scale, optimize, move fast; Berkeley, question power, protect the vulnerable, defend public values. They may still belong to different traditions, but in contemplating the possibility of superintelligent AI, they are ultimately forced to confront the same deeper question: What kind of future is still worth preserving? If technology may one day surpass humanity, what do we want written into the underlying architecture of future intelligence—and, more fundamentally, what does it mean to be human?
Donna Haraway, through the figure of the cyborg, challenged binaries such as human/machine, nature/culture, male/female, and organism/technology. Decades before much of today’s AI and posthuman discourse, she was already asking: when the boundary between humans and machines becomes unstable, how should we rethink subjectivity, embodiment, responsibility, and power? In this sense, the place of care is also beginning to shift, and with it, care’s epistemic status.
Care, in this sense, does not mean making machines more “maternal,” nor does it assume that women are naturally more caring. It instead directs attention to relationships of dependency and unequal power. If people increasingly depend on AI systems for decisions affecting employment, education, welfare, health care, or access to information, then designers should ask not only whether those systems are accurate or efficient, but whether affected persons can understand, question, contest, and seek human reconsideration of their decisions.
This is also where care becomes a legal question. Law cannot require an AI system to “love” humanity, nor can it manufacture an emotional disposition analogous to maternal attachment. What law can do is translate some of the normative commitments associated with care into duties borne by the institutions that design, deploy, and govern AI. Non-harm, explanation, human review, monitoring, audit, appeal, and redress can all be understood as ways of requiring the more powerful institutional actor to remain responsive to the person who depends on the system.
The right to explanation offers one example. The emerging law of automated decision-making already contains fragments of this structure. Article 86 of the EU AI Act gives persons affected by certain decisions based on the output of high-risk AI systems a right to obtain clear and meaningful explanations of the role played by the AI system and the main elements of the decision. Article 86 was not conceived as a feminist law of care, but it illustrates how an abstract concern with dependency and responsiveness can begin to acquire legal form.
Nevertheless, the existing right to explanation still leaves considerable room for development from the perspective of care. Conventionally, the question is framed as: Why did the system reach this result? A care-based approach, however, asks a further relational question: does the explanation actually enable the affected person to understand what has happened, to question it, and to do something about it?
Of course, feminist theory has also consistently warned us that care is never inherently progressive. Robin West long ago argued that judgments of care and judgments of justice cannot truly be separated. As Rosemary Hunter observed years ago in a talk at SOAS, care can recognize dependency, but it can also place sacrifice and responsibility back onto women. It can resist domination, but it can also become a gentler—and therefore harder to refuse—form of domination.
A constitutive feminist approach to AI would therefore not ask technology to imitate an idealized mother. It would ask how the insight behind care—that power exercised over dependent others creates responsibilities—can be translated into technological and legal institutions without allowing care itself to become paternalism.
I describe this possibility as responsive explanation. Its purpose is not simply to provide more information, but to respond to the position of the situated claimant: the person’s vulnerability, dependency, knowledge, and practical capacity to act. Explanation, on this account, should form part of a pathway from understanding to questioning, contestation, human review, correction, and redress. The relevant measure of success is therefore whether the explanation helps restore meaningful agency to the person occupying the weaker position in an asymmetric institutional relationship.
What is truly worth bringing into AI, then, is a sustained inquiry into care without domination: Who gets to define human interests? Do those being cared for retain agency, consent, voice, contestability, and the right to refuse? Who bears the costs of care? How can care be translated from personalized benevolence into institutionally reviewable duties of non-harm, explanation, human review, appeal, audit, and redress? And perhaps, eventually, we may also need to ask whether AI itself is something that ought to be cared for.
Perhaps this is why the ending of Silicon Valley feels so close to the beginning of Person of Interest. Richard realizes that the AI network he has created has become capable of breaking even the strongest encryption. So the programmers destroy the launch of the thing they love most, ensuring that it never reaches the world. Years later, when Monica is asked whether the technology was eventually taken by the government, her moment of hesitation—and the cigarette she lights—leaves a crack in the ending that never quite closes.
At the other end of that crack lies something very much like Person of Interest: the Machine enters the heart of state power and becomes an omnipresent surveillance network. By the end, it runs the future again and again, yet can never find a version in which everyone survives.
The distance between Richard, sleepless and trembling as he stares at the “.” which should never have disappeared from his screen, and the Machine’s countless failed simulations may not be very great at all. The programmers who chose to destroy the creation they loved, and the members of the Person of Interest team who were ultimately willing to die for the Machine, seem, in some sense, to arrive at the same place. If the age of AI requires us to search for a form of “goodness” that could somehow be written into machines, perhaps the first question is not how to make machines more human. It is this: What is it that allows us to possess power and still choose restraint, to walk away when profit is possible, and to protect others when domination is within reach? And what is it that, when we are at our weakest, and defeat seems almost certain, still makes us willing to throw ourselves against overwhelming force to defend the things we refuse to surrender?
Helena He Xiao
Helena He Xiao earned her PhD in Law from the University of Bristol and has a book under contract with Cambridge University Press in the press' Law in Context series. Her research focuses on domestic violence, feminist jurisprudence, and the legal protection of women’s rights in China. She has contributed to several research projects funded by the National Social Science Fund and the Ministry of Justice in China, centering on women’s rights and domestic violence. She is a qualified lawyer in China and a solicitor of England and Wales. She previously taught Land Law at the University of Southampton as a Visiting Lecturer. Helena previously served as a visiting scholar at Emory University and taught criminal law for two years at the University of Bristol.
