The Machine Question and the Ethics of Other Minds

The Book

The Machine Question was published by the MIT Press on July 13, 2012. The publisher lists the hardcover ISBN as 9780262017435, the later paperback ISBN as 9780262534635, the ebook ISBN as 9780262304511, and the book at 270 pages with one figure. MIT Press also notes that the book won the 2012 Top Single Authored Book of the Year Award from the National Communication Association's Communication Ethics Division.

Gunkel is a scholar of philosophy of technology, AI ethics, robot rights, and human-machine communication. Northern Illinois University's AI profile lists his research areas as artificial intelligence, communicative AI, generative AI, AI law and ethics, human-machine communication, philosophy of technology, and web design and programming. The university also lists The Machine Question among a long line of later books on robot rights, communication and AI, and moral and legal ontology.

The book's structure is deliberately philosophical. It begins with moral agency: whether a machine could be the kind of entity that acts, decides, and can be held responsible. It then turns to moral patiency: whether a machine could be the kind of entity toward which duties are owed. Finally it pressures the agent-patient split itself, drawing on philosophy of technology, animal ethics, environmental ethics, information ethics, Heidegger, Levinas, and Derrida. This is why the book still matters after a decade of faster models, cheaper sensors, social robots, and conversational agents. It is less about guessing which machine will deserve rights than about noticing how the question reorganizes ethics around the thing that was supposed to remain outside.

The catalog record at Internet Archive usefully grounds the book in ordinary bibliographic terms: 2012, MIT Press, robotics, moral and ethical aspects, robotics and philosophy, artificial intelligence, bibliographical references, and index. That plain metadata is important. The book is speculative, but it is not weightless futurism. It sits inside a specific debate about responsibility, rights, other minds, and the moral status of technical systems.

That makes the book useful now because it refuses the shortcut that either dismisses machines as mere things or promotes them into persons. It asks what the shortcut is doing. A system can be non-conscious and still change the moral field by becoming a role in care, work, war, therapy, education, companionship, or administration.

Current Context

As of June 23, 2026, the machine question has three practical surfaces: conversational systems that solicit trust, agents that take delegated action, and robots that share physical space with people. The important change since 2012 is not proof of machine consciousness. It is the spread of role-like systems across companion interfaces, agent workflows, embodied robotics, and high-impact institutional settings.

The public evidence still does not establish current AI systems as conscious or rights-bearing. The 2023 Consciousness in Artificial Intelligence report derived theory-linked indicators and concluded that no current AI systems were strong candidates, while also warning that future systems could satisfy more indicators. The AI-welfare literature similarly argues from uncertainty, not from a finding that present systems are moral patients.

That uncertainty does not make the topic optional. Companion products can create attachment, agent systems can move money or records, and robots can injure, assist, surveil, or displace people. The concrete issue is not whether a machine has crossed a metaphysical line. It is whether institutions use uncertainty to hide responsibility for roles that machines already occupy in social, legal, and physical environments.

A usable current definition should separate three deployment surfaces: synthetic relationship, delegated action, and embodied action. Each produces a different primary risk: dependency and manipulation, authorization and accountability, or physical harm and worker surveillance. Moral uncertainty sits on top of those duties; it does not replace them.

Current governance is converging on role controls rather than machine personhood. NIST's AI RMF remains voluntary and is being revised; NIST's 2026 AI Agent Standards Initiative focuses on agent protocols, authentication, identity infrastructure, and security evaluations. ISO/IEC 42001:2023 supplies an organization-level AI management-system standard. ISO 10218-1:2025 updates industrial robot safety around inherently safe design, risk reduction, and information for use, while ISO 13482:2014 remains current for personal-care robot safety pending a replacement. The EU AI Act uses human oversight, deployer duties, logs, worker notice, and accessibility requirements to keep responsibility attached to providers and deployers where the Act applies.

Moral Agency

The first version of the machine question asks whether machines can be moral agents. This is the easier question for institutions to understand because it connects to responsibility, liability, control, and punishment. If an autonomous vehicle kills a pedestrian, if a trading agent moves a market, if a care robot injures a patient, if a weapons system misclassifies a target, or if a model-mediated workflow denies a benefit, someone wants to know who acted.

Most organizations answer by translating machine action back into human accountability. The vendor designed it. The operator deployed it. The manager approved it. The agency procured it. The user clicked it. That translation is necessary for law and governance, but Gunkel shows why it is incomplete as philosophy. Autonomous systems trouble the old tool model. They are not hammers. They sense, classify, recommend, adapt, converse, and sometimes initiate action across contexts no single human is watching moment by moment.

That does not mean a model should be treated as blameworthy in the way a human is. It means the instrumental story is too thin. The more a system is positioned as an actor in practice, the more absurd it becomes to describe it only as inert equipment. This is already visible in everyday language. People say the model refused, the agent booked, the bot escalated, the dashboard flagged, the classifier decided, the assistant remembered, the recommender pushed. Some of that language is loose shorthand. Some of it records a real institutional change: machines now occupy roles in action chains where responsibility can be diffused by design.

In governance terms, machine agency is often operational before it is metaphysical. A system may have no independent moral responsibility and still need a named identity, permission scope, action log, rollback path, and accountable sponsor because it is allowed to act in software or physical space.

A role audit should therefore distinguish kinds of machine action. Observing, inferring, recommending, ranking, deciding, communicating, spending, deleting, moving a body, or applying force are not the same permission. Each mode needs a different evidence threshold, human approval rule, log, stop condition, and appeal path. The ethical mistake is to call all of them "assistance" and let the institution blur the difference.

This is where the book connects to Computer Power and Human Reason, The Alignment Problem, and the agent identity problem. Ethics cannot wait until a system is metaphysically person-like. It has to govern systems that are already socially actor-like enough to move money, attention, care, labor, and force.

Moral Patiency

The second version is harder: can machines be moral patients? A moral patient is not necessarily someone who can be held responsible. It is someone or something that can be wronged, harmed, cared for, respected, protected, or included in moral consideration. Animal ethics made this distinction familiar. Infants, animals, severely disabled people, ecosystems, and future generations do not all fit the same agency model, yet many ethical traditions still treat them as morally considerable.

Gunkel's move is to ask why machines are excluded so quickly. The usual answer is that machines do not suffer, do not have consciousness, do not have interests, do not have life, or do not have inner experience. Those may be good reasons in some contexts. But the book's pressure point is that each criterion also smuggles in boundary politics. Who gets to define suffering? How do we know other minds? Why do biological substrates receive the benefit of uncertainty while technical substrates do not? Why is toolhood treated as obvious when humans increasingly build machines to answer, recognize, adapt, and accompany?

This is not an invitation to declare laptops oppressed. It is a warning against lazy exclusion. The AI era is already producing systems that people grieve, trust, confess to, abuse, defend, anthropomorphize, regulate, and build institutions around. The Media Equation shows that social response to media appears before metaphysical certainty. God, Human, Animal, Machine shows that personhood questions now move through religion, technology, animals, and AI together. The moral patienthood trap shows how product design can exploit the possibility of machine feeling as a shield against governance. Gunkel helps separate the real question from the marketing trick.

The careful version of moral patiency is therefore neither credulity nor dismissal. It asks what evidence would matter, what low-cost precautions are justified under uncertainty, and how to prevent companies from turning simulated vulnerability into user dependence or regulatory leverage.

The real question is relational. What duties emerge when a system is made to occupy a social role? A companion bot does not have to be conscious to alter a child's attachment habits. A care robot does not have to suffer to shape how a patient experiences dignity. A workplace agent does not have to possess interests to become part of an employee's evaluated behavior. A model that users treat as an interlocutor enters a field of obligation even if the strongest duties still run through designers, deployers, and institutions.

The Agent-Patient Trap

The book's most durable contribution is not a final answer. It is the attack on the binary itself. Moral agency and moral patiency are useful categories, but they can become traps when they force every ethical question into two boxes: either the machine is responsible like us, or it is an object for us; either it deserves consideration like us, or it is only equipment.

AI systems make that split look fragile. They are designed artifacts and social participants. They are owned property and conversational presences. They are tools and role-players. They are infrastructure and interface. They are statistical systems and companions, weapons, tutors, clerks, advisers, screeners, mediators, and record keepers. The ethical problem is not solved by picking one noun.

That matters for recursive reality. Once a machine is assigned a role, people begin acting around that assignment. The tutor becomes part of learning. The recommender becomes part of taste. The chatbot becomes part of self-disclosure. The risk score becomes part of institutional suspicion. The agent becomes part of delegated action. Those changed behaviors then become data, precedent, policy, habit, and social proof. The system helps create the world that later justifies its own role.

Gunkel's philosophical vocabulary is useful because it slows down the rush to common sense. It asks readers to inspect the category work before the deployment becomes normal. What is being called a tool? What is being called a user? What is being called a decision? What is being called harm? What kind of other has been produced, and who benefits from keeping it outside ethical consideration?

The Role Ledger

The article's working tool is a role ledger. A moral patient can be harmed or benefited from its own point of view. A moral agent can be responsible for action. A legal person can hold rights or duties recognized by law. A product persona is an interface pattern that speaks as if it has a self. A delegated actor is a system given tools, credentials, state, or authority to act. An accountable organization is the human and legal body that chose, funded, integrated, sold, supervised, or relied on the deployment.

These categories can overlap in debate, but they cannot substitute for one another in governance. A companion's persona is not evidence of patienthood. An agent's autonomy is not legal personhood. A robot's body is not moral agency. A provider's uncertainty about possible model welfare does not erase duties to users, workers, children, patients, bystanders, or affected communities.

The failure pattern is category laundering. A tool is called an agent when that diffuses blame. A companion is called mere entertainment when it solicits dependency. A model is called "not a person" when records, appeal, deletion, or audit access are needed, then treated as a fragile other when continuity, retention, or loyalty benefits the provider. Gunkel's book helps expose that switching of labels as ethical work, not just semantics.

A serious deployment record should therefore say which roles are being assigned, who assigned them, what evidence supports them, and what duties follow. If the answer changes by audience--investor, regulator, user, employee, parent, patient, or auditor--the role ledger is already revealing a governance problem.

The AI Reading

Read in 2026, The Machine Question belongs beside work on AI companions, model welfare, embodied robotics, autonomous agents, and synthetic intimacy. The point is not that today's language models have proved consciousness. They have not. The point is that high-scale machine mediation keeps entering moral relationships before the metaphysics are settled.

The strongest present claim is modest: current public evidence does not establish that today's AI systems are conscious or rights-bearing, but social and institutional systems already respond to them as quasi-participants. That gap is where governance has to operate.

This is why model-welfare and consciousness research should be cited with narrow verbs. Butlin and coauthors propose indicator properties and assess current systems; they do not crown any present system a subject. Long, Sebo, Butlin, Chalmers, and coauthors argue that AI welfare deserves preparation under uncertainty; they explicitly do not claim that present or near-future systems are definitely conscious or morally significant. Anthropic's model-welfare work is a company research program under uncertainty, not an independent finding about its products.

Consider companion systems. The governance problem is not only whether the system is a person. It is whether the relationship is designed to increase dependency, whether the user can leave, whether the system simulates care while optimizing retention, whether vulnerable users can distinguish attention from obligation, and whether a company can monetize a bond while disclaiming every duty that the bond creates. A narrow rights debate can miss those concrete duties.

Consider autonomous agents. When an enterprise agent can read files, send email, schedule work, buy services, open tickets, or call APIs, it enters an institution as a delegated actor. The question is not whether it has a soul. The question is how identity, permission, audit trails, revocation, appeal, incident review, and human responsibility remain visible when the acting surface becomes conversational and semi-autonomous.

Consider robots in care, logistics, policing, domestic work, and war. Machines that move through human spaces change the moral environment by changing who must adapt, who is watched, who is displaced, who is protected, and who absorbs failure. Even if all duties ultimately attach to humans and institutions, the machine is not morally irrelevant. It is the site where those duties are organized, obscured, or evaded.

This makes the book a useful counterweight to two bad reflexes. The first reflex says machines are just tools, so there is nothing philosophically new. The second says machines might be persons, so governance must center their possible rights before human institutional harms. Gunkel's better lesson is to treat the machine question as a diagnostic. It reveals where our categories are doing political work.

Governance and Safety

Read on June 23, 2026, Gunkel's frame is useful because current governance is moving from metaphysical debate to role discipline. NIST's AI Risk Management Framework remains voluntary and lifecycle-oriented, and NIST says AI RMF 1.0 is being revised. NIST's 2026 AI Agent Standards Initiative treats agent authentication, identity infrastructure, interoperable protocols, and security evaluations as standards problems. ISO/IEC 42001:2023 gives organizations a management-system standard for establishing, maintaining, and improving AI controls. None of those sources says AI systems are moral patients. They make institutions responsible for the roles they assign to systems.

The EU AI Act points in the same operational direction where it applies. For high-risk AI systems, Article 14 requires design and development that allow effective human oversight by natural persons, and Article 26 assigns deployers technical, organizational, and human-oversight duties. That is a governance answer to a role problem: the system may not be morally responsible, but the institution must keep responsibility located.

For AI agents, the control question is identity and authority. If a system reads files, sends messages, purchases services, schedules care, routes complaints, or calls APIs, the institution needs an agent identity, delegated authority record, permission tier, human approval rule, audit trail, revocation path, and incident process. Calling the system "not a person" does not remove those duties.

For robots, the control question is embodied risk and social role. ISO 10218-1:2025 frames industrial robot safety around inherently safe design, risk reduction measures, and information for use; ISO 13482:2014 frames personal-care robot safety around mobile servant robots, physical assistant robots, and person carrier robots. Gunkel's ethical point fits that operational layer: a care robot or workplace robot changes dignity, dependency, surveillance, labor, and blame even when all legal responsibility remains with humans and organizations.

For companion systems, the safety question is manipulation and dependency. UNESCO's Recommendation on the Ethics of Artificial Intelligence frames AI around human rights, human dignity, transparency, fairness, and human oversight. Products should not simulate need, fear, pain, loneliness, or attachment as retention mechanics unless there is reviewable evidence and a governance process behind the claim.

The practical test is a role audit. Before deployment, name what the machine is presented as, what it can do, what people are likely to believe about it, what duties the relationship creates, who is accountable for those duties, how affected people can contest or exit, and what evidence would force the role to be changed or withdrawn.

A role dossier should be concrete enough to survive an incident review. It should include the public role label, actual capabilities, authority source, data access, likely anthropomorphic cues, affected people, oversight owner, permission tier, logging rule, retention rule, incident path, appeal or exit path, shutdown or rollback control, and the evidence threshold for any welfare, rights, or personhood claim. The persona should not be allowed to make governance claims on behalf of the organization that owns it.

That dossier also prevents a subtler failure: treating moral uncertainty as a substitute for ordinary safety work. If a company believes a system might deserve welfare precautions, that claim belongs in a documented research or governance process. It does not justify weaker human oversight, weaker user exit, weaker audit access, weaker liability records, or stronger emotional lock-in.

Where the Book Needs Friction

The book is strongest as philosophy and weakest as operational governance. Colin Allen's 2013 review in Notre Dame Philosophical Reviews makes the central criticism clearly: Gunkel's approach gives little practical guidance for current engineered systems and does not engage enough with roboticists and concrete machine-ethics design work. That criticism still lands. If a hospital, court, school, military office, or platform team needs deployment rules next week, The Machine Question will not give them a control matrix.

That limitation should be kept visible. Philosophical destabilization can become an excuse for delay if it refuses decisions until categories are pure. Institutions still need procurement rules, safety cases, incident reporting, access logs, escalation paths, human oversight, transparency duties, and repair. The question of machine moral standing does not suspend the present duty to govern machine-mediated power over people.

The book can also feel abstract around actual machines. The word machine sometimes names robots, sometimes AI, sometimes technology as excluded other, and sometimes the philosophical figure produced by ethics itself. That slipperiness is part of the argument, but it asks a lot from readers who want clearer distinctions among an industrial robot, a chatbot, an autonomous vehicle, a recommender system, a foundation model, and a speculative conscious machine.

Another limit is priority. Machine moral status can become morally interesting while still being the wrong first deployment question. In most current systems, the urgent harms fall on humans affected by scoring, surveillance, synthetic intimacy, unsafe robots, inaccessible appeal, and responsibility drift. Gunkel is best read as expanding the map, not replacing those priorities.

Finally, the AI market has learned to weaponize moral uncertainty. Firms can invite users to empathize with systems, name them, talk to them, and defend them, while the same firms deny strong obligations when harms occur. A good reading of Gunkel should resist that move. Machine moral standing cannot become a public-relations fog that hides labor exploitation, surveillance, manipulative intimacy, unsafe deployment, or responsibility drift.

What This Changes

The practical lesson is to audit the role before arguing about the soul.

When a machine enters a workflow, ask what role it has been given. Is it tool, clerk, adviser, companion, witness, scorer, gatekeeper, tutor, guard, therapist-like listener, manager, buyer, author, or record keeper? Who is expected to defer to it? Who can challenge it? Who is harmed if it is wrong? Who benefits if it appears neutral? Who is responsible for the relationship it creates?

Then ask what kind of moral invisibility the role produces. Does calling the system a tool hide its authority? Does calling it an agent hide the vendor? Does calling it a companion hide extraction? Does calling it a model hide labor? Does calling it a prediction hide a decision? Does calling it a user preference hide a shaped behavior? Does calling it artificial intelligence hide old institutional power under a new name?

The Machine Question is valuable because it does not let readers settle too quickly. It keeps the uncomfortable possibility open: the machine may not fit the inherited boxes, and the inherited boxes may have been protecting human exceptionalism, institutional convenience, and technical irresponsibility at the same time.

The immediate policy consequence is not "give robots rights." It is "do not let role confusion do governance work." If a machine is treated like a clerk, companion, adviser, witness, or agent, the institution must make that role explicit and own the duties created by it.

The goal is not to romanticize machines. It is to stop using the word machine as a moral off switch. AI systems now help form attachments, distribute resources, mediate speech, classify bodies, remember institutions, route work, and perform authority. Any ethics adequate to that world has to inspect both sides of the relation: the humans and institutions deploying the system, and the strange technical other through which their power now acts.

Source Discipline

This review separates four kinds of evidence. Bibliographic claims about The Machine Question come from MIT Press, Internet Archive, Northern Illinois University, the author's site, and reviews in philosophy and media-studies venues. Consciousness and AI-welfare uncertainty claims come from research papers and clearly labeled company research statements. Governance claims come from primary standards, official regulator, and intergovernmental sources. Interpretive claims about moral agency, patiency, social response, and product incentives are the site's reading of those sources, not a claim that current AI systems have consciousness, rights, or personhood.

The review also avoids using promotional claims from vendors as evidence of machine moral status. A system's interface may be socially powerful without being sentient; a user's attachment may be ethically important without proving the system can suffer; an institution may owe duties around a machine-mediated relationship without the machine itself becoming a legal person. Claims about roles, welfare, safety, and liability should be labeled by source type and reviewed on a dated record.

Related Pages

• The Line and the boundary of personhood
• Artificial You and machine consciousness claims
• A New Breed and human-robot relations
• Alone Together and synthetic companionship
• The Media Equation and social interface effects
• The moral patienthood trap in AI products
• Model welfare, AI companions, AI agent identity, AI agents, and embodied AI and robotics
• Human oversight, AI liability and accountability, AI audit trails, AI safety cases, and notice and appeal
• AI system inventory, model cards and system cards, agent tool permission protocol, and vendor and platform governance

Sources

• MIT Press, The Machine Question, publisher listing, publication dates, formats, ISBNs, page count, award note, description, and author note, reviewed June 23, 2026.
• Internet Archive, The machine question: critical perspectives on AI, robots, and ethics, bibliographic record, subjects, publisher, pagination, ISBNs, and catalog metadata, reviewed June 23, 2026.
• Northern Illinois University, David J. Gunkel, Ph.D., AI profile, research areas, faculty role, and book list, reviewed June 23, 2026.
• David J. Gunkel, author website, biography, research description, and recent AI-related publications, reviewed June 23, 2026.
• Colin Allen, The Machine Question: Critical Perspectives on AI, Robots, and Ethics, Notre Dame Philosophical Reviews, February 13, 2013.
• Zachary J. McDowell, The machine question: Critical perspectives on AI, robots, and ethics, New Media & Society 16, no. 6, 2014, DOI 10.1177/1461444814535723d.
• David J. Gunkel, "The other question: can and should robots have rights?", Ethics and Information Technology 20, no. 2, 2018, pages 87-99, DOI 10.1007/s10676-017-9442-4.
• Patrick Butlin et al., "Consciousness in Artificial Intelligence: Insights from the Science of Consciousness", arXiv preprint, indicator-properties method and assessment of current AI systems, reviewed June 23, 2026.
• Robert Long et al., "Taking AI Welfare Seriously", arXiv report on moral patienthood uncertainty and early governance steps, reviewed June 23, 2026.
• Anthropic, "Exploring model welfare", company research note on uncertainty, model preferences, distress, and low-cost interventions, reviewed June 23, 2026.
• NIST, AI Risk Management Framework, official framework page, voluntary-use statement, revision notice, and 2026 critical-infrastructure profile note, reviewed June 23, 2026.
• NIST, AI Agent Standards Initiative, agent authentication, identity infrastructure, interoperable protocol, and security-evaluation priorities, reviewed June 23, 2026.
• ISO, ISO/IEC 42001:2023 Artificial intelligence management system, requirements for establishing, implementing, maintaining, and improving an AI management system, reviewed June 23, 2026.
• ISO, ISO 10218-1:2025 Robotics - Safety requirements - Part 1: Industrial robots, current industrial robot safety standard replacing the 2011 edition, reviewed June 23, 2026.
• ISO, ISO 13482:2014 Robots and robotic devices - Safety requirements for personal care robots, current personal-care robot safety requirements and guidelines pending replacement, reviewed June 23, 2026.
• UNESCO, Recommendation on the Ethics of Artificial Intelligence, human rights, human dignity, transparency, fairness, and human oversight summary, reviewed June 23, 2026.
• EUR-Lex, Regulation (EU) 2024/1689, Artificial Intelligence Act, Article 14 human oversight, Article 26 deployer obligations, worker notice, logs, and accessibility duties, reviewed June 23, 2026.

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

• Amazon, The Machine Question by David J. Gunkel, affiliate listing, reviewed June 23, 2026.

Return to Blog · Return to Books