The Word That Ate Itself
How “tolerance” stopped meaning what you think it means — and why that matters more than you realise
There is a word in the English language that used to describe a specific and demanding virtue. It required you to encounter something you genuinely believed was wrong, to possess the power to suppress it, and to deliberately choose not to. It was not passive. It was not easy. It was the disciplined decision to let a thing you disapproved of continue to exist, because you recognised that the alternative — stamping out every disagreement — was worse.
That word is tolerance.
It does not mean that any more.
What it means now, in most of the places where it actually gets deployed — university conduct policies, platform moderation guidelines, HR training documents, public discourse — is something closer to the opposite. It means: you may not judge. Not that you should refrain from acting on your judgment. That you should not have one. Or if you do, that you should never express it, because expressing it is itself a form of harm.
This is not a small shift. It is not a case of a word gradually drifting to mean something slightly different, the way “nice” once meant “foolish” and now means “pleasant.” What has happened to tolerance is more precise, more deliberate, and considerably more dangerous. The word has been captured.
What Tolerance Actually Meant
To understand what has been lost, you have to understand what tolerance originally required. The classical account — running from Locke through Mill to the contemporary philosopher Rainer Forst — has three non-negotiable components.
First, objection. You must actually disapprove of the thing you are tolerating. If you are indifferent to it, you are not tolerating it. You are ignoring it. Tolerance without disagreement is not a virtue; it is apathy with better branding.
Second, power. You must be in a position to do something about it. A prisoner does not “tolerate” his cell. A person without any capacity to interfere with a practice is not tolerating it; they are simply enduring it. Tolerance requires that you could act and choose not to.
Third, forbearance. You deliberately refrain from using your power to suppress the thing you disapprove of. This is the active ingredient. It is where the moral work happens. You hold your fire, not because you lack ammunition, but because you have decided that a world in which people are free to be wrong is preferable to a world in which only your version of right is permitted.
Remove any one of these and the word collapses. Without objection, it becomes indifference. Without power, it becomes impotence. Without forbearance, it becomes coercion dressed in a gentle vocabulary.
The critical thing to notice is that judgment is not only compatible with tolerance — it is a precondition for it. You cannot tolerate what you have not first evaluated. The entire virtue depends on your having looked at something, decided it is mistaken, and then — precisely because you are committed to something larger than your own certainty — chosen to let it stand.
This is demanding. It asks you to hold two things simultaneously: the conviction that someone is wrong and the restraint not to silence them for it. Most people find this uncomfortable, which is why genuine tolerance has always been rarer than its advocates suggest.
The Pivot
Something happened. I can date the acceleration to roughly the last fifteen years, though the seeds were planted earlier. The shift is visible across three institutional domains — universities, digital platforms, and legislation — and the pattern is identical in each.
The word tolerance stopped governing actions and started governing evaluations.
In the classical framework, tolerance tells you what you may not do. You may not fire someone, expel them, deplatform them, or punish them for holding a view you find repugnant. You may, however, think they are wrong. You may say so. You may argue against them, criticise their position, refuse to endorse it, and explain in detail why you believe they are mistaken. Tolerance constrains interference, not assessment.
In the captured framework, tolerance tells you what you may not think — or at least what you may not express. To call someone’s position wrong is to be intolerant. To evaluate a practice negatively is to violate a norm. The word has turned from a constraint on power into a constraint on judgment itself.
The pivot point is precise enough to formalise: tolerance is captured at the moment when expressing a negative evaluation of a protected position becomes, itself, a sanctionable offence. Not acting on the evaluation. Not discriminating on the basis of the evaluation. The evaluation itself.
How the Capture Works
The mechanism is a specific kind of speech act that I call accusation-as-shielding. It works like this.
Someone says: “You are intolerant.”
That sentence looks like a description. It looks like someone is observing that you have failed to meet a standard. But it is doing something more. It is simultaneously performing two operations. First, it accuses you: you are guilty of a moral failing. Second, it shields the thing you evaluated: your judgment of it is now evidence of your defect, not evidence about its quality. The content of your criticism is never addressed. It does not need to be. Your willingness to criticise at all is the offence.
Notice how this reverses the burden of proof. Under classical tolerance, if you want to sanction someone’s speech, you must demonstrate that they have done something — threatened, coerced, discriminated in some concrete way. Under captured tolerance, if someone claims your speech caused them harm, you must demonstrate that it did not. The accusation does not require evidence of an act. The act of evaluation is the evidence.
This is not hypothetical. It is written into institutional policies. The University of Exeter’s disciplinary regulations make the mechanism explicit: misconduct includes behaviour that “could reasonably be considered” to violate dignity, “irrespective of whether or not distress or harm was intended.” The critical word is irrespective. Intent is irrelevant. The speaker’s reasons are irrelevant. Whether the statement was true is irrelevant. All that matters is whether someone experienced the evaluation as harmful.
Oxford’s Code of Practice states that the university ensures academic discourse is conducted “respectfully.” Whether debate is “respectful” depends in part on its conclusions. A debate that reaches a conclusion experienced as disrespectful by participants has, by the policy’s own terms, failed the respectfulness test — regardless of whether the conclusion is well-evidenced, logically sound, or true.
Cambridge includes an explicit statement that the perception of the recipient determines whether behaviour constitutes harassment, “regardless of whether this effect was intended.” Edinburgh established a working group specifically to manage the “interaction of” free expression with dignity norms. Glasgow places its academic freedom provisions alongside the Equality Act’s harassment definition, creating a structural tension between the right to evaluate and the right not to be evaluated.
The Chicago Principles stand as a counter-example. The University of Chicago’s 2015 report states bluntly that the university’s “fundamental commitment is to the principle that debate or deliberation may not be suppressed because the ideas put forth are thought by some or even by most members of the University community to be offensive, unwise, immoral, or wrong-headed.” Notice what is absent: any evaluative gateway through which the effect of speech on a recipient can override the right of a speaker to evaluate.
The criticism of the Chicago Principles is itself an instance of the capture mechanism at work. Critics argue that the principles “come at the expense of” marginalised students’ wellbeing — framing the protection of evaluative speech as an act of harm, and thereby performing the very accusation-shielding move the principles were designed to resist.
The Three-Stage Pattern
Across every institutional domain I have examined — universities, platforms, legislation — the same three-stage transition occurs.
Stage one: act-constraint. Norms target interference, threats, coercion, and concrete discrimination in access. Judgment is presupposed and unconstrained. Universities discipline students for assault, plagiarism, vandalism. Platforms remove content that constitutes direct threats. Laws prohibit incitement to violence. No one is sanctioned for having or expressing a negative evaluation.
Stage two: evaluative gateway. Norms adopt language that extends violations to the effects of evaluative speech on recipients’ subjective experience. Universities introduce “dignity” and “respect” policies whose definitions include speech experienced as hostile. Platforms expand prohibited content from threats to “harmful stereotypes.” Legislation defines harassment by reference to the effect of unwanted conduct on the recipient’s dignity, regardless of intent.
Stage three: judgment-as-sanction-target. Evaluative judgment itself becomes disciplinable. Academics are investigated for expressing views that cause “harm.” Platform users are suspended for stating positions classified as hateful. Citizens are arrested under communications statutes for speech deemed “grossly offensive” — a standard the House of Lords has interpreted as independent of whether the speech was true.
This is not a slippery slope argument. It is a documented institutional trajectory. The evidence is in the policies themselves, version-tracked and publicly available.
The digital platforms provide the clearest case because their policies are versioned and their changes are dateable. Meta’s hate speech policy expanded from prohibiting direct threats to prohibiting “harmful stereotypes” in August 2020. In September of the same year, it added prohibitions on “dehumanising comparisons” that had previously been permitted. YouTube moved from removing content that “promotes violence” to content that “alleges that a group is superior in order to justify discrimination.” A cross-platform study by Mchangama and Alkiviadou found that the average number of protected characteristics listed in platform policies has expanded steadily, and the categories of prohibited speech have multiplied from act-based to evaluation-based across every major platform they examined.
In January 2025, Meta announced a significant rollback, renaming its category “Hateful Conduct” and permitting forms of expression it had previously prohibited. The rollback is itself evidence that the expansion was recognised, internally, as unsustainable. You cannot draw an infinitely fine boundary through a continuous space of evaluative content and expect your moderators — or your users — to find it. Leaked training materials show the granularity of distinctions moderators were expected to make: permitted phrasing versus prohibited phrasing of functionally identical claims, separated by nothing more than word choice. The expansion is not inferential. It is textual.
Why It Spreads
The most important feature of captured tolerance is not the initial prohibition but the way it propagates. Start with a small, defensible set of protected positions — a finite seed, let us call it. Then observe what happens.
Three propagation mechanisms do the work.
Logical propagation. If critiquing a protected position is prohibited, critiquing its implications is also prohibited, because evaluating an implication requires evaluating the premise from which it follows. If you cannot question the claim, you cannot question anything the claim entails.
Contrastive propagation. If affirming position A is protected, then expressing position not-A constitutes an implicit negative evaluation and is therefore sanctionable. You do not need to mention the shielded claim directly. Merely stating its contrary is sufficient to trigger the accusation.
Reformulation propagation. Rephrasing a prohibited evaluation does not escape the prohibition. This is documented in leaked Meta training materials, which show moderators attempting to draw distinctions between prohibited and permitted ways of saying functionally identical things — an exercise that generates infinitely fine boundaries through a continuous space of evaluative content. The boundaries are not principled. They are classificatory. They cannot hold.
These three mechanisms, operating on a finite seed set, will — given enough institutional time and sufficient connectivity among the propositions in the domain — expand the set of prohibited evaluations until it covers everything. The formal proof is straightforward. The seed set is finite and non-empty. Each propagation rule adds new propositions to the prohibited set. The set of evaluable propositions in any institutional domain is finite. The sequence is monotonically increasing and bounded above. It converges to a fixed point. Under standard connectivity conditions — which hold in any domain where propositions are linked by logical consequence, contrastive relations, and reformulation — the fixed point is the entire set.
The conclusion is stark: under captured tolerance, the space of permissible judgment collapses to the empty set.
This is not a rhetorical flourish. It is a structural consequence. If you shield a finite set of propositions from evaluation and allow the shielding to propagate by logic, contrast, and reformulation, you will — as a matter of formal necessity, not moral extremism — end up in a position where no evaluation is permissible.
The Forstater Test
The case of Maya Forstater illustrates the mechanism with uncomfortable precision.
In the initial Employment Tribunal decision, a judge held that gender-critical beliefs — the view that sex is biological and immutable — were “not worthy of respect in a democratic society.” Notice what happened. The tribunal did not assess whether Forstater had discriminated against anyone. It did not evaluate whether she had harassed, threatened, or coerced anyone. It assessed whether her belief was acceptable. Holding the belief — evaluating a factual claim and reaching a particular conclusion — was itself the offence.
The Employment Appeal Tribunal reversed the decision, ruling that gender-critical beliefs are protected under the Equality Act. The reversal reasserted the distinction between evaluative judgment and coercive action. But the initial decision is a paradigmatic instance of the capture mechanism at work in legal reasoning: the act of evaluating was collapsed into an act of harm, and the evaluation was sanctioned without reference to any concrete act of interference.
Two Kinds of Judgment
The confusion exploited by captured tolerance depends on a systematic equivocation between two senses of the word “judgment.”
The first is discriminatory judgment in the logical sense — the ordinary cognitive capacity to distinguish, evaluate, categorise, and assess. This is what you do when you decide that one argument is stronger than another, that one policy is more likely to work than another, that one claim is better supported by evidence than another. It is the basic operation of rational thought.
The second is moral or juridical judgment — the exercise of authority to impose consequences. This is what a court does when it convicts, what an employer does when it fires, what an institution does when it sanctions.
Semantic capture operates by collapsing the first into the second. Evaluative discrimination — the act of thinking critically about a claim — is redescribed as a moral act with coercive implications. Once this collapse is accepted, any act of evaluation can be treated as an exercise of power, and the accusation “you are intolerant” becomes unanswerable. You cannot defend yourself by pointing to the quality of your reasoning, because your reasoning is itself the offence.
This matters because the first kind of judgment — discriminatory judgment in the logical sense — is not an optional luxury. It is a precondition of every other intellectual activity. Without it, there is no science, no philosophy, no law, no criticism, no accountability, and no honest conversation. A university that prohibits discriminatory judgment in the logical sense has prohibited scholarship. A platform that prohibits it has prohibited thought. A society that prohibits it has prohibited itself from knowing anything at all.
Why “More Tolerance” Cannot Fix This
Here is the part that most discussions of this problem get wrong.
The standard liberal response to intolerance is to call for more tolerance. But if the word has been captured — if “tolerance” now means “you may not evaluate” rather than “you may not interfere” — then calling for more tolerance intensifies the problem. You are not calling for more restraint. You are calling for more prohibition on judgment. Every appeal to tolerance, under capture, is an appeal to extend the veto.
This is why the Paradox of Tolerance, as usually discussed, misses the point. The standard framing asks: should we tolerate the intolerant? That question presupposes that tolerance is functioning normally and the only issue is where to draw the boundary. The actual problem is that the word itself has stopped functioning. The boundary-drawing question is a distraction.
The remedy is not more tolerance. It is the replacement of tolerance-talk with explicit, checkable norms. Instead of a single prestige term that can be captured and weaponised, you need a differentiated vocabulary. You need explicit protections for evaluative speech. You need explicit prohibitions on concrete acts of discrimination. You need explicit procedural standards with clear burdens of proof. And you need these to be stated separately, so that protecting one does not require abolishing the other.
The Higher Education (Freedom of Speech) Act 2023 — which the UK government subsequently halted — represents a partial attempt at this. By explicitly stating that free speech includes the right to express views that others may find offensive, it attempted to decouple the protection of evaluative judgment from the prohibition on discriminatory action. The fact that the Act was halted is itself evidence that the political contestation over whether the capture should be reversed remains unresolved.
The Stakes
The collapse of permissible judgment is not an abstract formal curiosity. It has consequences that show up in measurable institutional behaviour.
Eric Kaufmann’s 2021 survey of academic freedom across the United States and United Kingdom found that significant proportions of academics reported self-censoring their research or teaching to avoid accusations of insensitivity. The self-censorship data is particularly revealing because it documents the perlocutionary effect of the accusation-shielding mechanism. You do not need to sanction many people. You need to sanction a few, visibly, and let the accusation-form do the rest. The anticipation of the accusation is sufficient to reshape behaviour across an entire institution.
This is not fragility. This is rationality. If expressing an evaluation carries the risk of a disciplinary investigation in which you must prove you did not cause harm — irrespective of whether you intended to — and the standard of “harm” is determined by the subjective experience of the complainant, then the expected cost of evaluative speech rises to the point where silence becomes the dominant strategy. People are not cowards for choosing silence under these conditions. They are doing arithmetic.
The result is that the most important function of a university — the systematic evaluation of claims — is degraded precisely in the domains where evaluation is most needed. The topics on which evaluative judgment is most urgently required are the topics on which it is most likely to be accused of intolerance.
What Remains
Civility is not the same thing as silence. You can insist that people argue without personal abuse, without threats, without intimidation — and still permit them to conclude that a position is wrong, a policy is mistaken, a claim is unsupported by evidence. The first set of constraints governs how you speak. The second governs whether you speak. Captured tolerance conflates them. Genuine civic discourse requires them to be kept firmly apart.
The word tolerance described a genuinely difficult virtue: the willingness to live alongside what you believed was wrong, because you valued freedom more than conformity. That virtue required judgment. It presupposed it. Without judgment, tolerance is meaningless — a word describing nothing, applied to everything, enforced by accusation.
Restoring the capacity for public evaluation does not require cruelty. It does not require the abandonment of decency. It requires the recognition that a society which cannot distinguish between evaluating a claim and attacking a person has lost the ability to think clearly about anything at all.
The remedy is not to use the captured word more loudly. It is to stop using it as a substitute for the explicit norms — protection of speech, prohibition of coercion, procedural fairness — that a functioning society actually needs.
You cannot fix a broken tool by pressing harder. You have to pick up a different one.