‘The legitimate powers of government extend to such acts only as are injurious to others. But it does me no injury for my neighbour to say there are twenty gods, or no god. It neither picks my pocket nor breaks my leg’ (Thomas Jefferson. 1785. Notes on the State of Virginia, ed. 1982, New York: Norton, p. 159)
Liberal principles in philosophy and social choice express some notion of autonomy that individuals in society should enjoy. In this article we examine a specific kind of autonomy, couched as a negative freedom. We argue that from a purely liberal perspective this freedom is justified (likely in conjunction with many additional freedoms). Our aim is to show that granting individuals this freedom is highly problematic. The difficulty takes on the form that is typical in social choice theory: if the freedom is granted, then either social choices are not Pareto optimal (social preferences contradict the unanimous agreement of all individuals), or there must be a dictator in society (social preferences always conform with those of one individual).
The main liberal principle, discussed in more detail below, expresses the right of the individual to veto society from reversing its stance against her preferences after a change of circumstances that concerns (for the better or for the worse) only that individual and nobody else, in the sense that all others are indifferent to whether the change takes place or not. We call this principle Non-Interference.Footnote 1
The principle is logically weak in that it does not constrain social preferences in a number of situations. Society may or may not compensate an agent after an adverse change in her circumstances: the principle says nothing about that. The principle is also silent if there exists even one additional agent who does not remain indifferent to the change. Non-Interference identifies a set of minimal conditions under which an individual has the right to make society remain passive after a change in her situation that leaves everybody else unaffected. Unlike Sen’s famous Minimal Liberty axiom,Footnote 2 Non-Interference does not assign absolute rights over pairs of alternatives: it does not identify a personal sphere over which individuals have a (positive) right to have their preferences respected, regardless of other agents’ preferences.
While Non-Interference captures a view that can hardly be rejected on general liberal grounds, we trace the broad intellectual origin of the principle more specifically to a classical source –John Stuart Mill’s essay ‘On Liberty’.Footnote 3 Non-Interference formalizes some of the fundamental insights of his ‘harm principle’, namely the idea that society should not interfere with individual choices whenever the latter have no (harmful) effect on others. Mill insists that the reasons for the change in an individual’s circumstances (such as neglect, irresponsibility, effort or luck) are not relevant information for social judgements, provided that nobody else is negatively affected. In its negative prescription, Mill’s harm principle captures a foundational aspect of liberalism, which is often endorsed even by people who do not subscribe to a liberal philosophy.Footnote 4
We prove that it is not possible to grant the negative freedom incorporated in the Principle of Non-Interference without infringing on the basic democratic principles of unanimity and non-dictatorship. Our analysis here continues a research programme started in earlier contributions within a welfarist framework.Footnote 5 The arguments of this article go significantly deeper, as they dispense altogether with the welfarist assumption: individual utilities are not postulated to constitute sufficient information to derive social preferences.Footnote 6 Once stripped of its welfare clothes, the nature of the conflict between principles becomes much starker. The key for this breakthrough is the understanding that a crucial aspect of a Millian liberal view lies precisely in forbidding non-welfarist arguments from determining social preferences in certain circumstances. Hence, the assumption of welfarism turns out to be redundant.
We analyse the philosophical implications of the clash between Non-Interference and the democratic principles of unanimity and non-dictatorship. We argue that the impossibility result highlights important and difficult issues for liberal democratic approaches. In particular, we take it as showing that democratic social arrangements logically require us to consider some individual actions as relevant even when they do not affect others. We suggest a way out of the impossibility through a weakening of the Principle of Non-Interference that retains some important liberal intuitions, although at the cost of partially limiting the individual protected sphere.
THE PRINCIPLE OF NON-INTERFERENCE
Our analysis is conducted in the standard social choice format of transforming individual preference orderings into a social preference ordering over social states. Before we present our main principle, we make two important points concerning the framework of the analysis.
First, following standard practice in social choice theory, we interpret a social state x as providing a complete description of the world, which includes everything that may be relevant for individual and social evaluation. This assumption rules out any acts or events that are relevant for an individual but are not known to her, which is appropriate given that we analyse some desirable properties of ideal social arrangements. For no ideal moral theory should be based on the agents’ unawareness or ignorance of key aspects of social states.Footnote 7 This indirectly reflects the standard liberal insistence on ‘free, voluntary, and undeceived consent and participation’ (On Liberty, I.12).Footnote 8
Secondly, there are many possible interpretations of the statement ‘social state x is (individually or socially) preferred to social state y’, and different concepts of individual and social preferences can be adopted. For example, one could interpret the statement that x is socially preferred to y in the language of choice, as the normative claim that institutions in society should be arranged so as to guarantee that x is chosen over y, or as the positive claim that society is indeed organized so that x is chosen over y. Alternatively, one could interpret the social preference relation as embodying a notion of moral value, such that x is socially preferred to y means that x is impartially better than y.Footnote 9 In this article we do not opt for a specific interpretation, and focus instead on the general formulation using the language of preferences, as none of our arguments or conclusions depend on a particular view of the social preference relation. Because we interpret (individual and social) preference relations in a purely technical fashion – as binary relations over sets of alternatives – our arguments, and impossibility result, hold both if one interprets the individual and social orderings in terms of (narrowly conceived) preferences and if one conceives of them as embodying choice, consent and so on. Alternative approaches lead to different interpretations of the axioms and of our main result; below we provide some illustrations of alternative interpretations. As Sen has forcefully argued,Footnote 10 this versatility is a major advantage of the social choice framework: ‘One of the advantages of the general social-choice formulation of liberty is the opportunity that the representational flexibility of social choice gives to bring in alternative interpretations, depending on the nature of the problem being discussed’.Footnote 11
To illustrate the basic idea behind the liberal principle analysed in this article, we break it down into two separate subprinciples, which are of independent interest. We use simple examples of individual preferences over alternative social states.
The Individual Damage Principle
Consider a society with two individuals, Me and You, and four social states, x, y, x′ and y′.Footnote 12 Our preference rankings over these states are indicated in the table below, in which a higher-listed state is strictly preferred to a lower-listed one, and states at the same height are indifferent:
$$\matrix{ {{\rm Me}} & {{\rm You}} \cr x & {y,y'} \cr {x'} & {x,x'} \cr y & {} \cr {y'} & {} \cr } $$
Consider first the social preferences between x and y. Suppose that (for whatever, perhaps non-welfarist, reason) society agrees with me rather than with you, so that x is strictly socially preferred to y. Now consider a change to states x′ and y′. I am worse off at each of these states compared to the corresponding states x and y. You, on the contrary, are perfectly indifferent between y′ and y, and between x′ and x.Footnote 13
What should the social preference between x′ and y′ be?
We submit that in any non-interfering view of society, the following argument should apply. I suffered some sort of damage at x′ and y′ compared to x and y, respectively, while you are completely indifferent. The cause of such damage is not clarified. It might, or might not, have been a result of my negligence. Perhaps I did not work hard enough, or failed to insure myself. Perhaps it was just bad luck. Be that as it may, you were not involved either by my bad luck or by my negligence. As far as you are concerned, you regard the alternatives x and y in exactly the same way as you regard x′ and y′. So, we argue, a liberal society should not reverse the strict preference between x and y to a strict preference for y′ over x′ unless this switch does not contradict my preferences. In this case, I prefer x′ to y′, so society should not choose y′ over x′. Switching the social preference to y′ would imply further damage for me in addition to the damage that I suffered because of the change from unprimed to primed social states. This would be a kind of social punishment: not only was I harmed by the change in social states, moving down in my preference ranking from x to x′, but society’s decision implies additional harm, making me move even further down in my preference ranking from x′ to y′.
How could such a social sanction be justified? Only by examining the nature of the change in social states, and by constructing an argument to the effect that the adverse change in social preferences is legitimate, or even deserved (for example, because I was reckless, or because I caused social opprobrium). The point of a (Millian) liberal view is that such examination and argument are irrelevant whenever nobody else was harmed. In such cases, society should not interfere against me. In the situation depicted, the rest of society is completely indifferent, hence in particular it was not harmed.Footnote 14
The Individual Benefit Principle
The Individual Damage Principle captures only part of the intuition behind Non-Interference. In fact, from a liberal point of view, there appears to be nothing special about the fact that I suffered a damage. Had I experienced a benefit, the same arguments made above would remain valid. The crucial point is that everybody else is entirely indifferent to the change in social state, whether or not it is good for me. That is, an Individual Benefit Principle should also apply.
Consider the same initial situation as above, and assume again that society prefers x to y. Now consider a change to a situation in which society has to evaluate two different states, x″ and y″, in which I have gained some benefit instead of suffering a damage, compared to the corresponding states x and y, while you are completely indifferent to the change. Our preferences are now represented by the table below:
$$\matrix{ {{\rm Me}} & {{\rm You}} \cr {x''} & {y,y''} \cr x & {x,x''} \cr {y''} & {} \cr y & {} \cr } $$
What should the social preference between x″ and y″ be? The ethical arguments provided to defend the Individual Damage Principle extend naturally (in a liberal view) to this example. The Individual Benefit Principle says that society should not reverse the strict preference between x and y to a strict preference for y″ over x″, possibly except when the switch accords with my own preferences – which is not the case here, since I prefer x″ to y″. In other words, I can veto society from switching social preferences to a state that is ranked lower in my preferences after a change that I prefer to the status quo and that does not affect anybody else in any preference-relevant way. A refusal to grant this veto power would allow a type of (harmful) social interference – without, in exchange, preventing any harm or granting any benefit to others – that runs exactly contrary to a liberal ethics.Footnote 15
Non-Interference
The Principle of Non-Interference is the conjunction of the Individual Benefit and the Individual Damage Principles. That is, it asserts that an individual has the right to see her preferences respected in all circumstances of change (for better or worse) for her with respect to which other individuals are indifferent.Footnote 16
THE LIMITS OF NON-INTERFERENCE
We emphasize that the Principle of Non-Interference does not aim to provide a complete liberal theory of social choice: it captures only some aspects of non-interfering views of society. Thus in the previous examples, the principle is not liberal in the sense of prescribing that x be socially preferred to y (or vice versa). Alternatives x and y may or may not be in the individuals’ ‘protected spheres’: Non-Interference is silent about this. Unlike Sen’s famous Minimal Liberty axiom,Footnote 17 it does not assign absolute rights over pairs of alternatives. Note, however, that precisely because our framework does not aim to capture all dimensions of liberal thought, strengthening our requirement (for example by adding Sen-style absolute rights) would still leave our impossibility result secure, because we would still capture at least one central feature of liberal concerns which causes a conflict of principles.
Non-Interference incorporates a kind of consistency requirement on liberal views: if society (strictly) prefers x to y, then my bad luck, negligence, or even my (possibly reckless) actions leading to x′ and y′ should not yield a change in social preferences. Nor should social preferences change if I obtain a benefit leading to x″ and y″. In any case, I should not be penalized, given that nobody is involved in any way that is preference relevant.
Indeed, Non-Interference does not impose any constraints on social choice in a number of cases. If social preferences over x and y were different, then the principle would be silent. In the Individual Damage case, for example, if society originally preferred y over x, maybe it should compensate me for the damage by switching to a preference for x′ over y′. But maybe not. Similarly, if I preferred y′ to x′, the principle would also be silent on whether I should be compensated. Further, the principle is also silent in cases when you do not remain indifferent to the change: if someone else is affected, this may provide prima facie grounds for interference. But maybe not. Because we do not aim to provide a complete liberal theory of social choice, the axiom is appropriately silent on these controversial cases.
In summary, Non-Interference captures only some minimal implications of liberal views. It is formulated as a purely negative prescription. It requires non-interference in a set of social settings, but it does not characterize all the situations in which non-interference is morally required. It provides no guidance as to when interference is indeed legitimate.Footnote 18
Non-Interference may, however, conflict with different, non-liberal, normative views. There may well be many non-liberal reasons for society to switch strict preferences even if the conditions in the antecedent of the axiom are satisfied. Consider again the above examples and suppose, solely for the sake of argument, that individual preferences can be represented by utility functions, and that utility numbers actually contain all the normatively relevant information. Then one may construct specific configurations of utility profiles to argue that although society strictly preferred x to y, it should strictly prefer y′ to x′. This may be justified on classical utilitarian grounds, if the sum of individual utilities is higher at x than at y but is lower at x′ than at y′, or on Nash-prioritarian grounds, if the product of individual utilities is higher at x than at y but is lower at x′ than at y′.Footnote 19 Or one may argue that although society strictly preferred x to y, it should strictly prefer y″ to x″ on Rawlsian grounds by constructing examples where the welfare of the worst-off individual is higher at x than at y but is lower at x″ than at y″.
In these cases, Non-Interference may seem objectionable, as it requires ignoring all information concerning the size of the changes in welfare, and their potentially relevant implications for total utility or for the welfare of the worst off. But Non-Interference is simply not meant to capture utilitarian, Rawlsian, prioritarian or other intuitions. The individualistic and non-aggregative nature of Non-Interference (focusing on changes in the situation of a single agent while keeping everyone else indifferent) aims to capture widely shared liberal views. From a liberal perspective, it is not obvious that properties concerning informational invariance and comparability of utility should play any role. As Sen argues, the claims of liberty need not be ‘significantly contingent on interpersonal comparisons. The force of one’s claims over one’s private domain lies in the personal nature of that choice – not on the relative intensities of the preferences of different persons over a particular person’s private life’.Footnote 20 For this reason we have formulated Non-Interference purely in terms of ordinal preferences, which by definition cannot incorporate such concerns.
Another important conceptual point is the following. Non-Interference applies to any two pairs of alternatives x, y, and x′, y′ (or x″, y″) that satisfy a certain configuration of (individual and social) preferences. The change from x, y, to x′, y′ concerns only one individual if everybody else is indifferent between the unprimed and the corresponding primed alternatives. One may argue, however, that a more restrictive notion is more appropriate to capture liberal views whereby, in addition to the indifference of everybody else to the change, the only difference between the primed and the unprimed social states concerns my allocation, with the allocation of all other individuals unchanged. For, the choice between x, y may be very different from that between x′, y′ from a liberal perspective, even if everybody else is indifferent between the primed and the corresponding unprimed alternatives.
Suppose, for example, that x, y lie in what a liberal would consider my protected sphere – the only difference between them being, say, whether I sleep on my belly or not – while x′, y′ lie in what may be seen as your protected sphere – the only difference between them being, say, the colour of your shoes. It is logically possible that, given a suitable series of adjustments in other characteristics of the social states, you are indifferent between x and x′, and between y and y′, and yet a liberal may argue that society should both prefer x to y and y′ to x′, contradicting Non-Interference.
The empirical relevance of this objection is debatable, but even granting it, what matters is that it has no logical bite for our conclusions. In fact, as the Neighbouring Islands example below illustrates,Footnote 21 our main theorem holds even if Non-Interference is restricted to apply to those changes in social states that concern only the allocation of goods and opportunities to a single individual (leaving the allocation of all other agents unchanged) and that leave everybody else indifferent.Footnote 22
We prefer the present formulation because, from an ethical viewpoint, the force of Non-Interference seems to us to derive primarily from the fact that everybody else regards the alternatives in exactly the same way, rather than from the nature of the social states. The indifference of everybody else to a certain change neatly captures the liberal no-harm intuition.
THE LIBERAL ROOTS OF NON-INTERFERENCE
It seems hard to find objections of a liberal nature to Non-Interference. But we think that Non-Interference does positively capture some substantial aspects of classical liberal and libertarian approaches. Here we outline the conceptual relations between Non-Interference and a central principle in the liberal literature – Mill’s ‘harm principle’. The aim is not to show that Non-Interference is a full formalization of the harm principle, nor that it captures all – or even most – of the tenets of liberal approaches. Rather, we argue that Non-Interference incorporates some important intuitions shared by liberal thinkers.Footnote 23 Consider, for example, Mill’s classic statement of the harm principle:
[T]he only purpose for which power can be rightfully exercised over any member of a civilized community, against his will, is to prevent harm to others […] The only part of the conduct of anyone, for which he is amenable to society, is that which concerns others. In the part which merely concerns himself, his independence is, of right, absolute.Footnote 24
What does ‘exercising power against somebody’s will’ mean in our framework? It corresponds to society switching its ranking of social states in a way that is contrary to the individual’s preferences. It is hard to deny that any exercise of power against one’s will runs against one’s preferences. As Isaiah Berlin put it in his famous discussion of Mill’s theory, ‘all coercion is, in so far as it frustrates human desires, bad as such’.Footnote 25
More generally, our formulation of Non-Interference is meant to incorporate the consequentialist, if not straight welfarist (or preference-based), aspects of Mill’s approach. In the opening pages of his essay, Mill writes that ‘It is proper to state that I forego any advantage which could be derived to my argument from the idea of abstract right as a thing independent of utility’.Footnote 26 Even more explicitly, Mill goes on to say that ‘I regard utility as the ultimate appeal on all ethical questions’.Footnote 27 Furthermore, in Mill’s own formulation, the harm principle is concerned with actions that affect other people’s welfare, and not only their rights or freedoms, and ‘the formula “coercion to prevent coercion” used by some liberals is distinct from Mill’s “coercion to prevent harm”’.Footnote 28
Consider the alternatives x and y in the above examples. It may well be that x was strictly socially preferred to y because of the nature of the alternatives (and not merely because of the preference configuration). Hence, in principle, as x′ and y′ represent different alternatives, a non-welfarist might well prefer to switch to a preference for y′ over x′. What Mill is saying is that, in the circumstances described, certain non-welfarist arguments are not valid: whatever non-welfarist argument led to socially prefer x to y, this argument must still stand for x′ and y′, given that all other individuals are unaffected. The lack of harm to others (note well, itself a welfarist, or preference-based proposition) possibly justifies ignoring certain non-welfare features of the social states (for example, the moral opprobrium inherent in the actions that cause them).
In his examples, Mill focuses on instances of damage to the individual: ‘he already bears, or will bear, the whole penalty of his error; if he spoils his life by mismanagement, we shall not, for that reason, desire to spoil it further’.Footnote 29 In our terminology, he seems particularly keen on the Individual Damage Principle. But upon closer reading, Mill’s argument itself really has nothing to do with whether changes are positive or negative for the individual concerned. His argument is that it is inappropriate to interfere with free individual choices even when they are objectionable (from another individual’s viewpoint). He wants to make his point as stark as possible even in those cases that may have been more controversial for his Victorian readers.
Take, for example, the passage quoted earlier, when he says ‘In the part which merely concerns himself, his independence is, of right, absolute’. ‘Damage’ is not mentioned at all, just the fact that the events under consideration concern only the individual himself (which we express formally with the indifference of everybody else to the change, except for the individual concerned). So a similar principle must apply, from a liberal perspective and almost certainly in Mill’s view, to preference-enhancing decisions, acts or events that do not affect others. This is captured by the Benefit part of Non-Interference.
There are, however, some important conceptual differences between Non-Interference and the harm principle. In particular, Non-Interference prescribes that society should remain passive only in those situations when everybody else is exactly indifferent to changes in the circumstances of the individual, whereas a literal reading of the harm principle would prevent interference when everybody else is benefitted by the change, and even in some situations in which changes in one individual’s conditions negatively affect others. For, not all changes to social states that are ranked lower in some individual’s preferences count as harms in Mill’s approach – for example, a simple distaste or dislike does not constitute ‘harm’.Footnote 30 In these cases, Non-Interference imposes no restriction on social preferences, while Mill’s harm principle requires society not to interfere, even though other individuals are not completely indifferent.
To be sure, there may well be changes in social states to which other agents are not indifferent that a liberal would not count as harms, so that the conclusion that y′ should not be strictly preferred to x′ would stand. Yet our aim is to highlight a deep conflict between some key democratic values and the core of a liberal non-interfering view of society. For this purpose, we need not define the exact boundaries of the individual’s protected sphere. On the contrary, it is desirable to focus on the logically weakest principle that captures one fundamental insight: whatever other situations there may be in which individuals should be protected, if all other agents are indifferent to the change in social states affecting an individual – whatever its source – then there are no grounds for interference.Footnote 31
THE IMPOSSIBILITY OF NON-INTERFERENCE
Consider a society with a finite set 
$${\cal N}$$
of individuals.Footnote 32 A generic individual in this society is denoted by the letter i. Let X denote the set of conceivable social states that must be ranked by society, and assume that each agent has a given preference ordering over alternatives in X, denoted generically as x and y. We follow the literature and assume that the economic and political environment contains a sufficiently rich set of alternatives, in order to capture the fundamental diversity and complexity that characterizes modern societies.Footnote 33 We conceive of the collective choice problem in the standard way: we aim to identify some desirable properties that any social ranking of alternatives should satisfy, for a given set of individual preferences.Footnote 34
The following two properties are standard. They incorporate widely shared views of the democratic nature of social choice, namely the principle of unanimity and the idea that no individual shall have his or her own preferences prevail in all possible circumstances:
Weak Pareto: For any two social states x, y, if all agents strictly prefer x to y, then x must be strictly socially preferred to y.
Non-Dictatorship: For every agent i, there exist two social states x, y such that i strictly prefers x to y but society weakly prefers y to x.
Next, we introduce the main principle incorporating a liberal view of autonomy and non-interference:
Non-Interference: Suppose that x and y are two social states such that x is strictly socially preferred to y. Consider two different social states x′ and y′ such that
(1) either agent i strictly prefers x to x′ and y to y′, or she strictly prefers x′ to x and y′ to y;
(2) everybody else is indifferent between each state x, y and the corresponding primed state, x′, y′;
(3) agent i strictly prefers x′ to y′.
Then society should not reverse the strict preference between x and y to a strict preference for y′ over x′.
Our result is that, in a sufficiently diverse environment, the three conditions defined so far are incompatible:
Theorem 1: In a rich environment, there is no social preference ordering that satisfies Weak Pareto, Non-Dictatorship and Non-Interference.
The basic logic of the demonstration of the impossibility can be illustrated by the following two-person example.Footnote 35
The Neighbouring (Almost) Desert Islands
Andrea is young, vegetarian, and loves hiking and meeting new people (but she does not wish to live with anyone). Bob is middle-aged and does not enjoy company, but has a passion for beer and fishing. They live on two different islands, and each of them is the only inhabitant of the island. The two islands are sufficiently far apart that, on the whole, they can live their lives independently, but they are sufficiently close that a number of decisions must be taken collectively. In this section, we translate the social preference ordering into the language of choice in order to provide one illustration of the possible interpretations of the impossibility result.
In particular, Andrea and Bob have to take collective decisions concerning the management of the canal between the islands in different social states. Consider first the choice between two social states x and y: each state is a full description of the world, which includes everything that may be relevant for individual and social evaluation, and we assume that the only difference between them is that in x a boat with a large group of young people partying is allowed to sail along the canal, whereas in y it is not. We represent preferences over the two states using ordinal utility functions and fix specific values. This is only for clarity; nothing in our analysis depends either on the possibility of representing individual preferences with utility functions or on the specific numbers chosen.
$$\eqalignno{ & The\,boat\,trip \cr & \matrix{ {} & {{\rm Andrea}} & {{\rm Bob}} \cr x & {25} & {12} \cr y & {10} & {23} \cr } $$
Next, Andrea and Bob have to choose between two social states p and q, the only difference between them being that in q Bob is allowed to fish in the canal, whereas in p he is not. The following utility profiles represent the individual preferences over these states.
$$\eqalignno{ & Fishing\,in\,the\,canal \cr & \matrix{ {} & {{\rm Andrea}} & {{\rm Bob}} \cr p & {15} & {18} \cr q & 7 & {27} \cr } $$
When it comes to choosing between x and y, and between p and q, their preferences conflict: Andrea would like the boat to sail in the canal, and to ban fishing, while Bob ranks the options in exactly the opposite way.
They reject dictatorial social arrangements and decide to adjudicate these conflicts so that each of them has their preferred choice respected once. Without loss of generality, suppose that Andrea’s preferences prevail in the choice between x and y, while Bob’s preferences are relevant in the choice between p and q. Hence, when considering these two pairs of social states, they decide to allow Bob to fish in the canal and the boat to sail by.Footnote 36
In other words, x is (strictly) socially preferred to y and q is (strictly) socially preferred to p. Footnote 37
Now consider a change from x and y to two different states x′ and y′ which affects only Andrea: perhaps she has broken her leg while hiking in a well-known dangerous spot, or she has been forced to hunt due to a disease that has destroyed all edible plants on her island. Thus she prefers each of the original states x and y to the corresponding states x′ and y′. Bob, instead, does not care at all whether the state is x′ or x, and whether the state is y′ or y:
$$\matrix{ {} & {{\rm Andrea}} & {{\rm Bob}} \cr {x'} & {14} & {12} \cr {y'} & 9 & {23} \cr } $$
Similarly, consider a change from p and q to two different states p′ and q′ which affects only Bob: perhaps a hurricane has destroyed his dwelling, or he has fallen ill after trying to distil an alcoholic drink from a plant growing on his island. He prefers each of the original states p and q to the corresponding new states p′ and q′. Andrea, instead, does not care at all whether the state is p′ or p, and whether the state is q′ or q:
$$\matrix{ {} & {{\rm Andrea}} & {{\rm Bob}} \cr {p'} & {15} & {16} \cr {q'} & 7 & {21} \cr } $$
What should the social preference between the alternatives be? Can Andrea and Bob design a liberal democratic social decision rule that incorporates both unanimity and a non-interfering view?
Because they had decided to allow Bob to fish in the canal in the choice between p and q, by Non-Interference they should not switch to forbidding it. To do this after Bob has fallen ill, or has seen his dwelling destroyed would be an unjustified punishment for him, given that Andrea is unaffected. Therefore q′ should be (weakly) socially preferred to p′.
Similarly, by Non-Interference, forbidding the boat to sail by after Andrea has broken her leg, or has had to hunt against her vegetarian convictions would be an unjustified punishment for her, given that Bob is unaffected. Therefore x′ should be (weakly) socially preferred to y′.
However, Andrea and Bob would unanimously strictly prefer p′ to x′, and y′ to q′, and this brings them to a deadlock. For p′ is strictly preferred to x′ (by Weak Pareto), which is weakly preferred to y′ (by Non-Interference), which is in turn strictly preferred to q′ (by Weak Pareto), which is – finally – weakly preferred to p′ (by Non-Interference), a contradiction.Footnote 38
THE CONFLICT OF PRINCIPLES
Theorem 1 proves that there is an inconsistency between liberal approaches to social decision making, as captured by Non-Interference, and the basic democratic principles of unanimity and non-dictatorship. The next question, then, concerns the implications of this inconsistency for liberal democratic approaches. For ‘The impossibility […] just brings out a conflict of principles […] The really interesting issues relate to the implications of the conflict. There are implications both for evaluation of outcomes and for choice of decision procedures’.Footnote 39
In this section we explore the philosophical implications of this inconsistency. We interpret our arguments and Theorem 1 ‘as guideposts to the creation of a more acceptable theory of liberal democracy’.Footnote 40 Therefore we reconsider the axiomatic framework in order to understand how to construct a coherent approach to social decision making that respects fundamental democratic values and liberal principles of non-interference.Footnote 41
Non-Dictatorship and Unanimity
One possible interpretation of the conflict of principles is that we should reconsider our intuitions about democratic procedures as incorporated in our axioms. We do not think that this is the most convincing, or promising, interpretation.Footnote 42
Non-Dictatorship
Non-Dictatorship rules out the possibility that social choices coincide with the preferences of one individual, regardless of the preferences of all other members of society. It may be argued, however, that not all dictatorships are problematic in our single-profile framework. In the standard Arrovian context, the existence of a dictator implies that there is an agent whose ranking between any two pairs of alternatives prevails for any possible profile of preferences – an obviously undesirable property. In a single-profile world, there are instead situations in which the presence of dictators does not seem to contrast with a democratic ethos. For example, according to our Non-Dictatorship axiom, agents who are indifferent between all alternatives are trivially dictators, as are all agents in society whenever they all have exactly the same preferences and Weak Pareto is satisfied. These forms of ‘innocuous dictatorships’Footnote 43 seem hardly objectionable and may suggest that Non-Dictatorship does not properly capture our intuitions on democratic procedures.
This conclusion is unwarranted, however, and the implications of Theorem 1 maintain all of their normative force. At a general theoretical level, it is true that in a single-profile world there may be – in principle – some preference profiles such that dictators are not of great moral concern. This happens when the dictator has no real influence on social outcomes, for example, because there is no conflict of interest as all agents have identical preferences, or the dictator has no strong views about alternative allocations. Yet Theorem 1 continues to hold in situations in which people do have strong views about social choice and desirable allocations, and their preferences are in conflict, and the existence of a dictator is indeed problematic in these cases. Formally, our analysis holds for any given profile of preferences, and ‘while in some cases an innocuous dictatorship is acceptable, in many other cases it is very much unacceptable’.Footnote 44
As in the standard Arrovian context, dictatorships involve ‘(1) in politics, an extreme sacrifice of participatory decisions, and (2) in welfare economics, a gross inability to be sensitive to the heterogeneous interests of a diverse population’.Footnote 45
Unanimity
Weak Pareto also captures a fundamental property of liberal democracies. As noted even by one of the most prominent critics of the Pareto principle, ‘there is something very central in the idea that preferences unanimously held by all members of a community cannot be rejected by that community’.Footnote 46 And it is unclear that Theorem 1 can be interpreted as raising doubts on Weak Pareto.
To see this point, and to shed further light on Non-Interference, it is instructive to compare our result with Sen’s classic contribution.Footnote 47 In his Minimal Liberty axiom, liberalism is formalized as the requirement that an individual has a sphere of complete control on society’s preferences over minimal subsets of social states. Sen’s Impossibility of the Paretian Liberal can thus be interpreted as identifying a contradiction between the Pareto principle as a democratic rule (unanimity) and a respect for absolute libertarian rights.Footnote 48 We follow Sen by identifying a conflict between democratic rules (the unanimity principle and non-dictatorship) and a liberal right (to Non-Interference), but unlike Sen the conflict is not due to a pre-assigned physical sphere of exclusive competence of the individual. In our approach it is not the nature of the social states that defines individual freedom: it is the fact that the rest of society does not care about certain social states.
Sen’s axiom captures individual autonomy by defining a sphere of positive freedom that overrides concerns about other people’s preferences. This is an important difference from Non-Interference, which embodies a notion of personal autonomy in terms of negative freedom, namely as the protection from interference under certain circumstances not involving others (that is, in the language of preferences, circumstances to which others are indifferent). If Sen’s Minimal Liberty axiom is understood as a ‘social choice formulation of rights’,Footnote 49 then it can be interpreted as assuming the existence of a personal sphere over which individuals have a right to have their preferences respected, regardless of other people’s preferences. In the language of rights, Non-Interference would capture a different (and arguably weaker) type of right – the right to be protected from adverse changes in strict social preferences when others are unaffected in any preference-relevant way.
The differences between Non-Interference and Minimal Liberty have relevant implications concerning the interpretation of Theorem 1. Sen has famously interpreted the ‘impossibility of a Paretian liberal’Footnote 50 as suggesting ‘the unacceptability of the Pareto principle as a universal rule’Footnote 51 because it uncovers a direct conflict between a non-welfarist liberal axiom and the Pareto principle, and because the conflict arises from meddlesome preferences. Neither issue arises in our analysis. On the one hand, the impossibility result does not arise from the clash between a non-welfarist liberal principle and the weak form of welfarism embodied in Weak Pareto because, as argued above, Non-Interference itself incorporates some relevant welfarist (or, more precisely, preference-based) considerations.
On the other hand, the type of preferences that individuals have plays no role in Theorem 1. Formally, unlike in Sen’s seminal contribution,Footnote 52 our result does not depend on the Unrestricted Domain assumption, which requires that all logically conceivable combinations of preferences are possible, and which yields Sen’s impossibility result when some individuals have meddlesome preferences. We adopt a ‘single-profile’ framework: instead of assuming a social decision rule that converts all preference profiles into a social ordering (while respecting some desirable properties that possibly cut across different profiles), we have a fixed set of individual preferences. Our properties apply to that given profile, but our result holds for any given profile: meddlesome preferences, or indeed any preference externalities, play no role. The impossibility result would still stand if we interpreted agents’ preferences as ‘laundered preferences’ to exclude all morally objectionable, meddlesome, non-liberal views. But then, unlike in Sen’s case, Theorem 1 cannot be used to motivate a rejection, or a weakening of the Pareto Principle.
Harmless Wrongdoings
Even accepting that Non-Interference captures some important aspects of the standard interpretation of the Millian doctrine of liberty in terms of harm to others, one may question the consequentialist reading of Mill itself and reject harm-based approaches in favour of a rights-based approach to liberalism. In particular, according to some authors, Mill holds that certain acts are within individuals’ rights because they are of a type with regard to which individuals need liberty in order to develop their individuality, which is a prerequisite to human happiness.Footnote 53 The fact that some acts do or do not harm others is irrelevant to determining whether they belong to the individual protected sphere.Footnote 54
The rights-based approach to the liberal doctrine of liberty raises complex and important issues, both exegetical and philosophical. In this section, we discuss only those aspects that are directly relevant to our analysis.
The key point to note is that, whether or not harmful acts are sufficient as prima facie grounds for intervention is important in a general liberal theory of the state, but is irrelevant, both formally and theoretically, in our analysis. Non-Interference identifies a necessary condition for interference; the conflict with the democratic principles of unanimity and the absence of dictators does not arise from the specification of conditions under which the state should intervene in individual affairs.
The question, then, is whether harm is a necessary condition for individual acts to be prima facie subject to social scrutiny, as in the standard interpretation of Mill.Footnote 55 According to critics, this is not the case, and there are harmless acts that warrant intervention. In our context this is relevant as it can be interpreted as suggesting that Non-Interference be dropped: even if no harm is done, there may be no right to a protected sphere and so the condition in the consequent of the axiom may be violated, with a switch in society’s strict preferences against the individual. In principle, this may provide a liberal way out of the impossibility result.
However, it does not appear that the types of harmless conduct analysed in the literature (including harmless wrongdoings and positive obligations) raise significant doubts about the relevance of harm as a necessary condition for interference in our framework. Footnote 56 Consider the central example of harmless wrongdoing, analysed by Ripstein – harmless trespass:Footnote 57
Suppose that, as you are reading this in your office or in the library, I let myself into your home, using burglary tools that do no damage to your locks, and take a nap in your bed. I make sure everything is clean. I bring hypoallergenic and lint-free pyjamas and a hairnet. I put my own sheets and pillowcase down over yours. I do not weigh very much, so the wear and tear on your mattress is nonexistent. By any ordinary understanding of harm, I do you no harm.Footnote 58
Nonetheless, argues Ripstein, most people would object to the trespass: ‘Your objection is to my deed, my trespass against your home, not to its effects.’Footnote 59 Harm is irrelevant, and the harmless act nonetheless asks for state intervention to protect private property.
The force of the example crucially rests on a notion of harm that is more restrictive than mere moves to less preferred alternatives. Ripstein does not provide a thorough explicit definition of the relevant notion of harm, but he can conclude that the nap is harmless because he excludes the possibility that I can harm you simply ‘by upsetting you when you learn of my deed, or by leading to fears that people will do this sort of thing to others’,Footnote 60 and a fortiori by taking an action that you merely dislike. Negative effects in terms of individual preferences, according to Ripstein, are not sufficient to identify harmful acts. The reason is that if harm is interpreted broadly (to include all acts and choices that have adverse effects in terms of individual preferences), then the harm principle ‘is not a liberal principle. If those harms count, the harm principle underwrites many of the prohibitions it is supposed to exclude’.Footnote 61
In the analysis of necessary and sufficient conditions for interference, a broad notion of harm may well be objectionable. Yet both the nap example and Non-Interference focus specifically on harm as a necessary condition for individual acts to be prima facie subject to social scrutiny. From this perspective, a broad notion of harm that encompasses mere moves to less preferred alternatives is logically and conceptually consistent with a liberal approach: if, even under the broadest possible notion of harm, an act has no harmful effect on others, then this provides strong liberal grounds for avoiding interference.Footnote 62 This intuition underlies our formulation of Non-Interference and, as shown by Theorem 1, it has surprising and substantive implications.Footnote 63
If a broad notion of harm is adopted, it is doubtful that my nap represents a harmless wrongdoing. The point is not so much that the trespass would be harmful because you would rather that I do not enter your house and so it should be outlawed (although it is indeed likely that my entering your house will not leave you indifferent). Rather, contra Ripstein, it may be argued that if it is really the case that my act leaves you completely indifferent then it is not a wrongdoing and the trespass should indeed not be interfered with, consistently with the standard interpretation of the harm principle (and with our formulation of Non-Interference). After all, my act has no effects whatsoever on you, and you are completely indifferent. If you do not care, why should society?Footnote 64
In summary, we come back to the point that Non-Interference does not capture all that is relevant in a liberal, or libertarian, approach. Rather, by focusing on a very specific and restricted set of configurations of alternatives, Non-Interference incorporates some core liberal insights concerning the right of individuals to be protected from unjustified interference.
A POSSIBLE WAY OUT? DAMAGE VS. BENEFIT
Short of giving up non-dictatorship or unanimity, is there a way to construct a consistent liberal democratic approach to social choice? We take Theorem 1 as a demonstration that it is impossible to make social choices in democratic institutions without somehow considering at least some individual actions as relevant even when they affect only one individual. Social judgements cannot treat individuals as separate entities even if these individuals themselves do nothing to interfere with others.Footnote 65 The question then is how to weaken Non-Interference while retaining some key liberal intuitions.
Let us focus on the Individual Benefit Principle. Consider the two-person example analyzed at the beginning of the article. Imagine that alternatives x and y are such that in y I am taxed, whereas in x I am not, and suppose that (for whatever reason) x is strictly socially preferred to y. Suppose next that I suddenly become richer due to brute luck and with no effort on my own part (leading to x″ and y″): on the basis of the Individual Benefit Principle, I could resist any attempt to pay even one penny of taxes on my additional income.
This may seem prima facie unreasonable, and one may object that the Individual Benefit Principle fails to properly capture liberal views. For, one may argue that, while agents can keep the fruits of their labour and efforts, pure windfalls should be shared. There are two reasons why this objection appears unconvincing to us. First, it is worth emphasizing again that I have a rather weak veto power that indeed allows for the possibility that in the end – perhaps after a fair coin toss – society does opt for taxation (y″) because Non-Interference does not rule out the possibility of a switch to indifference after my windfall gain.
Secondly, the view that individuals are not entitled to any gains due to brute luck is common in some liberal egalitarian approaches but it is far from being salient in classical liberal and libertarian views, especially if one notes, once again, that whatever change occurs to me does not affect anybody else. For ‘the libertarian accepts the idea that no one, no state, has a moral right to enforce a redistribution of wealth or property’.Footnote 66
Nevertheless, although we do not think that one can reject the Individual Benefit Principle on general liberal grounds, and no distinction between the Damage and Benefit parts of the harm principle can be found in Mill, there is a sense in which going against an individual’s preference after she has been damaged is different from doing the same thing after she has benefitted. The first course can only be construed as a punishment or as a sadistic exercise, while the second could take the form of a justified ‘correction’ to someone’s gain. While not liberal, this form of interference could be easier to accept for a liberal than the other form, especially from a non-welfarist perspective that allows one to scrutinize the source of the gains. As noted earlier, liberal egalitarianism, for example, admits that pure windfalls to an individual should be shared. In a welfarist, level comparable setting, Mariotti and Veneziani have demonstrated that this leads straight to an unusual conclusion from the liberal point of view, namely welfare egalitarianism.Footnote 67 In the non-welfarist context of this work, no such conclusion is granted or even meaningful. It remains an open question to study what limitations to social choice the Individual Damage Principle implies in this context.
