5.1.1 The Forms of Freedom

The concept of freedom is often discussed in terms of one’s “negative” freedom; that is, in terms of noninterference, nonaggression, noncoercion, or in general, the absence of external constraints on one’s ability to act.Footnote ³ A person enjoys negative freedom when they are not being actively interfered with or coerced. Negative freedom is often considered primary, for a couple of reasons. One is that limitations on negative freedom are intuitively obvious; if one is prohibited by law or physically prevented from taking an action, their freedom has been restricted. Another reason is that some shortfalls in negative freedom are especially restrictive, which makes negative freedom appear to be the most important form of freedom. For instance, if one has been kidnapped or forced at gunpoint to hand over one’s possessions, one’s current political freedoms are of secondary concern.

Yet, looking beyond the obvious importance of negative freedom to human life, the value of noninterference does not swamp the value of all other forms of freedom one could value. Freedom must be understood as in two ways extending beyond the mere absence of interference. To see why, consider two cases. First, consider the case of a woman living under an oppressive political regime in which she is permitted to vote only with her husband’s permission. Supposing that her husband does, in fact, allow her to vote, she would have negative freedom. After all, no one has coerced her, interfered with her, or physically prevented her from voting. But her husband could have chosen otherwise and could have interfered. Depending on someone else’s approval to vote is a deeper sort of unfreedom underneath her thin, negative freedom to vote. Second, consider the case of a person who is not dependent on the approval of someone else to vote, but who does require a wheelchair for mobility. If there are stairs that they must navigate to access their polling place, and there are no ramps or alternative means of casting a ballot, they are not free to vote, their ample negative freedom (insofar as no one coerces, prohibits, or physically blocks access to voting) aside. They are unfree because they cannot effectively do what they wish to do, namely vote.Footnote ⁴

One conception of freedom that extends beyond negative freedom, corresponding to the case of the first woman, is called “republican freedom.” Beyond mere noninterference, republican freedom requires non-domination, which is to say, the absence of arbitrary power or domination. On this conception, even the possibility of interference counts as unfreedom. Noninterference that occurs at the whim of a benevolent ruler does not count as genuine freedom. For republican freedom, it is not enough that there is no interference. Rather, as Philip Pettit puts it, interference must be “robustly” absent, meaning that interference is “absent over variations in how far others are hostile or friendly.”Footnote ⁵ For the first woman to be free, her husband’s benevolence cannot play a role in her ability to vote; she must be able to do it without needing his permission.

The republican theory of freedom does not emphasize that the fullest sort of freedom requires having not just choices, but autonomous choices. Pettit, for instance, argues that his account “does not require that resources […] must be present over variations in your personal skills, the natural environment, or the structure of society,” because “[i]n order to choose freely between the options in a certain choice, it is enough that you actually have such assets available.”Footnote ⁶ The republican theory holds people to account for the choices they make, no matter why they made them or what challenges they faced leading up to the choice. Yet, if the affective and deliberative challenges to human agency are real, it must be in part through their influence on our choices. To fully understand our freedom, we must take into consideration not just our choices, but how those choices get made.

Another conception of freedom that is distinct from negative freedom, corresponding to the case of the second voter, is “positive freedom.” It involves not only the absence of external constraints, but the ability to effectively carry out one’s interests. While negative freedom is defined in terms of (and, to some extent, presupposes) the capacity for autonomy,Footnote ⁷ positive freedom requires the ongoing exercise of self-government. Republican freedom, moreover, cannot by itself guarantee positive freedom. This is because the republican account is primarily oriented toward the alleviation of subjugation, or “defenseless susceptibility to interference,”Footnote ⁸ rather than to the positive development of persons’ talents and capacities. But this glosses over the fact that one might be undominated by all others but nonetheless have interests that one cannot pursue because they lack the capacities to effectuate their desires or realize their interests. For the second voter to be free, they must not only be undominated by all others, their polling station must also have the relevant accessibilities, allowing them to actually vote rather than merely have the (empty) right to do so.

“Unfreedom” in the positive sense can be tricky to diagnose or discern, in part because it can become fully internalized. Consider, for instance, a third person who has been raised in an insular religious community. Suppose that in this community, people adhere to a narrow set of religious precepts, their access to information is restricted by their religious leaders, and their social roles are defined in terms of highly gendered categories. By the time the person reaches adulthood, they are likely to have internalized their community’s standards about the nature of personhood and about what counts as an adequate range of options. Unlike the first woman, this person is not externally constrained, in the sense that someone could prevent them from acting according to their desires, and unlike the second person, their choices have not been undermined or frustrated by inaccessibility. Indeed, they may even possess all the material resources necessary to exit the community. Yet there is an important sense in which they lack freedom, owing to how their space of possibility has been circumscribed; their agency itself has been short-circuited in a way that they cannot repair or (possibly) even recognize. To be clear, this is a generic, hypothetical example. Gesturing at this kind of example does not justify any inference about the authenticity or “false consciousness” in any real cases. Rather, it is an example to show that freedom is not reducible to persons’ abilities to act on their preferences, as preferences can be formed in response to oppressive conditions, which are themselves antithetical to freedom.

It’s worth pausing to explain this a bit more. We have distinguished negative freedom (freedom from external constraints) and positive freedom (ability to effectively carry out one’s interests). One might argue that these apparently distinct facets of freedom are reducible to a single conception. Gerald MacCallum, for example, argues that we can reconcile negative and positive freedom under a single, tripartite conception of freedom: between (1) a person, (2) their goals (their “doings” or “becomings”), and (3) the relevant constraints.Footnote ⁹ From this perspective, understanding freedom as primarily being about external constraints or as primarily being about individuals’ effective capacity to act is a mistake. Rather, what matters is a person, some action they wish to take or way they would like to live, and whether there are constraints on the person’s ability to take the action or to live the way they would like. Hence, a person driven by an addiction might be free of external constraints, but unfree insofar as their addiction prevents them from acting or living in a way that comports with their higher-order desires.

But even a conception that collapses positive and negative freedom cannot adequately explain our third case. The problem there is that despite the fact that there are no external constraints and despite the fact that the person is able to act upon their values, it is odd to say that they are free tout court. Being raised in oppressive circumstances may preclude them from developing reasonable sets of values and preferences. The limitations of freedom in this case are, as Christman puts it, to their “quality of agency.” That is, a person who has had unreasonable constraints on their ability to develop their own sense of value has diminished “effectiveness as an agent.”Footnote ¹⁰ More is needed to secure their effectiveness as an agent than the mere absence of external constraints and ability to act upon their preferences. Quality of agency requires that their beliefs and desires be both authentic and competently acquired for an adequate range of options. It also requires social and political structures that support those beliefs and desires and that one has sufficient affordances to act on those beliefs and desires. This quality of agency view can explain how challenges to authenticity, such as in cases of severe addiction, limit freedom. For instance, we might imagine a person who is driven by addiction but who also has enough resources and opportunities to sustain their habits in perpetuity.Footnote ¹¹ Such a person lacks freedom in the fullest sense because their addiction limits their efficacy as an agent.

Our view of freedom aims to capture both freedom as quality of agency (which itself encompasses both negative and positive freedom) and non-domination (which encompasses republican freedom). In this sense it is as ecumenical as our conception of autonomy from Chapter 2. Specifically, we recognize both non-domination and autonomy as vital facets of freedom.Footnote ¹² An agent’s freedom requires what we will call “ecological non-domination.” It is ecological in the sense that one’s freedom encompasses both facts about oneself (quality of one’s agency) and their environment (absence of external constraints and non-domination). One is not free without both.

Notice that it is possible to enjoy some aspects of autonomy without full republican freedom (or even negative freedom). Consider, for instance, the “Russian oligarchs”; the group of wealthy Russian kleptocrats who rapidly acquired enormous wealth in the wake of post-Soviet privatization. They enjoy unfettered positive freedom, in that they have at their disposal the means to purchase world-class football clubs,Footnote ¹³ record-breaking superyachts,Footnote ¹⁴ and former royal estates.Footnote ¹⁵ Yet, at the same time, their enjoyment of noninterference is not all that robust: Their personal safety depends on remaining in good favor with the regime. Mikhail Khodorkovsky’s oil company (Yukos) came to control a sizable fraction of Russia’s oil supply. Yet when Khodorkovsky engaged in a power struggle with the government, he was imprisoned for nearly a decade over trumped-up fraud charges. Similarly, the chairman of the country’s biggest telecom, Vladimir Yevtushenkov, was placed under house arrest by Russian authorities under suspicion of money laundering.Footnote ¹⁶ Indeed, to some extent even the president of Russia, Vladimir Putin,Footnote ¹⁷ faces this sort of precarity: Both his wealth and safety depend on retaining political power, and this in turn requires at least minimal cooperation with the oligarchs. For him, interference is absent, but perhaps not robustly absent. That is, it may not be absent over variations in the friendliness of others.

5.1.2 The Value of Freedom

It is still an open question as to the relation between freedom and morality. Why, in other words, is freedom good or morally significant (if it is)?

As in our discussion of autonomy earlier in this book, we can understand the value of freedom by considering its boundaries. We have already seen that autonomy is not good without qualification. For example, using one’s autonomous capacities to undermine others’ freedom is bad. Consider the oligarchs just mentioned: not only does their freedom permit them to capture a sizable proportion of the wealth and national product of Russia for their personal benefit; they also have a long history of using their freedom as a tool for promoting the domination and diminished quality of agency of others. With this sort of example in mind, John Danaher argues that freedom is an “axiological catalyst” – that it “makes good things better and bad things worse.”Footnote ¹⁸ Yet even this seems too strong when we consider cases that raise the “paradox of choice.”Footnote ¹⁹ This paradox is that although we often believe that having more options is good, in fact having a large number, wide diversity, or substantial nuance in our choices leads to anxiety and stress rather than happiness or satisfaction. A greater range of choices is simply not always better, for either the chooser or those who depend on their choices.

These cases suggest that it will be impossible to coherently explain either freedom or autonomy as intrinsically good. Of course, we need not ignore the fact that freedom and autonomy are almost always good. An easier proposition to defend is cast in terms of unfreedom: shortfalls in freedom (that is, either shortfalls in self-government or instances of domination) are prima facie bad. This commitment is compatible with freedom being outweighed by any number of other values in any number of cases, such as when we take the freedom of oppressors to be outweighed by the value of human rights. There may be cases where the protections licensed by the assumed value of freedom are morally and politically outweighed by the public interest, such as public health mandates. The shift in emphasis from freedom to unfreedom is motivated by the fact that in practice, it is easier to discern what is objectionable (or not) about a given freedom deficit than to discern how freedom (conceived of as an intrinsic good) must be limited.

This account has substantive implications. It holds that freedom deficits are usually morally bad and that unfreedom can be thought to serve morally good purposes only in certain circumstances. Figuring out how to best promote human freedom, then, is not as simple as settling debates about which sorts of interference would be objectionable to rational actors. Unlike freedom, which can be defined in the abstract, unfreedom can only be defined in terms of the specific foibles of human agency, specific architecture of human environments, and specific implications of social choices. As we show in the next section, the effects that shortfalls in freedom can have on our freedom are well illustrated in terms of the challenges to freedom mentioned earlier.

5.2.1 Affective Challenges to Autonomy

The first sort of challenge to freedom is that human behavior is driven by affective influences, such as fear, anger, resentment, or addiction, as opposed to purely being driven by rational attitudes. These affective states, we will argue, can undermine the authenticity of people’s preferences and desires, threatening our freedom by threatening our autonomy.

People are so driven by affective states that it might seem difficult or impossible to clearly pinpoint any kernel of rational attitudes that lie underneath. Humans, it might seem, experience affective influences “all the way down,” in the sense that our authentic attitudes cannot be distinguished from any other attitudes one might have in a principled way.Footnote ²⁰ But in some cases, the influence can be so dramatic that it is hard to accept that the resulting behavior could be freely chosen or even independently motivated. The AAA Foundation for Traffic Safety has found, for instance, that around 4 percent of drivers each year have gotten out of their cars to angrily confront another driver, and 3 percent have purposefully run into another car for the same reason.Footnote ²¹ Anger is not the only source of authenticity-undermining affective influence: It is difficult to believe that a person could have an authentic desire or preference to go to an internet cafe and play a game for fifty straight hours before dying of exhaustion, yet this has happened on several occasions.Footnote ²²

As has been known since B. F. Skinner, people’s susceptibility to operant conditioning can be used by third parties (such as casinos) to reinforce certain patterns of behavior over others. Many digital platforms employ conditioning strategies in the same sort of way, engendering affective states in their users that undermine the authenticity of their choices. The targets of optimization are different, of course; mobile developers optimize in terms of clicks, views, watch times, and so on, rather than lever pulls or revenue benchmarks. However, the underlying strategy – engendering some sort of artificial dependency – is the same.

As Nir Eyal points out, engendering artificial dependencies is a crucial element of the general strategy of keeping users “engaged.” He outlines a four-step method for effectively “Skinner boxing” a digital platform.Footnote ²³ The first step involves a trigger, which draws the user’s attention to the app or platform. The quintessential modern trigger is, of course, the “push notification,” which literally forces you to pay attention to it enough to dismiss it. The second step of Eyal’s method involves getting users into acting on the trigger. The best way of getting users to do this is to get them to anticipate some sort of reward. The third step involves tying user behavior to these rewards but making the rewards variable, in the sense that they are sometimes highly rewarding but at other times mundane, in a way that is unpredictable. Dopamine surges whenever the brain has been conditioned to expect a reward, but without variability, the experience becomes unsatisfyingly predictable. The fourth step, finally, involves getting users to make some sort of investment in or commitment to the app, to maintain their engagement with it in the future.

The Duolingo platform offers an illustration of how Eyal’s model can be employed. The app uses push notifications to draw the user’s attention to the app, sometimes coupled with a guilt trip. When a user has stopped using Duolingo for a period, they will receive a notification depicting the very cute Duolingo mascot in tears, stating that “Language Bird is crying,” that a failure to go back to learning a language will lead the mascot to “eat a poison loaf of bread,” and that the next email to the user will be an e-vite to Language Bird’s funeral.

Skill badges, which represent user achievements, are displayed at the end of rounds in a pleasingly unpredictable way. Finally, the platform has a number of mechanisms for investment: users’ earned skills decay over time, requiring ongoing practice, and the platform itself also contains a microtransaction market for premium avatars, allowing users to invest in the app using real money.

In several ways, then, Duolingo uses its users’ affective states to undermine their autonomy and, thus, limit their freedom. For the most part,Footnote ²⁴ its freedom-impinging tactics do not raise anyone’s hackles, because there is a background presumption in any discussion of it that language acquisition really is valuable for its users (thereby serving their global autonomy if not quite their local autonomy). The same tactics, however, are more troubling in the context of employment, where people can be compelled to act in dangerous ways to protect their basic livelihood. Consider the practices of the ride-hailing company Uber.

Uber has come under substantial criticism on the basis of how its algorithms have negatively affected its drivers.Footnote ²⁵ The company uses these systems for a variety of purposes: to track passengers, to anticipate market demand, and (at one point) even to identify and deceive regulators, media, and law enforcement. As we discuss in Chapter 7,Footnote ²⁶ the company employs several practices to keep its drivers working longer than they might otherwise. Here we will discuss two sorts of practice, arguing that one serves to undermine the freedom of drivers while the other is more morally benign.

Both sorts of practice can be linked to the strategy outlined by Eyal. The first involves Uber’s user interface design choices. The driver app, for instance, is configured by default to remind drivers of their goals through push notifications, and it is also configured by default to queue another rider before the current rider has been delivered. Presumably, the motivation for Uber to employ this particular set of defaults rather than the alternative sets is that their chosen configuration is statistically more likely to induce drivers to accept more riders than the others. (In fact, it is likely that Uber has discerned the precise overall effect of this design choice through massive-scale A/B testing.)

Nevertheless, Uber’s design choices do not necessarily undermine drivers’ freedom in this case. It is plausible that drivers have their own reasons for such settings being one way or the other. It is plausible that someone might want to work a shift determined by a particular monetary goal or period of time and, thus, to simply queue as many customers as fast as possible without having to continue manually making choices at each possible choice point. Uber would certainly not increase the freedom of users by doing away with the choice altogether. So long as Uber does not covertly reset the defaults, or design the interface in a predatory fashion, the company’s UI design choices do not strike us as an issue of serious moral concern (or even, for that matter, an issue of freedom at all).

The second practice for keeping drivers working longer than they would otherwise, however, is less apparent to the drivers and much more troubling from the standpoint of autonomy. It involves such practices as “surge pricing,” which is a cost-multiplier that raises the price of a ride (sometimes dramatically) in a location at a time when the demand for drivers is high or the supply of drivers is low. Some of the additional revenue from this cost markup is passed to the drivers, so the surge-pricing system gives drivers a financial incentive to work more than they would otherwise, because they will be able to make more money than they would have otherwise. The “surge-pricing” mechanism is a familiar device in modern retail – it’s just a form of dynamic pricing – but essential to Uber’s use of it is an element of uncertainty and variability that is not readily apparent to the drivers at first: when drivers accept a ride at a “surge priced” rate, they do not know whether the ultimate payment for the ride will be surge priced.Footnote ²⁷

The fact that these practices are opaque to drivers might lead one to question Uber’s motives. Our concern here, though, is the drivers’ autonomy. A recent report describes the phenomenon of “chasing the surge” at length.Footnote ²⁸ Since 2016, Uber has employed the “boost” system, which grants automatic surge pricing to drivers who have completed a certain number of rides the previous week, in a tiered arrangement from “Bronze” to “Platinum.” One driver described his relationship with Uber in the following way:

The authors of the report note that “[t]he stress of these games, be it chasing a surge or platinum, was a refrain in 50% of our interviews in surveys.”Footnote ³⁰

What is the moral upshot of the Uber case? According to our view of the value of freedom, the verdict is mixed. The nudges Uber employs in its user interface, which for the most part serve obvious functions for its users (and which are also relatively low stakes and transparent), are unproblematic as a class. From the perspective of alleviating the domination of its users or promoting their autonomy, it is not clear that Uber could have done better.

Uber’s Skinner-boxing of its drivers to “chase platinum,” in contrast, is considerably more troubling and gives rise to a freedom-based complaint. The authenticity condition of autonomy precludes users being in affectively compromised states; if people’s actions lack authenticity – in the sense that those actions serve to alienate the person from their basic values – then those actions will not be autonomous. However, the boost aspect of the Uber surge-pricing system exists precisely to place users in such a state and, in so doing, urges them to “chase” overwork even at the expense of their health. The same freedom-impinging features that seem somewhat innocuous in the context of Duolingo and language acquisition become more troubling when a dangerous degree of overworking is a potential effect.

5.2.2 Deliberative Challenges to Agency

Exploiting our emotional vulnerabilities is not the only way to hack human agency. Even if we set aside the fact that human agency can be undermined by noncognitive affective states, human cognition itself is limited in ways that can be exploited to limit our quality of agency.

Our actual deliberative process is both bounded and inaccurate; the information we receive as reasoners is limited by our computational capacities and fitted to our choice environments, and our processing of the information we do get is subject to a variety of errors and biases.Footnote ³¹ Many of these biases – the availability bias, anchoring bias, conjunction fallacy, base-rate neglect, and so on – are well known.Footnote ³² To this end, in the next chapter we discuss problematic search suggestions and algorithmically curated news feeds, which allows people to expose themselves only to information that reinforces their existing beliefs. Here, we focus more on the boundedness of human agency.

Algorithmic systems, particularly digital platforms, introduce new sources of economic value – namely personal and consumer datasets at big data scaleFootnote ³³ – as well as new sources of risk – data breaches and other forms of exposure.Footnote ³⁴ These systems can be so complex both for developers and for end users that it is impossible to live up to the Reasonable Endorsement Test in practice. To be sure, the choice environment of our digital daily lives does not live up to this moral standard: most people do not and could not always read all of the fine print to which they claim to consent and the idea of “common terms” has been stretched thin by massive swathes of boilerplate that are unreadable by humans in practice.Footnote ³⁵ In practice, we face an avalanche of digital “pseudo-contracts” – contracts that are so convoluted that they assume impossible levels of human competence and thus fail to represent any actual “meeting of the minds” between the parties.Footnote ³⁶

Our constant subjection to these “contracts” violates our procedural independence, by exploiting the limits of our epistemic competence and thereby undermining our consent. One study, for instance, found that “only one or two of every 1,000 retail software shoppers” actually access the contract “and that most of those who do access it read no more than a small portion.”Footnote ³⁷ In 2014, Europol conducted an experiment in which they offered access to a Wi-Fi hotspot behind a contract that granted access only if the recipient signed a so-called Herod clause, in this case, to agree “to assign their first born child to us for the duration of eternity.”Footnote ³⁸ Six people still signed the contract. In 2015, a Guardian journalist resolved to read all the terms of conditions for all his services and wound up reading 146,000 words in a single week.Footnote ³⁹

Most modern web services now have end-user licensing agreements (EULAs), but Brainly, a peer-to-peer learning platform in which users can provide homework help to one another, offers an especially striking example of a pseudo-contract. The platform’s end-user licensing agreement is expansive. First, the company is permitted to force users into arbitration. Second, it is permitted – despite its claim to protect users with a privacy policy – to license user content to third parties, distribute it through any media, or even sell personal data outright as part of bankruptcy, even after its users have terminated their accounts. Third, it is permitted to change the terms of service at any time without notice.

Each of the components of the Brainly pseudo-contract can be seen to exploit the undermined capacity of its users to reflect on their values, motivations, and decision-making in the digital environment. The company’s ability to automatically force users into arbitration allows it to settle disputes with its users on legal terrain that is broadly unfavorable to them; its privacy policy serves as a smokescreen that obscures the extent to which it can freely trade or sell user data; its right to change its terms of service at any time prevents and discourages users from reading any one version of it and contributes to the overwhelming deluge of information users must cope with.

This pseudo-contract exploits its users’ limited quality of agency. By holding them responsible for self-government in a context where this is humanly impossible, the contract exploits their lack of understanding of their circumstances for corporate benefit. In many cases, the stakes are sufficiently low: the opacity of Brainly’s EULA aside, it is difficult to argue that anyone has been harmed by having their content unknowingly open-sourced to the educationally minded public. Yet users have proven eager in the past to allow their likenesses to be sold by Instagram,Footnote ⁴⁰ and (as we discuss in Chapter 8) they have more recently allowed their (and their friends’) personality profiles to be mined by Cambridge Analytica, even if they later come to regret these choices. In the context of complex algorithmic systems, epistemic competence – and therefore, freedom – is impossible to achieve.

5.2.3 Social Challenges to Freedom

So far in this section, we have argued that one way that human agency can be undermined is if its processes come to be directed by affective states rather than by conscious deliberation, violating the authenticity condition. Another way agency can be undermined, we have argued, relates to the fact that our cognitive capacities themselves are bounded, violating the competence condition. In this subsection, we explore a third way that agency can be undermined: through the influence of the affective states and erroneous deliberative processes of other people, in the context of an epistemologically toxic social environment.

Consider, for instance, YouTube’s recommendation algorithm, whose autoplay mechanism drives users toward echo chambers and other ideologically extreme and conspiratorial content. By YouTube’s own analysis, almost three-quarters of viewing time spent on the site is driven by this recommendation system as opposed to their viewers’ independent actions,Footnote ⁴¹ so the operation of this system is of overriding significance in determining what content people are exposed to. When left unsupervised by humans, it fosters an unsafe environment for children: it allows bad actors to expose them to distressing parody content, such as a fake Mickey Mouse cartoon depicting eye-gougingFootnote ⁴² or a fake Paw Patrol episode where one of the main characters attempts suicide.Footnote ⁴³

YouTube’s recommendation system has also arguably been a source of violent self-radicalization. The reason has at least in part to do with YouTube’s business model and with the evolution of its incentive structure. Originally, the YouTube recommendation algorithm had been designed simply to maximize the number of clicks,Footnote ⁴⁴ but this proved to offer content creators an incentive to post “clickbait” videos, ones that entice users to initially click on the video but that do not necessarily hold their interest through the duration of the video. In light of this, YouTube changed its algorithm in 2012 to optimize more for watch time than for number of clicks so that “creators would be encouraged to make videos that users would finish, users would be more satisfied and YouTube would be able to show them more ads.”Footnote ⁴⁵ The change worked – watch time increased by 50 percent each year from 2012 to 2015 – but it also conferred an advantage to those content creators who produce naturally engaging videos: conspiracy theorists.

Guillaume Chaslot, a former Google engineer who left the company over issues with the development of the YouTube algorithm, wrote a piece of web software that examines which videos follow others via the recommendation system. When the Guardian examined the one thousand top-recommended videos, it found that YouTube had an 85 percent chance of recommending a pro-Trump video rather than a pro-Clinton video. The newspaper also interviewed several conspiracy theorists, whose videos normally receive only a few hundred views but whose traffic increased dramatically right before the 2016 election, and it found that most of these content creators got their traffic from the YouTube recommender system rather than from external links.Footnote ⁴⁶ These videos, which almost always have titles like “WHOA! HILLARY THINKS CAMERA’S OFF … SENDS SHOCK MESSAGE TO TRUMP” and “Irrefutable Proof: Hillary Clinton Has a Seizure Disorder!” are more engaging than they are factual. The natural result of this state of affairs is a social environment in which divisiveness and sensationalism are in themselves advantageous traits for content, quite apart from whether or not that content reflects misinformation or authoritarian aims.

Given the affective and deliberative challenges to agency that we have already discussed, we should not be surprised that sensational and conspiratorial content grounded in fear, anger, and resentment turns out to deliver greater engagement than the messy realities of legitimate news reporting. But it is worth noticing how toxic social environments can aggravate the other challenges to agency: When users who are already prone to fear-conditioning are fed an escalating diet of misinformed, radicalizing content, their autonomy is socially comprised. For such users, both the possibility of autonomy and the possibility of high-quality agency are ruled out from the start.

5.2.4 Summarizing the Challenges to Freedom

As we have seen in this section, drivers who have become caught up in the gamified ridesharing ecosystem, information consumers who have become overwhelmed by a deluge of labyrinthian pseudo-contracts, and conspiracy theorists who have become radicalized all have been made less free in a broadly similar way: either their autonomy has been undermined or their quality of agency has been diminished.Footnote ⁴⁷ However, the traditional conceptions of freedom (negative, positive, republican) that we canvassed at the beginning of the chapter do not provide an adequate explanation for these shortfalls of freedom. That is because drivers, information consumers, and conspiracy theorists all seem to satisfy the traditional conceptions. They are not subject to external constraints, they are not subject to arbitrary exercises of power, and they often even have the capacity to make choices to realize their preferences. The challenges to freedom we have outlined here make clear that freedom requires autonomy, high-quality agency, and non-domination.

Taking these three challenges seriously motivates an important shift in how we think about freedom. Recall that our account of freedom is ecological non-domination, which includes quality of agency and republican freedom. That account captures the emotionally volatile, cognitively limited, and fundamentally social nature of human agency. It is also consistent with the fact that it remains possible to coherently act even under some degree of individual, psychological disunity, such as when one is internally conflicted about a course of action. Surely the authenticity of Uber drivers, the competence of overwhelmed service users, and online radicalization of conspiracy theorists are not all-or-nothing affairs. People in those circumstances are neither fully free nor acting purely on reflex. When we reconsider human agency in light of its distinctive challenges, any plausible account of freedom must cope with this complexity.

We are finally able to make more precise the sense in which these users of technology have been made unfree: their beliefs, preferences, constraints, and values have been formed in a way that conflicts with agency. Freedom requires, in addition to non-domination and effective choice, that one’s beliefs, preferences, constraints, and values themselves have been formed without undue influence. It is only when people are both undominated and unstricken by such malformations that they can be said to be free in the fullest sense.