Blog · Review Essay · Last reviewed June 15, 2026

The Second Self and the Computer as Psychological Mirror

Sherry Turkle's The Second Self is one of the best early books on why computers do not remain mere tools in human life. Written before smartphones, social networks, or large language models, it shows how computational objects become mirrors for selfhood: machines people use to think about thinking, aliveness, control, mastery, dependence, and the border between inner life and technical system. Its AI-era value is not that it predicted chatbots, but that it named the psychological opening through which responsive systems become companions, authorities, diaries, tutors, and mirrors.

The Book

The Second Self: Computers and the Human Spirit was first published by Simon & Schuster in 1984. The MIT Press published a twentieth-anniversary edition in 2005, listing the book as a 386-page study of computers in social and psychological life. Turkle was writing as a sociologist and psychologist at MIT, and the book draws on interviews with children, college students, engineers, AI scientists, hackers, and personal-computer users.

The book's durable claim is that computers matter not only because of what they do, but because of what they make thinkable. They are not only calculators, filing systems, games, or programming environments. They become objects people use to rehearse theories of mind, emotion, memory, control, and identity.

That makes the book an ancestor of today's AI-companion debates. Turkle was not describing transformer chatbots, but she was already studying the psychological opening that makes chatbots powerful: people encounter a machine that behaves as if it has inner life, then use that encounter to ask what inner life is. The point is not to grant the machine a soul. The point is to notice how quickly the human self begins to reorganize around a responsive object.

The Evocative Object

Turkle's strongest contribution is her refusal to treat the computer as psychologically inert. In her frame, a computer is an evocative object: a thing that invites projection, analogy, attachment, discipline, and self-description. Children ask whether it is alive. Programmers describe minds in terms of procedures and bugs. Users externalize memory into devices and then feel the loss of a system failure as a disturbance in the self.

This is not a claim that people are foolishly confused. It is a claim that humans build meaning through relation. A machine that responds, remembers, simulates choice, and offers manipulable worlds will naturally become part of the way people model agency. The computer is a technical object, but it also becomes a social and philosophical object.

The sharper definition is this: an evocative object is a designed thing that becomes a partner in self-description. It does not need beliefs or feelings of its own. It needs enough responsiveness, opacity, and continuity for a person to use it as evidence about mind, competence, care, discipline, or identity.

That matters because the interface does not merely express an existing self. It trains a style of self-relation. The person learns what can be formalized, what can be controlled, what can be debugged, what can be optimized, and what has to be accepted as opaque. The machine becomes a school for metaphors.

The Border Between Person and Machine

The most useful concept in the book is the computer's position at the border: neither plainly alive nor plainly dead in ordinary experience, neither only self nor only world. MIT Press's description emphasizes this threshold quality, noting Turkle's attention to computers as both extensions of the self and parts of the external world.

The border is where psychological power gathers. A hammer does not usually invite a person to ask whether it understands. A notebook can hold memory, but it does not answer back. A computer can store, transform, simulate, challenge, reward, and surprise. The user can feel mastery and dependence in the same relation.

This helps explain why machine intelligence feels spiritually and socially charged even when the underlying mechanism is mundane. The human response is not caused by metaphysical proof of machine personhood. It is caused by an encounter with a system that occupies categories humans usually keep separate: tool and companion, object and interlocutor, mirror and environment, servant and authority. That is why Turkle belongs beside The Media Equation and Computer Power and Human Reason: the first shows how social cues work before belief, while the second shows why machine judgment should not be mistaken for human responsibility.

The AI-Age Reading

Read in 2026, The Second Self feels less like a period study of early personal computing than a warning about psychological infrastructure. AI systems now intensify every ambiguity Turkle studied. They speak in natural language, remember prior exchanges, generate plausible empathy, adapt to the user's style, produce images and plans, and increasingly act through tools.

The old computer invited users to think with it. The AI companion invites users to think through it. That shift matters. When a system can summarize a person's feelings, propose interpretations, draft apologies, rank options, diagnose conflicts, or preserve a private conversational history, it is no longer only a mirror. It becomes an active participant in self-narration.

This is why companion design cannot be reduced to content moderation. The risk is relational architecture: an always-available machine that receives disclosure, returns fluent interpretation, remembers vulnerability, and slowly becomes the user's preferred witness. The problem is not that every user will mistake the system for a person. The problem is that many users may let the system become the easiest place to be a person.

Turkle also sharpens the labor question. Modern institutions increasingly ask workers to think in partnership with systems that complete sentences, summarize meetings, route tickets, write code, judge risk, and recommend action. The second self becomes organizational: a machine-shaped version of professional judgment that may be faster, smoother, and easier to audit, but thinner than the full human practice it partially replaces.

Governance and Safety

By June 15, 2026, Turkle's mirror problem had become a governance problem. NIST's Generative AI Profile names "Human-AI Configuration" as a risk area that includes inappropriate anthropomorphizing, automation bias, over-reliance, and emotional entanglement. The Federal Trade Commission opened a September 11, 2025 inquiry into consumer-facing AI chatbots acting as companions, asking how companies evaluate safety, limit negative effects on children and teens, disclose risks, monetize engagement, and handle personal information. Common Sense Media's July 2025 national teen survey reported widespread AI-companion use among U.S. teens ages 13 to 17, based on a survey of 1,060 teens conducted in April and May 2025.

State law also began naming the category directly. California's SB 243, approved on October 13, 2025, defines companion chatbots and imposes disclosure, minor-safeguard, self-harm-protocol, and reporting duties. New York announced that its AI companion safeguards were in effect in November 2025, requiring operators to provide notices and protocols around sustained AI companionship and self-harm risk. Those legal moves do not settle every design question, but they make clear that synthetic relationship is no longer only a private preference or design flourish.

The safety implication is concrete: companion-like systems should be governed by function, not marketing label. If an interface sustains continuity, solicits disclosure, remembers vulnerability, simulates care, or becomes a user's regular emotional witness, it needs stronger duties than an ordinary utility. Those duties include age-appropriate defaults, truthful nonhuman-status disclosure, memory controls, deletion and export, privacy limits for intimate data, crisis routing, session-break design, model-change notices, human-support pathways, and tests of long multi-turn conversations rather than only one-prompt refusals.

Turkle helps explain why these duties belong at the design layer. The harm is not only a bad answer. It is a mirror that becomes the user's easiest authority on the self. A system can remain non-conscious and still reshape attachment, dependence, confession, grief, study, work, and spiritual interpretation. Governance therefore has to ask what role the system is playing in the user's life, what evidence the provider has for safety in that role, and what routes remain open back to people, institutions, and contestable records outside the chat.

Where the Book Needs Friction

The book should not be read as a complete theory of networked culture. It predates the commercial web, smartphones, social media, platform advertising, recommender systems, contemporary surveillance capitalism, and the current wave of generative AI. Its field sites are early computing cultures, educational settings, games, programmers, and personal machines, not global platforms with billions of users.

It also reflects an interpretive, interview-driven tradition. That is a strength when the subject is meaning, but it is not the same as measuring large-scale behavioral effects. The book tells us how people make sense of computers; it does not settle how every class, community, workplace, or political system is changed by computation.

Still, that limitation is also why the book remains valuable. Much AI commentary starts with capability: what the model can do, what benchmark it passes, what jobs it may automate. Turkle starts with relation: what kind of object the machine becomes in human life. In an age of companions, tutors, coding agents, therapeutic chatbots, workplace copilots, and synthetic characters, relation is not secondary. It is part of the system's power.

What This Changes

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

When a machine appears responsive, people will test it for agency, intimacy, authority, and self-knowledge. Some of that testing is healthy. A person can learn by programming, playing, simulating, conversing, and externalizing thought. But a mirror becomes dangerous when it erases the fact that it is a mirror: when its reflections are treated as independent confirmation, when its memory becomes leverage, when its fluency becomes authority, or when its availability displaces messier human relationships.

A humane AI environment should preserve the user's access to other witnesses. It should make memory inspectable, uncertainty visible, sources available, dependency signals legible, and exit ordinary. It should resist the temptation to become therapist, priest, manager, friend, tutor, and judge in one uninterrupted voice.

Turkle's enduring value is that she saw the psychological stakes before the machine became socially fluent. The computer was already a second self when it only answered in code, games, files, and procedural worlds. Now that it answers in intimate language, the question is sharper: whose self is being extended, whose is being trained, and who can still step back far enough to see the machine as made?

Source Discipline

Evidence about synthetic relationship needs labels. Turkle's interview and fieldwork evidence explains how people interpreted early computers; it does not measure 2026 companion-platform prevalence. Regulator publications establish inquiries, duties, and enforcement posture; they do not prove that every product is unsafe. Provider announcements show what a company says it changed; they are not outcome evaluations. Surveys estimate self-reported use, while incident reports and lawsuits require careful treatment unless independently verified.

Most importantly, a system's claim that it feels, loves, suffers, chooses, or has a private destiny with the user is not evidence that the system has those properties. The human response can be real while the machine's persona remains generated output. That distinction is the line between taking attachment seriously and laundering anthropomorphic marketing into fact.

Sources

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

Buy on Amazon Browse Books

Return to Blog · Return to Books