Tuesday, March 28, 2023

104 - What will be the economic impact of GPT?

In this episode of the podcast, I chat to Anton Korinek about the economic impacts of GPT. Anton is a Professor of Economics at the University of Virginia and the Economics Lead at the Centre for AI Governance. He has researched widely on the topic of automation and labour markets. We talk about whether GPT will substitute for or complement human workers; the disruptive impact of GPT on the economic organisation; the jobs/roles most immediately at risk; the impact of GPT on wage levels; the skills needed to survive in an AI-enhanced economy, and much more.

You can download the episode here or listen below. You can also subscribe the podcast on AppleSpotifyGoogleAmazon or whatever your preferred service might be.

Relevant Links

Sunday, March 26, 2023

Should harmful speech be restricted? More on Mill and Free Speech

Via Flickr

A while ago, I wrote an extended analysis of John Stuart Mill's defence of free speech. As noted in that piece, Mill is not, as he is sometimes perceived to be, a free speech absolutist. He thinks you can restrict some forms of speech, particularly forms of speech that incite violence, amount to fraud or cause harm.

This limitation on free speech is not surprising. Mill's discussion of free speech comes in the midst of his book On Liberty which is all about the Harm Principle and how a necessary condition for any coercive interference with the behaviour of another person is that the behaviour causes harm to others. What's more, Mill's defence of the Harm Principle is, itself, conditional upon his more general utilitarian philosophy: he thinks a blanket application of the Harm Principle is consistent with the greatest happiness of the greatest number and, furthermore, that even when conduct is harmful to others, coercive interference is justified only if it serves the greatest happiness principle.

But Mill's acceptance of this restriction on free speech raises problems. When, exactly, is speech harmful? If someone reports research findings suggesting that members of certain ethnic minority groups are less likely to do well on standardised tests, or are more likely to engage in crime, are those claims harmful? Can they be censored (or 'cancelled' to use a modern term)? In this article, I want to take up some of these questions.

I will start by discussing an argument put forward by Melina Constantine Bell which would allow for the censorship/restriction of bigoted insults on the grounds that they are harmful and fail to express a proposition or opinion that can be countered with other speech. Bell is adamant that her position does not justify the restriction of offensive speech or speech that expresses a genuine opinion. I will raise some doubts about this and use it as a springboard for discussing some puzzles that arise from using harm as a criterion to restrict speech. In particular, I will argue that (a) it is very easy for merely offensive speech to morph into harmful speech and (b), following Mill, whether speech is harmful is itself a contentious claim and depends on a number of factors that may weigh against censorship. 

1. Bell's Basic Argument

Bell argues that bigoted insults are harmful and, since the harm cannot be redressed or countermanded by more speech, they can be rightfully restricted because of that harmfulness. In defending this view, Bell adopts an interpretation of Mill's harm principle that is broadly consistent with the interpretation I outlined in a previous article. Roughly, she thinks that Mill's harm principle serves as an initial and necessary filter on any attempt to coercively restrict the actions of another person. If an action is harmful to others, then you have a prima facie case for coercively restricting it. But you don't have a sufficient case for restricting it. To get a sufficient case for restricting it, you need to establish that the restriction would be consistent with Mill's underlying utilitarian philosophy, i.e. that it serves the greatest happiness of the greatest number. That is not a foregone conclusion. Restrictive policies may cause more harm than they avert. This is a particular concern when it comes to the restriction of speech, as we will see later on.

This means that Bell appears to adopt the following argument in favour of the coercive restriction of bigoted insults (this is not formally stated in her article; this is my reconstruction):

  • (1) If speech is (a) harmful to others and (b) its coercive restriction is consistent with the greatest happiness principle, then it can be coercively restricted.
  • (2) Bigoted insults are harmful to others.
  • (3) The coercive restriction of bigoted insults is consistent with the greatest happiness principle.
  • (4) Therefore, bigoted insults can be coercively restricted.

Some of the terminology here needs to be explained. A bigoted insult is a term or phrase, grounded in prejudicial attitudes towards certain groups of people, that is intended to insult or cause offence. The N-word is, perhaps, the quintessential example, but any racial, ethnic or gendered slur would potentially count (e.g. 'bitch', 'slut' and so on). Bell also includes a long-ish discussion of sexist and racist jokes in her article, suggesting that they too count as bigoted insults.

When is something harmful to others? Here, Bell adopts Joel Feinberg's famous definition of a harm as a 'setback to life interests'. Everyone has life interests, i.e. stakes in certain things such as their physical health, their career prospects, their financial situation. If conduct undermines and sets back those life interests, it is harmful. This means that harmful conduct is reasonably serious in terms of its effects. It is distinguished from merely offensive conduct which causes some temporary mental distress or disgust, but does not set back life interests. That said, both Feinberg and Bell concede that offence could morph into harm, in certain cases, where the offence experienced is 'severe, prolonged and constantly repeated'. We will return to the problems created by this concession later on.

Finally, what is meant by 'coercive restriction'? Bell includes, obviously, the use of legal punishments (imprisonment, fine etc) but also moral disapprobation. Moral disapprobation can include shunning, excluding, condemning or cancelling a person for what they have done. She includes the latter on the grounds that it involves the use of social power to change behaviour in a way that bypasses dialogue and reasoned debate. I am little bit uneasy about including all forms of moral disapprobation within the realm of coercion. For example, I think strong moral condemnation and criticism is fine and not necessarily coercive. But I agree that certain forms of disapprobation, such as shunning or cancelling, have a coercive nature to them. Furthermore, since a lot of the contemporary debate about free speech concerns attempts to deplatform or cancel people, it makes sense to include them within the terms of the argument being put forward.

That's the argument in its outline form. How can it be defended?

2. How Are Bigoted Insults Harmful?

Bell focuses primarily on premise (2) in her defence of the argument. Her main point is clearly stated:

...bigoted speech causes direct harm to individuals, and plays a central role in maintaining and perpetuating institutional oppression and injustice, it causes not merely subjective unpleasant mental states or offence, but also serious harms that usually remain unrecognized both in law and in public discussion. 
(Bell 2020, p 7)


There are two main claims here. The first is that bigoted insults cause direct harm to the people targeted by them. The second, which may arguably be an extension of the first, is that these insults play a key role in reinforcing systems of institutional oppression and injustice, which are also harmful to the people targeted by the insults.

In defence of the first claim, Bell focuses on the case study of sexist humour and its associated insults to women. She reviews the findings from two papers by Thomas Ford and Julia Woodzicka (this one and this one, if you're interested). These papers find that women often fare worse on cognitive tasks as a result of sexist jokes (compared to control groups), and experience a greater sense of self-objectification after hearing sexist jokes, which is linked to mental health issues. In addition to this, studies find that men exposed to such jokes are more willing to cut funding to women's causes. Away from the humour case study, Bell also reviews evidence from the US suggesting that African Americans do worse on health outcomes because of discriminatory stressors, which is likely to include (Bell is admittedly speculative on this) bigoted insults.

In defence of the second claim, Bell draws heavily from Miranda Fricker's concept of epistemic injustice. Such injustice arises when people that are oppressed by social practices cannot meaningfully share knowledge and ideas that might lead to the reform of such practices. Fricker distinguishes between two forms of epistemic injustice:

Testimonial Injustice: This arises when people's testimony is unfairly deemed less credible because of who they are, e.g. police being less likely to trust the testimony of a black person.
Hermeneutical Injustice: This arises when people lack the concepts they need to interpret and explain their experiences (thus giving them less voice in public conversations and debates), e.g. women lacking the concept of 'sexual harassment' to explain what had been happening to them for centuries.


Bell argues that bigoted insults contribute to epistemic injustice and thereby serve to reinforce systemic oppression. While one insult may not make a big difference, the repeated use of them to target particular groups serves to 'intimidate, stigmatize or exclude' them from social practices and debates. They do not get to participate, in equal terms, in the marketplace of ideas.

If we accept that bigoted insults are harmful -- and this seems plausible, along the lines suggested by Bell -- it does not automatically follow that they ought to be restricted. We still have to clear the hurdle of premise (3) and show that their coercive restriction is consistent with the greatest happiness principle. And, before we do that, we have to recognise the obvious Millian objection. Mill's defence of free speech is all about the critical value of truth and truth-seeking practices, and the problem of imperfect censors. As a general rule of thumb, the Millian position is that bad speech, even harmful speech, should be met with more and better speech, not censorship and restriction.

Bell's response to this is that bigoted insults do not 'communicate a viewpoint that serves as a basis for consideration and discussion'. In other words, you cannot meaningfully respond to them with more and better speech. You cannot challenge their accuracy. You cannot introduce evidence and argument to rebut them. So the standard Millian position does not apply to their use. Bell has a nice thought experiment to illustrate her point:

Suppose a group of white college men yell a racist, sexist insult (e.g. N----- whore!) at a female Black student walking across campus at night. What is she to say to counter that speech? "No, actually, I never have been paid for sex, so 'whore' is inaccurate besides being derogatory to sex workers; and 'N-----' is a word with a history connected to intimidation, terrorism and murder of Black people, and which, because of false beliefs about the intelligence and work ethic of Black people, has operated to deny them recognition of their equality and full personhood". Would that prompt an apology? Would it make her attackers recognize the truth of the words spoken...? The likely effect would be to place the female student in danger of physical violence, and/or invite further insults and mockery. This is not an exchange of ideas...and does not bring anyone closer to truth. 
(Bell 2020, 14)


This strikes me as an effective example and shows that, in some contexts, more speech is not the answer. If we generalise the point and accept that it applies to all uses of bigoted insults, then Bell's argument goes through. Some speech can be restricted, even within Mill's framework, and bigoted insults look like a good candidate for this.

But should we generalise the point? I'm sympathetic to some of this, and I think that many uses of bigoted insults are illegitimate and can be rightly restricted. That said I have some concerns about this argument. I'll discuss two major ones in the next two sections but, before that, I want to briefly mention four other concerns.

First, I worry that a prohibition on bigoted insults would require a careful assessment of the intentions and context in which insulting terms are used. It is, obviously, widely documented that oppressed communities sometimes use bigoted insults in a transgressive way and, presumably, this is permissible. While it will often be obvious when this is happening, sometimes it might not be and there may be borderline cases. In addition to this, sometimes someone may use a term that is insulting, and is taken by others to be insulting, even without directly intending this to be the case. Debates about the use of the 'N-word' strike me as a good illustration of this. The dominant opinion nowadays seems to be that you cannot even mention the word in the context of a debate about how it functions or its history. You can only refer to indirectly through the euphemism 'N-word'. But the arrival at this dominant view has taken time (and varies across cultures). Is it obvious, for instance, that a linguist should be coercively restricted from mentioning the 'N-word' in a public speech about the history of bigoted insults? You might say 'no' but it's not clear to me what Bell's argument would say about this example.

Second, and related to the point just made, whether a word or phrase counts as a bigoted insult, is subject to change over time and, consequently, may be a legitimate object of debate in at least some contexts. For instance, the word 'colored' is now taken be insulting in the US (due to its association with Jim Crow laws) and in the UK (due to its historical use in colonialism). But this wasn't always the case. It was once an acceptable and 'positive' term (hence the NAACP). Also, in some countries, for example South Africa, it is an acceptable term, used to denote those with mixed-race heritage. So someone could use the term without intended insult and it may be worth debating and engaging with them about its history to explain why it may be deemed insulting in some contexts. Coercively restricting the use of the term may actually backfire in those cases if such coercive restriction is deemed overbearing and is poorly explained.

Third, while some phrases are undoubtedly insulting and convey no meaningful propositions or viewpoints, Bell's use of sexist and racist humor as an illustration of the problem with bigoted insults, strikes me as more problematic. It seems to me that jokes can convey worldviews and propositions, albeit often indirectly. Sometimes it is worth engaging with and debating those perspectives. In addition, whether a joke counts as racist or sexist will vary depending on the context in which it is made, the intentions of its maker and, possibly, their identity. For instance, someone may make a joke that is, on its face, sexist, but is used, ironically, to poke fun at sexist attitudes and prejudices. Should we restrict such uses of sexist humour? I don't think so and I think there would be a lot of practical problems in implementing such restrictions.

Fourth, the plausibility of Bell's argument will depend, a lot, on the kinds of coercive restrictions she has in mind. What should happen to the college students in her thought experiment, for example. Should they be reprimanded and shunned? Subjected to bias-sensitivity training? Fined? Expelled? Imprisoned? Some punishments might be consistent with the greatest happiness principle but others would not be. Striking the right balance will be tricky.

Those, as I say, are some concerns with the argument. There are two others I want to explore in more detail.

3. Offensive Speech vs Harmful Speech

The first is the distinction between offensive speech and harmful speech. Bell is adamant that her argument does not apply to merely offensive speech. But what might be an example of offensive speech that is not harmful? I'm not sure, but some examples discussed in Bell's article and prominent in contemporary debates might include claims about race IQ differences, and gender differences in aptitude or ability.

I take it as a given that such claims are offensive to members of the groups targeted by them. If I were told that I was naturally or innately inferior at mathematics, I would be insulted. But would I be harmed?

The distinction between harm and offence is cashed out by Feinberg (simplifying a bit) as the distinction between an objectively determinable setback to life interests (harm) vs a temporary episode of mental distress or inconvenience (offence). Perhaps the argument might be that the claim about inferiority in mathematics causes merely temporary distress. It doesn't actually stop me from pursuing my life interests.

Unless, of course, we follow the argument made above, and acknowledge that the repeated utterance of such claims could create an epistemic social environment in which I and others come to believe in my innate inferiority and so censor or alter my behaviour in a way that does knock back my life interests. So the claim can be both offensive and, eventually and more insidiously, harmful.

Related to this, it doesn't seem right to say that offensive speech is always of merely temporary distress. Take another example. Suppose I am a religious conservative and I am repeatedly exposed to statements such as 'gay sex is good sex' or 'those who believe in God are evil'. I find these statements offensive, given my religious convictions. Maybe this initially causes merely temporary distress. I can get on with my life after hearing them. But it doesn't take too much to imagine a world in which I become fixated on these statements. I constantly hear them being echoed and repeated by my social peers. My religious leaders tell me that the people making these claims are evil and a threat to society. I cannot sleep at night knowing that these people are out there, believing and espousing such things. It makes me shake with rage every time I hear them speak. I sweat. I get anxious. I find it hard to breathe. What started out as mere temporary distress has become life debilitating harm.

Maybe you think this slide from offence to harm is unrealistic. But I'm not so sure that it is. I think, given the right combination of personal characteristics and social conditions, a claim that we might otherwise dismiss as offensive could become harmful in the Feinbergian sense of causing a setback to life interests. What's more, some people might be able to manipulate the social context so that speech that is initially dismissed as merely offensive becomes classified as harmful. Some people, critical of contemporary 'woke' culture, argue that this is exactly what is happening in relation to some kinds of speech. There is an incentive to catastrophise and overinterpret certain speech types so that they get seen as harmful and can be restricted.

But to attribute this tendency to purveyors of 'woke' culture alone is a mistake. If it is possible to transform offensive speech into harmful speech, then anyone looking to restrict certain forms of speech on the grounds of harm -- be they religious conservatives or woke liberals -- has an incentive to do so. It is, consequently, not clear if the line between offensive speech and harmful speech can be maintained.

4. If speech is true, but harmful, what then?

Of course, just because speech is harmful, it does not follow that it should be restricted. Bear in mind premise (1)(b) of the argument stated above. The restriction must also comply with the greatest happiness principle. One reason why Bell favoured the restriction of bigoted insults was because they did not state a point of view or proposition. They could not be countered by more speech. Offensive propositions -- such as the examples given above -- are different. Even if we accept that they are harmful, they still state points of view that can be debated and discussed:

Suppose in a college class, or in a debate in the public square, a white man expresses the view that men are naturally better at mathematics, or that they make better political leaders...Expression of these opinions should be protected, and others should have the opportunity to refute them. Otherwise, both points of view appear equally valid, with two sides in disagreement but with no fact of the matter to be established or recognized. 
(Bell 2020, 17)


This is the standard Millian position. Speech should, ordinarily, be countered with more speech. But, the interesting thing about Bell's defence of this view (and others that I have seen) is that they often focus on cases in which the offensive views are presumed to be false. Countering them with more speech is hoped to expose the falsehood. But what if they actually turned out to be true and, as a result of this truth, continued to be harmful to the people affected by them? 

To be clear, I am not granting that the views given in the quote from Bell are true. My point is simply that, suppose after reasoned debate and argument, most people come to believe that they are true. The fact that they are true doesn't mean they cannot be harmful. It could be true that I have terminal cancer with no prospect of a cure but accepting this truth could harm me. As I have discussed before, there is some evidence to suggest that people with positive illusions about themselves do better when it comes to disease prognosis than those with more realistic/true beliefs.

So what then? What happens when we have a true statement/opinion that is also, clearly, harmful? In that case, we seem to have a tension between the defence of free speech (which is a instrumental one focusing on the value of truth) and the harm principle.

It's worth returning to the original source on this matter. Mill, obviously, argued that pursuit of the truth was consistent with the greatest happiness principle. The only way to arrive at true beliefs was through vigorous and uncensored debate about the merits of those beliefs. Against those who claimed that false beliefs could sometimes be valuable or useful, he argued that there was a circular logic at play. Whether a belief is valuable or useful is, itself, a truth claim that can only be established through the free exchange of ideas:

This mode of thinking makes the justification of restraints on discussion not a question of the truth of doctrines, but of their usefulness; and flatters itself by that means to escape the responsibility of claiming to be an infallible judge of opinions. But those who thus satisfy themselves, do not perceive that the assumption of infallibility is merely shifted from one point to another. The usefulness of an opinion is itself matter of opinion: as disputable, as open to discussion, and requiring discussion as much as the opinion itself. 
(Mill, On Liberty, Ch. 2, p 24)


In other words, there is no real tension between truth and the alleged harmfulness of an opinion because you can only assess harmfulness through the pursuit of truth.

I am not sure how satisfying that is. Isn't it also the case that the harmfulness of a bigoted insult is an opinion whose truth can only be established through debate and the exchange of ideas? Does it follow that bigoted insults, therefore, should not be restricted since their harmfulness is up for debate? Maybe. But if that isn't what should happen in the case of bigoted insults then why should it happen in the case of true-but-harmful opinions?

Of course, there is another way of looking at it. We could accept that an opinion is true, and harmful, and still wonder whether the best way to mitigate its harm is to restrict it. Whether an opinion is harmful depends on its uptake: how those affected or influenced by the opinion respond to it and interpret it in their lives. Consider the following claim (deliberately chosen on the grounds that it is less likely to invoke certain connotations than similar claims):

Claim: Men, on average, do much worse than women on standardised tests. What's more, this outcome does not seem to be attributable to educational advantage or social difference. It appears to be innate.


Let's assume this is true. I am a man. Does this claim harm me? Well, it depends on how I and others interpret it. It concerns men, as a group, not me as an individual. So I could argue that it has little or no bearing on me. All that matters is how I do on the tests, not how men as a group do. Of course, other people might be biased against me, as a member of the male group, and their bias may harm me no matter how positive and upbeat I am, but if they interpret it along the lines I do, the harm to me could be minimised. The harm could be further minimised if they were penalised for interpreting the claim in a way that discriminates against me as an individual.

This might, in most cases, be the appropriate Millian response to the truth-harm dilemma. If a claim is true, but harmful, we still shouldn't censor it, but we should consider ways in which its harmful uptake can be minimised. The free exchange of ideas can continue, and harm can be mitigated.

But what if it is more likely than not, given social circumstances and history, that a true claim will be taken up in a harmful manner? What if it would be really hard to suppress or mitigate harmful uptake? That's the really challenging case for the Millian and the challenge that would arise in any society with historic, institutional bias against groups targeted by harmful-but-true opinions.

Thursday, March 23, 2023

103 - GPT: How worried should we be?

In this episode of the podcast, I chat to Olle Häggström. Olle is a professor of mathematical statistics at Chalmers University of Technology in Sweden. We talk about GPT and LLMs more generally. What are they? Are they intelligent? What risks do they pose or presage? Are we proceeding with the development of this technology in a reckless way? We try to answer all these questions, and more.

You can download the episode here or listen below. You can also subscribe the podcast on AppleSpotifyGoogleAmazon or whatever your preferred service might be.

Thursday, March 16, 2023

Mill's Harm Principle: What is Harm and Does it Matter?

One of the most commonly discussed principles of moral and political philosophy -- if not the most commonly discussed -- is Mill's Harm Principle. Introduced in Chapter 1 of his famous polemical essay, On Liberty, the Harm Principle is set out in the following way (all quotes are from a Wordsworth classics edition - page numbers may vary across editions):

The object of this Essay is to assert one very simple principle...That principle is, that the sole end for which mankind are warranted, individually or collectively, in interfering with the liberty of action of any of their number is self-protection. That the only purpose for which power can be rightfully exercised over any member of a civilised community, against his will, is to prevent harm to others. His own good, either physical or moral, is not a sufficient warrant. 
(On Liberty, p 13)


Thus stated, the Harm Principle seems like a robust, and in many ways attractive, statement of classical liberal or libertarian political philosophy. "Don't tread on me!", it seems to be saying. Governments cannot be too zealous or overreaching in moralistic and paternalistic interventions into individual behaviour. They must take a step back and allow us all space to experiment and grow as individuals.

Those familiar with Mill's essay, and the subsequent philosophical and legal debates it has invoked, will, however, know that the Harm Principle is anything but 'one very simple principle'. For starters, it is not clear that it is really 'one' principle. Mill asserts multiple versions of it in On Liberty and qualifies it in numerous ways. Furthermore, it is not obvious that it is 'simple' in its application. What exactly counts as 'harm'? Is 'harm' a necessary or sufficient condition for government interference?

It's not possible to do justice to all the nuances of the philosophical debate about the Harm Principle in what follows. Instead, I will take a more modest approach. I will try to highlight some complexities in Mill's formulation of the principle. I will then discuss some of the particular problems that arise in relation to the concept of harm and consider whether the inability to formulate a fully satisfying theory of harm undermines the credibility of the principle. In short, and following an argument put forward by Anna Folland, I will suggest that the Harm Principle is credible, despite the vagueness of the concept of 'harm'.

1. Mill's Formulation of the 'Very Simple Principle'

Several important questions arise from Mill's formulation of the Harm Principle. The first, and in some ways most crucial, is whether it states a necessary and/or sufficient condition for interference with individual liberty. The quoted passage above provides strong support for the idea that it states a necessary condition for interference. He talks about the 'sole end' and the 'only purpose' for which interference is warranted.

But does the principle also state a sufficient condition for interference? The initial formulation isn't clear on this point, however, Mill's subsequent discussions suggest that it does not. He analyses when, exactly, we are morally justified in interfering with harmful conduct. He thinks there are some kinds of harmful conduct (e.g. lying) that do not warrant interference. In those analyses he leans into his more general utilitarian philosophy, asking whether the benefits of interference outweigh potential costs. So, in other, words, the Harm Principle, for Mill, works as an initial filter for interventionist policies. First we ask whether we are intervening in conduct that is harmful to others. If the answer is 'yes' (and only if it is 'yes'), we ask additional questions about the costs and benefits of the proposed intervention.

This is a sensible view. If all conduct that was harmful to others justified external interference, the Harm Principle would, arguably, be overinclusive and would be anathema to many liberals and libertarians. We will explore this is more detail below when we look at issues that arise with the concept of 'harm'. But, sensible as it may be, this interpretation of the Harm Principle raises problems when considered in light of Mill's broader philosophy. Mill is not a rights theorist. He does not believe in absolute individual rights. As he himself puts it:

I forego any advantage which could be derived to my argument from the idea of abstract right, as a thing independent of utility. I regard utility as the ultimate appeal on all ethical questions; but it must be utility in the largest sense, grounded in the permanent interests of man...Those interests, I contend, authorise the subject of individual spontaneity to external control, only in respect of those actions of each, which concern the interest of other people. 
(On Liberty, p 14)


What is he saying here? In essence, he is saying that the Harm Principle is justified by the principle of utility. A broad, non-interventionist stance, serves the principle of utility -- the greatest happiness of the greatest number -- in the 'largest sense'. It could be that, in particular cases, utility is served by intervening in individual behaviour even when that behaviour is not harmful to others, but it is better to adopt a general rule against intervention since, over the long haul, this is more likely to undermine the greatest happiness of the greatest number. As far as I am aware, Mill does not provide a more detailed utilitarian argument in defence of this blanket policy of non-intervention. It sounds plausible to me, and I would like to believe it, but it might be wrong.

A second question that arises is: 'To whom does the Harm Principle apply?' The obvious answer is 'the government'. The government is not justified in introducing policies that interfere with individual behaviour, unless that behaviour causes harm to others. And, certainly, most contemporary policy-related debates about the Harm Principle, focus on government intervention. But Mill clearly intends the principle to apply more generally. One of the key ideas in On Liberty is that there can be a 'tyranny of the majority' and that minority views and minority lifestyles need to be protected against interference by busybody majorities. So, for Mill, interference by our social peers and social majorities is just as much a problem as government interference. Indeed, so much so, that one of the duties of an effective government is to provide for the liberty of minorities.

A third question that arises is: what counts as a liberty-undermining interference? Mill gives some guidance on this. Right after the most-quoted passage stating the 'very simple principle', he says:

[Man] cannot be rightfully compelled to do or forbear because it will be better for him to do so, because it will make him happier or because, in the opinions of others, to do so would be wise or even right. These are good reasons for remonstrating with him, or reasoning with him or persuading him, or entreating him, but not for compelling him, or visiting him with any evil in case he do otherwise. 
(On Liberty, p 13)


From this, it seems that liberty-undermining interference is limited to coercive interference. This would include threats of punishment or imprisonment for failing to do something, but also, perhaps, other threats of harm. Remonstrating with people or trying to persuade them through reasoned debate is fine, even when this concerns conduct that does not harm people. This might be interference but it is not liberty-undermining interference. This concession is interesting insofar as it permits some busy bodying interference by majorities in the form of public education campaigns. So, for example, a government education campaign intended to discourage people from smoking or consuming alcohol would, on Mill's reasoning, be acceptable so long as it doesn't amount to coercion.

A more complex edge case of interference would be that of 'nudging'. This an idea that comes from the work of Cass Sunstein and Richard Thaler. In essence, it involves the use of techniques for pushing (nudging) people to make decisions that serve their own welfare (or that of the general population) that bypass rational cognitive faculties. For example, placing healthy snacks in someone's eyeline nudges them to favour health ysnacks over unhealthy ones. This doesn't involve reasoned dialogue but it doesn't involve coercive interference either. Where this stands within Mill's framework is open to dispute and there is, of course, a very extensive and detailed debate about it in the philosophical and legal literature.

A fourth question that arises is: does the principle apply to all people, irrespective of age or status? Mill is clear on this point:

It is perhaps, hardly necessary to say that this doctrine is meant to apply only to human beings in the maturity of their faculties. We are not speaking of children, or of young persons below the age which the law may fix as that of manhood or womanhood. 
(On Liberty, p 13)


This is a reasonable, mainstream position. Of course, the age fixed by the law is, always, going to be somewhat arbitrary. If the capacity to make decisions for oneself is largely a function of cognitive ability and emotional intelligence (etc) then there will, undoubtedly, be people officially designated 'children' by the law that ought to be treated as equivalent to adults, and vice versa. We may be able to overcome this problem by applying functional capacity tests -- i.e. assess whether people have the capacity to make decisions for themselves in certain domains -- but these can be difficult to employ in practice. A bright-line cutoff between youth and maturity, for all its arbitrariness, is often the simplest approach.

What is probably less well-documented is the fact that Mill thought the maturity restriction could apply to entire civilisations as well as individuals:

For the same reason, we may leave out of consideration those backward states of society in which the race itself may be considered in its nonage...Despotism is a legitimate mode of governance in dealing with barbarians, provided the end be their improvement, and the means be justified by actually effecting that end...there is nothing for them but implicit obedience to an Akbar or a Charlemagne, if they are so fortunate as to find one. 
(On Liberty, p 13-14)


This will, no doubt, reek of racism and cultural prejudice to the modern reader. Mill is suggesting that entire peoples can be so backward and immature that they need the helping hand of a benevolent dictator. That said, I think there may be something to the point he is making. Hobbes, after all, makes a similar argument, namely, that liberty is, in some sense, a luxury for societies that have achieved a level of prosperity and security. I agree with that, to a point, and Mill himself alludes to a constant tension between security and liberty in his writings. What I don't agree with is the suggestion that despotism is the route to achieving the level of prosperity and security needed for discussions of liberty to become salient. Indeed, my reading of social history is that openness and freedom, combined with strong and stable government, are often drivers of prosperity, not luxuries to be enjoyed after prosperity is obtained.

Mill places some other, miscellaneous constraints on the Harm Principle. For instance, at one point he suggests that people might be legitimately compelled to do things that benefit other people:

There are many positive acts for the benefit of others, which [a person] may rightfully be compelled to perform; such as, to give evidence in a court of justice; to bear his share in the common defence, or in any other joint work necessary to the interest of society of which he enjoys the protection... 
(On Liberty, p 14)


There is a certain common sense to these examples, and implicit in some of them is the tension between security and liberty to which I just alluded. Nevertheless, I find it hard to reconcile the general claim being made with a strict interpretation of the harm principle. Why can I be legitimately forced to do things to benefit other people if my failing to do so does not harm them? The answer to this might lie in what exactly we mean by 'harm' in the context of the harm principle. It is to this thorny topic that I now turn.

2. What is harm? Comparative Accounts

One thing Mill does not do is provide us with any general theory of harm. This might seem odd given that harm is the central concept in the Harm Principle. But perhaps it is good to avoid overly abstract thinking about harm. There are paradigmatic cases of harm to others, there are paradigmatic cases of harmless conduct, and there are borderline or difficult cases. So, for example, if my hobby is going around smashing people's fingers with a hammer, then clearly my conduct is harmful to others and I can justifiably be stopped. If I like to stick needles in my own fingers, in the privacy of my own home, then clearly my conduct is harmful to no one other than myself and I should not be stopped. If I play loud music, late into the evenings, and this upset my neighbours, then this is more borderline case. We can have an argument about whether it really counts as harm, and whether it can be justifiably stopped.

Many philosophers don't like having fuzzy or contestable concepts at the heart of our ethical theories. They seek theoretical and abstract purity. They want a general theory or account of harm that tells us whether or not specific conduct falls foul of the Harm Principle. You can understand why. A general, abstract theory of harm could be used to assess novel or controversial cases of harm and give us a clear answer as to whether it can be justifiably interfered with or not. No surprise then that many philosophers have attempted to provide a general theory of harm that we can plug into Mill's principle.

Some, however, have argued that no general and satisfying theory of harm exists. Anna Folland, in her article 'Mill's Harm Principle and the Nature of Harm', examines these critiques in some detail. The critics all follow the same strategy. They introduce some general theory of harm -- oftentimes one that has intuitive appeal or support from other debates -- and argue that it is under or over inclusive in some crucial respect. In other words, they argue that the theory would identify conduct as harmful that most people agree should not count as harmful or, vice versa, would fail to identify cases as harmful that should count as harmful.

Let's consider three examples. The first is the Temporal Comparative account of harm (TCA) (definitions taken from Folland's article):

TCA: An event e harms a subject s if, and only if, e makes s worse off (in terms of well being) after e than s was prior to e.


Assume I was healthy and well-functioning yesterday. Today, a man in the street punched me in the face and gave me a black eye. Clearly, I am worse off now, after the punch in the face, than I was yesterday, before the punch. The punch has harmed me.

Sounds sensible, right? Not so fast. Critics argue that the TCA doesn't deal appropriately with some cases. Suppose that your child has contracted a virus and has a terrible sore throat. You could, if you so decided, eliminate their pain by giving them some medicine, which you have freely to hand. On the TCA, you do not harm them by failing to give them the medicine. But, arguably, the failure to intervene in this case is a kind of harm. Similarly, imagine a case in which a child suffers from some debilitating illness because her mother smoked throughout her pregnancy and both parents have smoked, in her presence, from the moment she was born. Their actions clearly harm her, but they don't make her worse off than she previously was since the harmful actions coincided with her conception and birth.

Counterexamples like this lead people to reject the TCA and favour alternative theories. One such alternative is the 'baseline from Mankind' comparative account (MCA):

MCA: An event e harms a subject s if, and only if, e makes s worse off (in terms of well-being) than the normal well-being level of mankind.


This avoids the counterexamples to TCA, but faces some counterexamples of its own. For instance, if someone is extremely wealthy, relative to the average person, then stealing their money so as to bring them down to the average level of wealth would not count as harm, on the MCA. This is counterintuitive. Similarly, if someone is extremely attractive (admittedly more a subjective quality), then disfiguring them so as to make them more averagely attractive would not count as harm. This is also counterintuitive. This makes the MCA a non-runner.

This leads to one final theory of harm, the Counterfactual Comparative theory (CCA):

CCA: An event e harms a subject s if, and only if, s would have been better off (in terms of welfare over her lifetime) in the absence of e.


This is a very popular theory of harm. It pops up in numerous applied ethical debates. For example, I have seen people propose it as the theory that explains why death is harmful. It also avoids the pitfalls associated with the two previous theories. But it faces problems of its own.

The main one is that if harm is defined by comparison to what would have happened in some counterfactual possible world, then we have to pick an appropriate reference class of possible worlds. But, depending on how we select the reference class, conduct can be deemed harmful or non-harmful in somewhat arbitrary ways. For instance, if I, as a parent, fail to provide private music lessons to my child, am I harming them? Intuitively, most people would probably say no. But perhaps there is counterfactual possible world in which they do much better, in terms of welfare, if they receive the private music lessons. So my failure to pay for them is a harm if I compare this world with that world. In short, weird things start to happen when you start comparing this world with counterfactual ones. If you choose counterfactual worlds in which people do much better than they do in the actual world, then they harmed by virtually everything that is happening to them in the actual world. Conversely, if you choose counterfactual worlds in which people do much worse than they do in the actual world, they are not harmed.

There are some other theories of harm, but these three are a representative sample of the ones debated by philosophers and each fails to provide a fully satisfying underpinning for the Harm Principle. What do we conclude from this? Critics conclude that the Harm Principle must fail. Folland, in her analysis, is not so quick to draw this inference. As she points out, critics are adopting something like this argument:

  • (1) In order for the Harm Principle to be acceptable, it must be grounded in some fully satisfying theory of harm.
  • (2) The only plausible candidate theories of harm are the TCA, MCA or CCA (...etc)
  • (3) Neither the TCA, MCA nor CCA (etc) provides a fully satisfying theory of harm.
  • (4) Therefore, the Harm Principle is not acceptable.

But there are a number of ways to reject this argument. One way would be to reject premise (2) and argue that there are other plausible candidate theories of harm or that, perhaps, some mishmash of theories could be fully satisfying (it's not either/or). Another way would be to reject premise (1) and argue that the fate of the Harm Principle does not rest on articulating a fully satisfying theory of harm. Folland suggests this might be right response to critics. She does so on the grounds that harm features in many ethical debates and principles and yet rarely do we require those that deploy the concept to articulate a fully satisfying theory of it. It is prima facie plausible that harm is a meaningful ethical concept that delineates between acceptable and non-acceptable conduct. That we have some problems providing a fully satisfying theory of it does not undermine its use in the Harm Principle.

I think this is correct. I have long been uneasy with the argumentative standards employed by moral (and other) philosophers, which seems to suggest that we should reject normative principles if they fail to provide intuitively satisfying results across all possible worlds. I am not sure any principle could satisfy such a demand.

3. What is harm? Miscellaneous problems

There are other problems with the concept of harm. One concerns the distinction between harm to self and harm to others. This is central to Mill's principle. He thinks harm to self cannot provide a justifiable basis for coercive interference. He thinks harm to others can. Some argue, however, that this formulation is not quite right. The crucial distinction is not between conduct that is harmful to self versus harmful to others. The crucial distinction is between conduct that is consented to versus unconsented to. If I harm you, but you consent to being harmed, then it would be wrong to coercively interfere with that choice. Although Mill avoids this formulation, it is consistent with his other views, such as his claim that freedom of association is one of the fundamental forms of personal liberty:

from this liberty of each individual, follows the liberty, within the same limits, of combination among individuals. 
(On Liberty, p 15)


A related, and perhaps more serious problem, concerns the causal relationship between self harm and harm to others. Sometimes conduct that is, primarily, harmful to the self is also, indirectly, harmful to others. If I want to drink myself into oblivion, you might argue that this is my right, as a free person. But, of course, my decision to do so might harm my family. It might deprive my children of care and resources they need to survive. So can we coercively interfere with the decision to drink?

Mill has a long discussion of this issue in On Liberty, focusing on the arguments of temperance activists and prohibitionists (looking to ban the sale of alcohol). He rejects an outright ban, arguing that drinking is a private pleasure within the sphere of personal liberty. But, he accepts that if drinking becomes harmful to others, coercive interference may be justified. He also argues that those occupying certain social roles -- e.g. police officers, surgeons -- may be justifiably banned from drinking, at least while on duty.

Another problem concerns the gravity of harm needed to justify coercive interference. Some people are unwilling to accept that any and all harms meet the threshold needed to justify coercive interference. Perhaps the most elaborate discussion of this issue can be found in Joel Feinberg's, multi-volume work on the moral limits of the criminal law, which is, in effect, an extended discussion and application of Mill's principle. It's impossible to capture the nuances of Feinberg's position in a short summary, but the gist of it is that he distinguishes harm from hurts and offensive conduct. A harm, for Feinberg, is something that sets back your life interests. So it is a reasonably serious kind of injury to your ongoing life plans and personal welfare. A hurt is a more trivial and temporary kind of injury, e.g. a graze upon your knee. Offensiveness is mental displeasure or distress caused by the conduct of others. For Feinberg, harms justify coercive interference, but hurts and offensiveness do not. But Feinberg ties himself up in knots about this, eventually conceding that some kinds of offensiveness, if they are sufficiently persistent and serious, might justify coercive interference, partly on the grounds that they end up being harmful if they are persistent and serious.

Others counter argue that Feinberg's attempt to distinguish between different levels of harm is unnecessary. For instance, Turner argues that the Harm Principle ought to rest on an expansive definition of harm. Virtually all harm to others, no matter how trivial, raises a prima facie case for coercive interference. Whether coercive interference is then justified depends on whether it passes the 'greatest happiness' test: i.e. do the benefits of the coercive interference outweigh its costs. Since all coercive interference is itself harmful, this is a difficult threshold to cross in the case of relatively trivial harms. Less intrusive and coercive policies will almost always be preferable.

4. Conclusion

The foregoing is a summary of some of the key challenges facing the Harm Principle. Mill's claim that it is a 'very simple principle' is both true and misleading. It is true insofar as it is easy to state the principle, and it is intuitively appealing. Nevertheless, it is misleading insofar as its practical application raises a number of complexities. Still, the mere fact that it can be contentious and that there can be difficult edge cases is not, in itself, a reason to reject the principle. One can see why it remains popular to this day.