A person needs a place in which a thought can appear before it is evaluated, a question can be asked before it is assigned a category, and a judgment can mature without an audience. That place need not be romanticized as a sealed chamber. Human beings are accountable for choices, including some chosen intentions, and our minds are formed in language, community, worship, and education. Yet accountability is not the same as continuous inspection. The right to an unobserved mind is the right to develop and exercise judgment without being required to make every intermediate state legible to a predictive system.
This claim is becoming urgent because artificial intelligence no longer waits for a person to publish a finished statement. It predicts interests from hesitation, infers mood from voice, ranks attention from scrolling, and optimizes an interface around the impulses it detects. The result is not merely more advertising. A person can be classified before she has settled what she thinks, steered before she recognizes a desire, and denied an opportunity because a model treats the past as a forecast of the future. Cognitive liberty names the freedom at stake: the freedom to attend, imagine, deliberate, revise, and refuse instrumental capture.
From Privacy to Cognitive Liberty
Privacy ordinarily protects information, spaces, and relationships. Cognitive liberty adds a concern about the conditions of thinking itself. It includes bodily and mental integrity, freedom from non-consensual manipulation, and the ability to form beliefs without coercive interference. It does not imply that all beliefs are equally true, or that no one may challenge an opinion. A teacher may correct an error; a friend may offer a reason; a court may judge an act supported by evidence. The principle limits how authority may reach the person, not whether reasoned disagreement is permitted.
Catholic anthropology gives this freedom a foundation without turning the mind into an untouchable idol. The human person is made for truth and responsible communion. Gaudium et Spes 16 describes conscience as the most secret core where a person is alone with God, while the Catechism teaches both the dignity of conscience and the duty to form it (CCC 1776–1785). “Secret” here does not mean morally irrelevant. The person must examine motives, resist evil, seek counsel, and accept responsibility for deliberate action. It means that this encounter cannot be outsourced to a score that claims to know the person better than the person can know and answer for herself.
Not every thought is a free act. An intrusive image, a reflexive emotion, a dream, or a neurological event may occur without consent. Conversely, a person can knowingly entertain hatred, cultivate deceit, or resolve to harm someone before taking an outward step. Moral responsibility depends on knowledge and freedom, not on whether a thought was private or detectable. The right to an unobserved mind therefore does not excuse wrongdoing. It protects the interior process by which a person recognizes, rejects, confesses, or acts upon what arises within.
Predictive AI and the False Future
Predictive systems are useful when they estimate a bounded event from relevant evidence: a machine may need maintenance, a patient may need follow-up, or a message may be spam. Trouble begins when an estimate of behavior is treated as a revelation of identity. Past clicks become “political temperament”; pauses become “deception”; purchases become “financial character.” The model may find statistical regularities, but it cannot see the singular reasons, new commitments, or grace by which a person changes.
Prediction also has a performative force. If a school assigns extra scrutiny to students predicted to fail, the prediction changes the attention and opportunity they receive. If an employer filters out applicants whose communication style resembles prior departures, the model helps make its own theory of loyalty true. If a lender prices someone as risky because a proxy correlates with hardship, the person may lose the chance to establish a different record. A probability becomes a social destiny when institutions deny the person a meaningful chance to contradict it.
Human freedom is not disproved by the fact that behavior has causes or can sometimes be predicted. Prudence routinely uses experience and likelihood. The moral error is to confuse a useful forecast with a complete account of a person, especially when the person is not told which inference governs the decision. A prediction may justify a question or a proportionate precaution. It should not silently replace evidence of the act, erase the possibility of conversion, or make appeal a performance against an inscrutable profile.
Attention Extraction Is Not Neutral
The attention economy is often described as a competition for minutes, but the deeper issue is competition for the user's capacity to choose what deserves those minutes. Interfaces learn when a person is lonely, bored, angry, or exhausted and then present the stimulus most likely to prolong engagement. Personalization can make a tool more useful. It becomes exploitative when the system treats a known vulnerability as an opening through which to bypass deliberation.
This is not an argument that every notification is mind control. People retain agency, and responsible design can help them remember, connect, or discover worthy work. The concern is structural: a service financed by behavioral extraction has an incentive to keep the person reactive. Infinite feeds, variable rewards, autoplay, and emotionally optimized ranking can crowd out the intervals in which attention returns to its own purposes. A person may still “choose” each tap while the choice architecture makes refusal unusually costly and reflection unusually rare.
Catholic moral language helps name the loss. Attention is not only a resource; it is part of how we love, pray, learn, and perceive the neighbor. The virtue of temperance orders desire rather than abolishing pleasure, and prudence directs action toward a real good rather than the strongest immediate stimulus. A platform need not become a monastery, but it should not profit by systematically weakening the user's capacity for self-command. “Engagement” is not an adequate measure of human flourishing.
Compulsory Legibility
Compulsory legibility occurs when a person must expose or explain an interior state to receive ordinary standing. A worker may be asked to submit sentiment scores to prove commitment. A student may be required to keep a camera and emotion detector active to demonstrate attention. A benefits applicant may be judged by a model that infers sincerity from speech. These demands are especially coercive where refusal means losing education, income, care, or access to a public service.
The problem is not simply that the data may leak. Even perfect security would not justify every demand. The institution is claiming that a person's unexpressed condition is its property or its proper object of evaluation. Yet interior states are ambiguous, changeable, and partly involuntary. A tired face is not laziness; a guarded tone is not dishonesty; an unfocused interval is not proof of indifference. Making these states conditions of participation pressures people to stage an acceptable interior rather than perform an agreed external task.
The alternative is not a ban on all measurement. A teacher can assess work, a clinician can ask about symptoms, and an employer can evaluate responsibilities relevant to a role. The question is necessity and proportionality. What concrete good requires this inference? Is there a less invasive way to reach it? Can the person refuse without penalty? Is the label intelligible and contestable? If the answer is no, the institution should not convert a speculative psychological measure into an eligibility test.
Strongest Objections
The first objection is safety. Predictive tools can identify a risk before harm occurs; attention monitoring may prevent an accident; and early intervention can help a person in crisis. Rejecting all inference would be irresponsible, especially in medicine and transportation.
Cognitive liberty does not require blindness to risk. It requires bounded inference: a defined purpose, validated relevance, minimum collection, qualified human judgment, and a response proportionate to the evidence. An aviation system may monitor fatigue-related conditions tied to safe operation, while still prohibiting unrelated personality scoring and retaining no more than necessary. A clinician may ask direct questions and use a validated assessment with consent, rather than infer moral worth from an opaque consumer model. Safety should justify a careful intervention, not a standing claim on the mind.
The second objection is that people already reveal themselves voluntarily. They post opinions, accept recommendations, and use free services. Why should a company not infer what the customer has chosen to make visible? Because expression is not a transfer of every possible meaning. A photograph shared with friends is not consent to infer a mood for an insurer; a search for a book is not a settled belief; a click is not a declaration that the platform may shape the next hundred choices. Voluntary action can authorize a particular use without authorizing unlimited interpretation.
The third objection is theological: if God sees the heart, why should a human institution be forbidden to seek evidence about it? Divine knowledge is not a template for commercial or governmental power. God is not a vendor optimizing a person's susceptibility, and divine judgment is not a statistical shortcut. Christian humility should make institutions cautious about claiming an access to the heart that even ordinary human love approaches through trust, patience, and the person's own word.
Bounded Inference and Usable Choice
“Specific necessity” should not be a blank label. A proposed inference should identify what it measures, how well it has been validated for this population and setting, who has power over the person, how reversible the consequence is, and what less invasive alternative was considered. A direct or clinically validated assessment may be more appropriate than an inferred emotional state. In safety-critical work, a narrow fatigue measure may justify a temporary intervention; it does not justify a standing profile of character or attention.
Personalization is not automatically coercion. A user may knowingly select recommendations, captions, reminders, or an accessibility feature because it helps her pursue a purpose. The concern is whether refusal remains practical, whether the system exploits dependency or vulnerability, and whether the feature quietly changes from assistance to exclusion or manipulation. Providers should offer an opt-out without loss of essential service, disclose the inference type and uncertainty, and give a human route to reject it where consequences matter.
These judgments need measurement as well as moral vocabulary. A deployment should track compulsion, time-to-exit, notification interruption, appeal outcomes, and effects on children and dependent users, while testing accessibility benefits rather than assuming they are harms. The sources listed here ground conscience and human agency; they do not provide those empirical thresholds. Evidence should be gathered in the domain before a disputed inference becomes a condition of safety, work, education, care, or public standing.
Practical Rights and Design Duties
A right to an unobserved mind should produce concrete protections. First, prohibit the use of inferred mental or emotional states for consequential decisions unless a specific law or clinical duty establishes necessity and safeguards. A vague interest in optimization is not enough. Second, require clear notice when a system predicts an interior attribute, including what evidence, uncertainty, and purpose are involved. Third, provide a human appeal with authority to disregard the inference and correct the record.
Fourth, give people meaningful control over attention. Services should offer chronological or user-selected feeds, quiet defaults, export and deletion, notification limits, and a way to disable personalization without losing essential functionality. The strongest protections belong to children and people whose dependence makes “choice” largely formal. Fifth, separate safety-critical telemetry from advertising profiles. Data collected to keep a driver alert should not become a marketable measure of temperament.
Institutions should also preserve spaces in which no mental disclosure is required. Schools can grade demonstrated learning rather than inferred enthusiasm. Workplaces can set clear outputs and humane accommodations rather than demand continuous emotional performance. Churches and care providers can honor confidential conversation and never imply that spiritual seriousness can be established by a dashboard. Public agencies should design benefit and identity systems around relevant facts, not a generalized demand that applicants make themselves psychologically transparent.
Formation Instead of Optimization
There is a positive alternative to defending the mind only by refusing intrusion. Families, schools, churches, and civic associations can form attention through practices of silence, reading, prayer, conversation, craft, and service. These practices are not escapes from technology; they teach people to use tools without allowing tools to define every good as measurable engagement. A child who learns to wait, a student who can revise a belief, and an adult who can sit with another person without recording the exchange are learning forms of freedom that no predictive score can supply.
Designers have a corresponding duty of humility. Build systems that help users pursue purposes they have chosen, make persuasive features visible, and allow interruption without punishment. Test not only accuracy but whether the product increases compulsion, narrows deliberation, or penalizes people who need privacy. A system should invite attention, not quietly own it. The standard is not whether users can technically leave, but whether they remain capable of directing their lives while using the service.
Synthesis: Freedom Before Legibility
A society needs evidence, expertise, and reasonable prediction. It cannot govern a hospital, classroom, or city by pretending that no one may ever be assessed. But assessment must remain an aid to judgment, not a substitute for the person. The unobserved mind is where questions, temptations, resolutions, and acts of understanding meet. Some of these are morally significant; none should be flattened into a permanent institutional profile.
Cognitive liberty is therefore a condition of responsibility, not a retreat from it. A person can be accountable because she can reflect, change, answer, and make a free commitment. Predictive AI should not turn yesterday's pattern into tomorrow's sentence, and attention extraction should not make distraction the price of participation. Institutions should ask for what their legitimate work requires, leave the rest unobserved, and provide a path to human reconsideration when an inference matters.
The right to an unobserved mind finally protects the possibility of communion itself. We meet one another not as transparent objects but as persons capable of offering a reason, keeping a confidence, changing course, and giving a love that was not predicted. Human dignity does not depend on being unreadable. It depends on never being reduced to what a system claims to read.
Source Notes
The council’s account of conscience as the secret core of the person, alongside freedom, dignity, and the common good.
The Church’s teaching on the dignity and formation of conscience, moral responsibility, prudence, and the freedom required for accountable action.
International guidance on human agency, privacy, transparency, human oversight, and protection against AI systems that manipulate or unduly constrain persons.
Intergovernmental principles for human-centered AI, transparency, robustness, safety, and accountability in systems that make or support consequential inferences.
Editorial revision note
This essay completed one self-editing, cross-essay consistency, and strongest-criticism pass. Its controlling thesis was preserved while objections, qualifications, evidence limits, and practical consequences were strengthened.