We often make bad choices. We eat sugary foods too often, we don’t save enough for retirement, and we don’t get enough exercise. Helpfully, the modern world presents us with a plethora of ways to overcome these weaknesses of our will. We can use calorie tracking applications to monitor our sugar intake, we can automatically have funds taken from our account to fund retirement schemes, and we can use our phones and smartwatches to make us feel bad if we haven’t exercised in a while. All of these might seem innocuous and relatively unproblematic: what is wrong with using technology to try and be a better, healthier, version of yourself?

Well, let’s first take a step back. In all of these cases what are we trying to achieve? Intuitively, the story might go something like this: we want to be better and healthier, and we know we often struggle to do so. We are weak when faced with the Snickers bar, and we can’t be bothered to exercise when we could be binging The Office for the third time this month. What seems to be happening is that our desire to do what all things considered we think is best is rendered moot by the temptation in front of us. Therefore, we try to introduce changes to our behaviour that might help us overcome these temptations. We might always eat before going shopping, reducing the chances that we are tempted by chocolate, or we could exercise first thing in the morning, before our brains have time to process what a godawful idea that might be. These solutions are based on the idea that we sometimes, predictably, act in ways that are against our own self-interest. That is to say, we are sometimes irrational, and these “solutions” are ways of getting our present selves to do what we determine is in the best interests of future selves. Key to this, though, is we as individuals get to intentionally determine the scope and content of these interventions. What happens when third parties, such as governments and corporations, try to do something similar?

Attempts at this kind of intervention are often collected under the label “nudging”, which is a term used to pick out a particular kind of behavioural modification program. The term was popularized by the now famous book, Nudge, in which Thaler and Sunstein argue in favour of “libertarian-paternalism”.

In line with what I have said above, the authors recommend that because there are reliable ways that we get things wrong (and hence act in ways that go against our own self-interest), we ought to implement certain “choice architectures” that guide our behaviour in desirable directions. The claim is therefore that it is possible to maximise individual welfare with this type of intervention. This kind of interference has been made even more effective by the rise of Big Data analytics.

Talk of “Big Data” refers to the fact that corporations and governments have massive amounts of data on each and every one of us that they can make use of for their own purposes. Usually that purpose is to try and make money or reduce costs, but sometimes, at least ostensibly, these groups say they want to help us be better. They want to “nudge” us in the right direction: From “opt-out” retirement plans to flies in urinals, these seem to be innocuous, and in fact properly beneficial, ways in which our choice architecture can be influenced by third parties to our own advantage. Moreover, through the massive amounts of data that are now available, it is possible to tailor nudges to specific individuals. The data may suggest that I am more prone to making impulsive decisions at night, while you are more likely to buy things early in the morning. I might be swayed by appeals to emotion while you might be susceptible to framing effects. Nudgers can take advantage of this in order to streamline their interventions and make them more effective, perhaps even turning a “nudge” into a “shove”. The combination of Big Data and the success of nudging could thus be conceived of not only as aid to our decision-making, but also as a threat.

The most intuitive objection to such a theory is that it goes against our individual liberty: nudgers get to decide what is best for nudgees, and so those on the receiving end of these interventions seem to have their freedom undermined. Ordinary folk are not trusted to make the “right” choice. While there might be cases where this could be justified (such as opt-out models for retirement and organ donation), this is not true for all cases and in all contexts.

However, I think there is an even deeper problem than this critique of liberty. What those in favour of nudging assume is that “welfare” ought to be understood in a “hedonic” fashion. What this means is that they argue that an individual can be said to be “doing well” so long as they, on average, have more pleasurable rather than unpleasurable experiences. Thus, nudging can be justified because it allows agents to act in ways that are desirable, meaning that they might indeed have more pleasant experiences than they would have in the absence of the nudge. So, even if it erodes the liberty of agents, because it allows them to avoid pain (that follow from bad decisions) and maximise pleasure (that follow from good decisions), nudging can be justified. What I want to do is outline, briefly, is why I think this narrow understanding of well-being is problematic in the case of nudging in particular, but also more generally.

That is because this hedonic account of well-being is far too narrow. It assumes that we are just pleasure maximizing, and that therefore whatever maximizes our pleasure must be good for our welfare. This is just wrong, and perhaps even exactly backwards. Oftentimes, it is precisely in the pursuit of what we find most meaningful in life that we experience a great deal of pain. Bodybuilding, marathon training, or making demanding career choices are all examples where we willingly put ourselves through painful experiences in the pursuit of worthwhile goals. Moreover, it is often the case that it is less about the goal itself and more about the process or the habits formed in the pursuit of the goal that make it especially valuable to us. Yes, running a marathon is a great accomplishment, but the discipline and mental fortitude one needs to have in training are not conducive to well-being in a narrow sense. Rather, they might contribute positively to our character over time, and are thus not merely instrumentally valuable. Nudging, by assuming a narrow and simplistic conception of well-being might therefore undermine well-being in this broader, more holistic sense. If we are not trusted to make our own mistakes and learn from them, how can we hope to be better versions of ourselves? How can we develop moral and intellectual virtues if we are not confronted with situations in which we can actively engage the moral reasoning that underpins many virtuous character traits? It seems that to account for these concerns we need a broader (or perhaps entirely different) conception of well-being, and of what it means to be “living well”.

Based on this, defenders of nudging would do well to adopt a different conception of well-being upon which to build their theory. It might even turn out that certain nudges are inimical to our well-being and having a broader account of well-being would allow us to specify and get a handle on this. This approach is exemplified in a recent paper by Steffen Steinert and Matthew Dennis, where they argue for a eudaimonic conception of well-being, which shares many of the characteristics of what I think a more inclusive account of the concept would entail. Of course, I have not defended a particular account of well-being here (although I hope to do so soon), the point, for now, has just been to point out some problems with such hedonic accounts, and show that the threat to freedom is not the only problem with nudging.

Getting a handle on the various ways that technology influences us is as important as it is difficult. The media is awash with claims of how this or that technology will either save us or doom us. And in some cases, it does seem as though we have a concrete grasp on the various costs and benefits that a technology provides. We know that CO2 emissions from large-scale animal agriculture are very damaging for the environment, notwithstanding the increases in food production we have seen over the years. However, such a ‘balanced’ perspective usually emerges after some time has passed and the technology has become ‘stable’, in the sense that its uses and effects are relatively well understood. We now understand, better than we did in the 1920s, for example, the disastrous effects of fossil fuels and CO2 emissions. We can see that the technology at some point provided a benefit, but that now the costs outweigh those benefits. For emerging technologies, however, such a ‘cost-benefit’ approach might not be possible in practice.

Take a simple example: imagine a private company is accused of polluting a river due to chemical runoff from a new machine they have installed (unfortunately this probably does not require much imagination and can be achieved by looking outside, depending on where you live). In order to determine whether the company is guilty or not we would investigate the effects of their activities. We could take water samples from the river and attempt to show that the chemicals used in the company’s manufacturing process are indeed present in the water. Further, we could make an argument where we show how there is a causal relationship between the presence of these chemicals and certain detrimental effects that might be observed in the area, such as loss of biodiversity, the pollution of drinking water, or an increase in diseases associated with the chemical in question.

In this example the issue is black and white: we have a guilty party who we can properly hold to account for their irresponsible use of a technology. However, part of the reason we can do this is because it is supposed that we expect the polluters to have known better, that is, there would be certain, foreseeable, consequences that would have followed from their activities. Let us call these kinds of effects of technology ‘hard’ effects. ‘Hard’ effects can be quantified, have a clear harm, and in addition causation can be established. These are not the kind of effects that I am interested in. It is in the ‘soft’ effects of technology where we encounter inherent ambiguity and contestability, and where getting a handle on what is really going on becomes quite a complicated, if not impossible task. ‘Soft’ impacts, by their very nature, are difficult to quantify, do not have clear and noncontroversial harms, and direct causation is difficult to trace.

For example, it has become common to attribute the seeming rise in political polarization to new forms of social media and their governing algorithms (Twitter, Facebook, etc.). This argument seems to make intuitive sense, as social media seems to be characterized by increasingly extreme views and perspectives, and decreases the costs associated with finding other like-minded individuals. But is this really the case?

Alberto Acerbi, for example, has recently argued that we have been seeing a rise in political polarization that extends to well before the introduction of social media, and that offline individuals might in fact be more polarized than those that spend their time online. He references a research study that measured levels of polarization from 1996, and where the researchers found that age was the biggest factor in predicting an increase in polarization. This actually fits well with data from the election of Donald Trump in the US and the Brexit campaign in the UK: older people tended to be more pro-Trump and pro-Brexit, and these are people who we expect would use social media the least. Therefore, we have higher levels of polarization in communities that spend less time on social media platforms. Of course, this does not settle the matter: social media analysis is a new and moving target for research, and the latest data suggests that the elderly might be using social media just as much as younger people, but perhaps for different purposes.

The point of this example is that the effects of the technology do not conform to the three conditions that underlie our ability to discern ‘hard’ effects. The causal picture is messy, and it is sometimes unclear whether the technology is the cause of some effect, or whether it is merely a contingent part of a decades long trend.

This is in part due to the fact that social media might be thought of as a kind of emerging technology: These are nascent technologies whose meaning, use, and impact is still unstable, where we do not yet know what their future consequences might be. As noted above, such technologies cannot be conceptualized in a ‘cost-benefit’ framework (given that we cannot specify these in sufficient detail) and so a ‘promises and perils’ framework might be more productive. Examples of other emerging technologies include (but are of course not limited to) humanoid robots, autonomous military drones, synthetic biology, and virtual reality.

For these kinds of technologies, we are better off framing their potential impacts in terms of promises or perils: The predictions of what a technology may afford are grounded in what they promise us (better health, more ecologically friendly living, a world without work) or their perils (reduced autonomy, AI overlords, threats to democracy). Such an approach is justified because of the nature of promises and perils: they admit of ambiguity and are not an evaluation of what is but of what could be, and so they better capture the uncertainty that comes with imagining the future than a costs and benefits analysis, which might be better suited to evaluating more ‘stable’ technologies.

As emerging technologies are by their nature in a nascent stage of development, detractors and defenders have to speculate (and often embellish) the effects that these technologies might have. Because the meanings of these technologies have not yet ‘stabilized’ and their place in society is not yet fixed, it is often difficult to disaggregate false promises from genuine ones. Decisions about who will fund the development of the technology, who will be able to make use of it, and the cultural context in which it will be embedded all play major roles in whether the technology will be a ‘success’ or not.

To go back to the example of polarization above, what we observe is an inherent ambiguity and contestability, and, perhaps most importantly, part of the effects we are attempting to measure are produced by the agents themselves. Thus, unlike in the case of a factory polluting a river, when it comes to soft impacts users more intimately co-produce the harms (or benefits) that are the object of our study. Such soft impacts, therefore, require an account of how we expect humans to behave when confronted with the technology, and so it is not enough to merely give an explanation as to how the technology might influence the environment. We require further specification about how an intervention might impact the habits, autonomy, and self-understanding that individual users have. This is not an easy task, as difficult questions include whether the ‘target’ of such studies should be agents’ mental states, their ability to use moral reason, or whether they are more virtuous as a result of the technology.

This means that when attempting to evaluate the potential effects of emerging technology we should be especially sensitive to the soft impacts that they may have. Attention to these soft impacts makes salient how a cost-benefit approach cannot fully capture the nuanced effects that emerging technologies can have on human social relations, culture, and politics.

In a previous article I argued that, when it comes to our moral appraisal of emerging technologies, the best normative framework to use is that of virtue ethics. The reasons for this were that virtue ethics succeeds in ways that consequentialist or deontological theories fail. Specifically, these other theories posit fixed principles that seem incapable of accommodating the unpredictable effects that emerging technologies will have not only on how we view ourselves, but also on the ways in which they will interact with our current social and cultural practices

However, while virtue ethics might be superior in the sense that it is able to be context sensitive in way that these other theories are not, it is not without problems of its own. The most significant of these is what is known as the ‘situationist challenge’, which targets the heart of virtue ethics, and argues that situational influences trump dispositional ones. In this article I will defend virtue ethics from this objection and in the process show that it remains our best means for assessing the moral effects of emerging technologies.

So, what exactly is the situationist challenge contesting? In order for any fleshed-out theory of virtue to make sense, it must be the case that something like ‘virtues’ exist and are attainable by human beings, and that they are reliably expressed by agents. For example, traits such as generosity, arrogance, and bravery are dispositions to react in particular ways to certain trait-eliciting circumstances. If agents do not react reliably in these circumstances, it makes little sense to traffic in the language of the virtues. Calling someone ‘generous’ makes no sense if they only acted the way that they did out of habit or because someone happened to be watching them.

One example of the influence of social factors comes from what is known as Asch’s conformity experiment. This was an experiment run in the 1950s by social psychologist Salomon Asch, with the intention of testing the extent to which social information could influence individual judgement. In the experiment, participants were taken to a room where they were joined by between six and eight other people, who they thought were also participants, but were in fact confederates of the experimenters. The experimenter then showed all of those in the room two cards. The first had a line of a specific length, and the second had three lines. The experimenter then asked each participant which of the three lines in the second card was the same length as the one on the first card (see the image below).

It is relatively obvious which of the lines would be the correct answer. However, in each case, the participant was always the last one to give their answer, and so the confederates would answer first. Each round was composed of 18 trials, and in 6 of these trials confederates gave the correct answer, but in the rest, they unanimously agreed on the “wrong” line. Over the course of all the trials, over 35% of the participants were influenced by the majority opinion. This is quite an incredible result, considering that, first, the stimulus in question was not ambiguous in any way, and second, in the cases where subjects were left to their own devices, they made no mistakes whatsoever. Thus, we have a case of significant social influence. To bring this back to our discussion on moral virtues: it seems we have an example here where situational influences unduly influenced the participants’ responses.

The situationist challenge is essentially a complaint about the fragility of individual dispositions, and the depressing ease with which situational influences better account for our behaviour (like in the conformity study above) than any talk about virtues. This kind of worry is especially pernicious in the age of the internet, where we are exposed to massive amounts of social information on a daily basis, from all different kinds of people, some who are genuinely trustworthy, and others who might have nefarious intentions. There is a very real worry that this kind of information overload can lead to the reduced autonomy of individuals and undermine their ability to act virtuously.

However, there are a few responses that the virtue ethicist can make against such complaints. The first comes from being a bit more nuanced in our understanding of individual character and its associated psychology by distinguishing between continence, incontinence, vice and virtue.

The continent person eventually does the right thing but is nonetheless conflicted when doing so. The incontinent person is also conflicted but ends up doing the wrong thing at the end of the day. The vicious person does the wrong thing unstintingly, and the virtuous person does the right thing unstintingly. Thus, we might respond to this challenge to virtue ethics by stating that just because we observe that virtue is a rare trait, this does not necessarily mean that those who act in vicious ways are doing so simply out of vice: they might be merely incontinent. When evaluating studies that claim to show that we lack dispositional virtues (such as the famous Milgram studies in obedience) we should be attuned to whether subjects exhibited displays of emotional distress when performing blameworthy actions (and in the case of the Milgram studies, this was indeed what happened). If so, we can say that they were incontinent, and thus we need not give up the perspective of virtue ethics, as it provides us with a coherent way of describing what might be going on.

The second response to the situationist challenge consists in contesting the implications of the data coming from research in social psychology. Indeed, it does seem as though there is a substantial amount of evidence from both social psychology and behavioural economics that suggests we are susceptible to biases and the use sometimes irrational heuristics. However, these situational influences and cognitive biases are part of what we might abstractly call “human nature”, and therefore must play a role in our moral theorizing. We are by our nature imperfect decision-makers, and it is therefore predictable that we will err in our reasoning and in our actions. This does mean that we are somehow, at base, irrational: we can reflect on our own fallibility and attempt to seek out circumstances that are conducive to our own flourishing, by avoiding precisely those ‘temptations’ that we know might lead us astray.

For example, think of our insatiable appetite for sweet foods. From an evolutionary perspective this trait makes sense, as the relative scarcity of such foods in our past, combined with their immense nutritional value, allowed this disposition to flourish. Modernity, however, places before us a plethora of sugary snacks that we would do better to avoid. Doing so is easier said than done, but a common strategy is to not go shopping when hungry: this reduces the temptation to buy high glucose foods (which provide a spike in blood sugar that makes us feel good for a short time but can end up increasing hunger levels later on). Thus, if you know you have a sweet tooth, this is a simple strategy to overcome your potential “irrationality” when confronted by this kind of temptation.

This suggests a shift in the way that we ought to think about the environments that we find ourselves in. Instead of viewing ourselves as passive consumers of circumstance, we should instead see things as though we are active producers of situations. This kind of approach is insightfully argued for by Mark Alfano, and he claims that instead of

“simply seeking and avoiding situations based on their virtue-conducive properties, we may take a more active role and create (both for ourselves and for others) situations with an eye to their virtue-conduciveness”

To go back to the example of sugary foods: we know refined sugar is bad for our health, and thus we can implement strategies to curate our environments in such a way that helps us avoid temptation by, in a sense, structurally disabling the temptation in the first place. For example, one might try to cook all one’s meals at home (which, again, I know is easier said than done).

Thus, the response to the situationist challenge essentially comes down to the claim that yes, there are myriad social influences on our behaviour and therefore our character, but that instead of feeling threatened by this, we can instead use it to our advantage. By paying careful attention to these contextual factors, we can design systems that promote human flourishing. Straightforwardly enough, this kind of approach can be applied to technology, as new technologies often influence our habits and ways of interacting with one another. Paying careful attention to the kinds of novel situations that technologies bring about can shed light on the potential moral consequences they might have. There is much more that could be said about this, and perhaps next time I’ll take a look at some specific technologies and investigate what virtues they promote or inhibit. Doing so now, however, would take me beyond the scope of this piece. In any case, I think I’ve gone on for long enough.

In 2007 Wesley Autrey noticed a young man, Cameron Hollopeter, having a seizure on a subway station in Manhattan. Autrey borrowed a pen and used it to keep Hollopeter’s jaw open. After the seizure, Hollopeter stumbled and fell from the platform onto the tracks. As Hollopeter lay there, Autry noticed the lights from an oncoming train, and so he jumped in after him. However, after getting to the tracks, he realized there would not be enough time to get Hollopeter out of harm’s way. Instead, he protected Hollopeter by moving him to a drainage trench between the tracks, throwing his body over Hollopeter’s. Both of them narrowly avoided being hit by the train, and the call was close enough that Autrey had grease on his hat afterwards. For this Autrey was awarded the Bronze Medallion, New York City’s highest award for exceptional citizenship and outstanding achievement.

In 2011, Karl-Theodore zu Guttenberg, a member of the Bundestag, was found guilty of plagiarism after a month-long public outcry.  He had plagiarized large parts of his doctoral dissertation, where it was found that he had copied whole sections of work from newspapers, undergraduate papers, speeches, and even from his supervisor. About half of his entire dissertation was stuffed with uncited work. Thousands of doctoral students and professors in Germany signed a letter lambasting then-chancellor Angela Merkel’s weak response, and eventually his degree was revoked, and he ended up resigning from the Bundestag.

Now we might ask: what explains this variation in human behaviour? Why did Guttenberg plagiarize his PhD, and why did Autrey put his life in danger to save a stranger?

The most intuitive way to explain this is to make an appeal to different character traits. It seems that Guttenberg lacked honesty, as he was willing to contravene German law and academic standards to save himself some time and effort. Inversely, Autrey was courageous, as despite the fact that he was placing himself in harm’s way, he nonetheless found the courage to act and save a stranger’s life.

Both courage and dishonesty are part of our moral vocabulary, and, more specifically, we think of them as virtues and vices. Virtues being those dispositions or character traits that we praise persons for possessing, and vices being those that we find them blameworthy for possessing. In virtue ethics, one of the three main branches of normative ethics (the other two being deontological and consequentialist ethics), virtues play a foundational role in moral appraisal, and the rightness or wrongness of an action is to be cashed out in terms of stable character traits. Thus, the focus point of virtue ethics is on agents. This is in contrast to the other two normative approaches mentioned earlier. In consequentialist ethics (of which utilitarianism is the most common offshoot), the focus is, naturally, on the consequences of people’s actions. In Autrey’s case above, we could say that he acted in a morally praiseworthy fashion because of the fact that his actions led to him saving a life, which increased (or maximized) overall utility (or ‘happiness’). Deontological ethics (of which Kant’s moral philosophy is the most famous offshoot) concerns itself with how we ought to show respect for persons and the principles that we act on. For example, in Guttenberg’s case, we might say that he violated the maxim “do not lie”, and therefore ought to be held responsible and blamed. However, for the consequentialist such maxims or principles play a secondary role, as the rule “do not lie” might justifiably be broken so long as it produces more desirable consequences than truth telling.

These traditional moral theories have dominated discussions about what it is we ought to do in many different situations. The question I want to now ask is which of them, if any, is up to the task of helping us chart a course through emerging technologies. I am not going to claim that one of these theories is the best all things considered, but rather suggest that one might be best in the specific case of emerging technologies.

So, while the sketch I have presented here is obviously not the full story, the point was merely to outline some essential components of the three ethical theories. Before seeing how they fare in the face of emerging technologies, we need to first get a handle on the key characteristics of these technologies: they often create novel moral situations, in which traditional moral concepts might be challenged. For example, self-driving cars might challenge what we think of as a ‘moral agent’, Lethal Autonomous Weapon Systems (LAWS) might challenge our traditional concept of ‘moral responsibility’, and the emergence of Big Data techniques challenge ideas of what ‘privacy’ entails in the digital age. Let us go through each theory one by one.

Consequentialist theories face problems due to the unpredictable effect that emerging technologies may have. If we have high levels of uncertainty regarding the potential consequences of our actions (and how they interact with technology) then the abstract calculation of utility seems almost impossible. It would seem that holding people responsible due to consequences that they could not have reliably foreseen would be unfair.

For the deontologist we find a problem emerges due to its focus on fixed or universal moral laws. With emerging technology, we must pay attention to the specific context in which it might be embedded, and while such universal principles may be of help, they are insufficient by themselves. Having an abstract moral principle does not help us when we are attempting to figure out the ways in which novel technologies might challenge our existing understandings of morality, by, for example, extending the class of entities that might be counted as moral agents (such as LAWS).

This leaves virtue ethics, which I think holds out the best hope for our dealings with emerging technology. This is because of its focus on prudential wisdom, and its emphasis on the ability to respond appropriately in a variety of situations. This avoids the problem of unpredictability, as the focus is not on consequences, but on inculcating certain moral virtues. It is here that we can note one of the essential features of virtue ethics: the focus on right action (or excellence) is not only about the action itself, nor is it wholly concerned with following the correct rules. Rather, the central point feature is in the cultivation of right actions over time, and the aforementioned virtuous states emerge from, however slowly, continued action. In this way virtues are a kind of muscle that we need to train properly. The way we build a virtue like temperance, for example, is by observing those who we believe extoll this virtue and attempting to model our behaviour on theirs. Such exemplary persons serve as the guiding star in our cultivation of the moral virtues. This is not to say that this is the only way to get a handle on the virtues, however. As noted above, reason responsiveness also plays an important role.

This provides us with a very useful frame to use when analyzing emerging technologies, as we are not stuck trying to find a priori criteria for moral truths but can rather focus on the conditions which enable human flourishing and use these in the design of various technologies.

Take digital communication, which includes instant messaging applications but which for my purposes will exclude things like video calls (once again calling attention to how we need to be specific when assessing any technology). The utilitarian might say that face-to-face interaction maximizes utility because people enjoy it more, and we should therefore try to minimize virtual interaction. The deontologist might say that computer mediated interaction is intrinsically less valuable than face-to-face interaction. Of course, these example arguments are merely illustrative – there might be much more compelling consequentialist and deontological arguments. In each case, however, notice that the conclusion is that the technology must either be good or bad, and there is little room for context. Virtue ethics, however, gives us the possibility of providing far more nuanced responses. We might say, as Shannon Vallor does, that

“The morally significant virtues of perseverance and commitment tend to have weaker expression in present online social environments where anonymity is combined with low entry and exit barriers, when compared with social environments (online and offline) with less anonymity and/or higher entry/exit barriers.”

Thus, virtue ethics gives us a way of understanding the effects that technology may have on agents as they interact with it over time and in different contexts. It can also make salient how exactly certain virtues might be promoted or hindered, and thus does not rely on either a strict technological determinism or on an entirely constructivist understanding of technology. “Technology” is not some monolith that when investigated gives us binary answers about how it ought to be used. Rather, each specific technology will have its own specific ways of influences our ability to flourish as humans, and virtue ethics provides us with a helpful way of tracking this influence.

We have always been a technological species. From the use of basic tools to advanced new forms of social media, we are creatures who do not just live in the world but actively seek to change it. However, we now live in a time where many believe that modern technology, especially advances driven by artificial intelligence (AI), will come to challenge our responsibility practices. Digital nudges can remind you of your mother’s birthday, ToneCheck can make sure you only write nice emails to your boss, and your smart fridge can tell you when you’ve run out of milk. The point is that our lives have always been enmeshed with technology, but our current predicament seems categorically different from anything that has come before. The technologies at our disposal today are not merely tools to various ends, but rather come to bear on our characters by importantly influencing many of our morally laden decisions and actions.

One way in which this might happen is when sufficiently autonomous technology “acts” in such a way as to challenge our usual practices of ascribing responsibility. When an AI system performs an action that results in some event that has moral significance (and where we would normally deem it appropriate to attribute moral responsibility to human agents) it seems natural that people would still have emotional responses in these situations. This is especially true if the AI is perceived as having agential characteristics. If a self-driving car harms a human being, it would be quite natural for bystanders to feel anger at the cause of the harm. However, it seems incoherent to feel angry at a chunk of metal, no matter how autonomous it might be.

Thus, we seem to have two questions here: the first is whether our responses are fitting, given the situation. The second is an empirical question of whether in fact people will behave in this way when confronted with such autonomous systems. Naturally, as a philosopher, I will try not to speculate too much with respect to the second question, and thus what I say here is mostly concerned with the first.

Additionally, I don’t wish to engage in trolley-problem style philosophical games. Rather, I want to investigate what I think is an interesting presupposition at the heart of what many people think about moral responsibility: that in order to be morally responsible, an agent should also be free. That is, before we even come to questions of whether a self-driving car could be responsible (as I suggested above), many people would dismiss this out of hand because it is clear that the car has no free will.

In a previous article I explored the epistemic benefits that we might accrue from our belief in our own freedom (even if such a belief turned out to be false). There I argued that it might not matter whether we can reliably cash out what free will “is”, metaphysically speaking, and whether we have it or not. Famously, P.F. Strawson made similar remarks with respect to our freedom. With respect to AI systems, this might mean that even if they do not have free will, they might be properly held morally responsible, because, as I will show, the conditions for holding an agent morally responsible are less metaphysically demanding that those for characterizing an agent as having free will.

In his influential Freedom and Resentment (1962) Strawson makes the compelling case that “traditional” ways of going about justifying our responsibility practices get things exactly backwards. Instead of looking toward objective or external conditions that need to be met in order for people to be morally responsible, we should look inwards, towards our actual practices of holding one another responsible. When we understand responsibility in this way, according to Strawson, we find that our “reactive attitudes” (such as resentment) are constitutive of these practices, and not merely inconsequential aftereffect. This pragmatic approach sidelines discussion of free will or whether our world is deterministic by accounting for all of our responsibility responses in terms of the reactive attitudes.

Strawson begins his analysis by focusing on a specific class of emotions – what he calls the reactive attitudes, which he argues play a constitutive role in the way we hold one another morally responsible. Strawson believes that these emotions are the bedrock upon which our practice of holding responsible rest, and that this is due to a natural disposition shared by all human beings to care about what others think of them.

When we add interactive technology to this equation something interesting happens. Namely, the possibility is raised that our personal feelings and reactions come to be attached not just to human beings but also to artificial systems. This is easy to imagine, as we already have favourite pens, chairs, and hats. It is not hard to see how a “socially interactive robot” might come to be part of the set of things that we care about, independently of any substantive metaphysical answer pertaining to the actual moral status of the artifact.

This is not to say that these systems have the same moral status as human beings, but that humans, in their interactions with these systems, might still come to expect certain things from them, and should these desires not be fulfilled, might be disappointed or confused at such an outcome. And sure enough, people already a) come to form attachments to technological devices, and b) these devices are being designed in ways that explicitly promote such attachment. Think of PARO, the therapeutic and interactive robot, designed to provide comfort to those in nursing homes or hospitals. Patients in the presence of such robots are encouraged to develop attachments to them, and this might lead to certain expectations. For example, consider this quote from the PARO website: “By interaction with people, PARO responds as if it is alive, moving its head and legs, making sounds, and showing your preferred behaviour. PARO also imitates the voice of a real baby harp seal.” It only seems natural that people would develop attachments to such entities. And again, this is independent of their “actual” moral status.

This argument essentially rehashes a very natural intuition: because I (as a human being) have moral full moral status (and all the abilities that come with it), I deserve consideration. This consideration extends to the things that I deem important and value, and the moral status of these objects is thus “derived” from mine in this sense. In this way the question of the moral contours of our relations to social robots (and artifacts more generally) is brought into focus: it might indeed be wrong to harm such objects, not because this harm matters to them, but because this harm matters to persons. The way we live with and interact with technology is often reflective of our own moral character, and so it makes sense to take seriously the moral implications such relationships might have, independently of whether the systems themselves have an intrinsic sort of moral status.

However, my question was not really concerned with whether we owe AI systems anything, but rather whether they might be responsible, in the sense of being on the hook, for various harms. The digression above was to show how our experience of various technologies quickly becomes clouded by questions of responsibility, quite independently of any substantive metaphysics, and that these attachments (to pens, cars, and books) are important to us.

So, back to my original theme. Because currently existing machines lack anything like the complex internal psychology that we humans have, it seems unreasonable to think that they would experience anything like the reactive attitudes mentioned above. However, this is not just distinctive of machines: take the example of psychopaths. We punish psychopaths when we find them guilty of performing morally bankrupt actions by placing them in prison. This punishment is independent of whether they “deserve” to be punished, or whether the punishment will lead to an improvement in their behaviour (due to their lacking an emotional component to their decision-making). However, there are clear pragmatic benefits to society with this kind of system in place, as it ensures that psychopaths are kept in effective quarantine and are not capable of harming others. In a sense, then, we regard them as natural objects, who are not fully morally responsible for their character or behaviour. They are treated as if they are things, and our “punishing” them comes not from considering whether they deserve it or not, but rather from considering the potential consequences of not doing so. It seems that our treatment of autonomous machines could proceed along similar lines. Specifically, when we find a machine to have performed some moral or immoral act and attempt to hold it responsible, we find that it is the wrong kind of agent to be a fitting subject of our responsibility ascriptions (at least for now). In this way we might adopt what Strawson termed the “objective attitude”, which commits us to treat agents as though they are not deserving of either praise or blame. Essentially, these agents become worthy of consideration in terms of social cohesion: we hold them “responsible” in ways that maximise social utility, as opposed to treating them as genuinely responsible agents. This might be one way in which we an come to understand the emotionally fraught relations we may come to have with ever more autonomous machines in the future.

From a heat dome in North America, people drowning in their basements in New York, and a climate famine in Madagascar, you would think we would have started to take the climate crisis seriously. This is to say nothing of the volumes of scientific evidence that support the theory that we are teetering on the edge of catastrophe and are in the middle of an extinction event. With this backdrop, you would be forgiven for thinking that an event with the purpose of addressing this crisis would propose significant changes to our current production and consumption patterns. As luck would have it, we seem to inhabit the worst of all possible worlds, where such a common-sense expectation is not met.

The 26^th Conference of the Parties (or, COP26) promised much but delivered little. Before the event, there was a genuine sense that this might be a turning point in the fight against climate catastrophe: maybe world leaders could come to together and, for once, put the long-term welfare of our planet and those who inhabit it over short-term profit. Unfortunately, what emerged from COP26 was not very much of anything. Although the so-called Glasgow Climate Pact, agreed to at COP26, “moves the needle” it is nowhere near enough to stop global warming from exceeding the critical threshold of 1.5°C above pre-industrial levels (our current pathway is for an increase of  2.4°C).

What remains clear (and what was reinforced at COP) is that there remains a gigantic disconnect between what is needed to get a handle on the climate crisis and what is being proposed. Talks of $100bn in aid from the developed to the developing world fall far short of what is actually required. John Kerry, the chief American negotiator, echoed this point by claiming that it is not billions that we need, but trillions (between $2.6tn and $4.6tn, per year).

After decades of U.N. climate meetings, endless calls to action from activists, and compelling scientific evidence that we need to do something drastic, the build-up of carbon dioxide in the atmosphere has continued to increase (at a growing rate, no less, with more than half of all CO2 emissions since 1751 emitted in the last 30 years). The  biggest culprits, naturally, are companies in the fossil fuel industry, who also happened to be the biggest delegation at COP26.

So what, in concrete terms, was the suggested route out of the crisis proposed at COP26? Perfectly in line with the pervasive neoliberal logic rotting the brains of most policy makers, the private sector was touted as a band aid for the burning building that is our planet. We still seem to (somehow) be stuck at the point of acknowledging (scientifically) that we are in the grips of a crisis but lacking the (political) coordination required to enact the radical action that the situation demands. The purported private sector “solution” would be one in which massive investment firms, such as BlackRock, direct trillions of dollars into low-income economies to accelerate their transitions away from fossil fuels. This will only happen, however, if the World Bank and the International Monetary Fund come in to “derisk” these “investments”. How the same neoliberal solution can be proffered at every crisis since the 1990s is beyond comprehension.

In any case, besides the “bailout” assumption that seems to be operative for private business, it also seems as if there is a version of market or technological “solutionism” at play. Market solutionism assumes that given enough time and resources, the abstract concept known as “the market” will solve our problems (and hence we are justified in setting our money printer machines to overdrive). Technological solutionism starts with the presumption that there should be technological solutions to our problems, and that other kinds of solutions are by definition inferior. These two assumptions come together in a deadly way with respect to the climate crisis, as market solutionism gives hope to neoliberal interventions, whilst technological solutionism maintains hope that a new technology will come along and save the day.

However, to think in this way is to entirely misunderstand the nature and scale of the crisis we are facing. In the first case, the problem is global and not local. While solutionism might be useful for fixing local issues, global issues require massive political and social buy-in. In order to implement a market or technological solution, the specific product needs to be introduced into new markets, adapt to local conditions, and not violate any local customs or norms. This takes time, is costly, and successful implementation on a global scale is not guaranteed. To tackle the climate crisis requires us to first achieve the political and economic support of those who have the power to make a difference.

Unfortunately, nothing seems capable of convincing those with such power to in fact do anything. As we edge, nay, plummet, over the cliff of catastrophe we need to acknowledge that we are beyond the point of reversing the damage that has been done. Current global warming levels are at about 1.2°C, and this increase cannot be reversed on human timescales, it can only be halted. The carbon that currently sits in the atmosphere will still be there in the year 3000. No matter how good our technology, it cannot undo what we have done. It can only give us more time to prepare for catastrophe.

With COP26 behind us and very little achieved, we are faced with the standard neoliberal bet: carry on as usual and hope that something unpredictable saves the day. Unfortunately, in this case, it might be the last bet we make.

An easy way to ruin any conversation is to start talking about philosophy. An easier way to do so is to mention free will. The issue of free will (whether we have it, if it is compatible with determinism, whether it even matters, etc.) has plagued philosophers for quite some time now. This is might be worrying, as it seems very important that we are free. How else can we fairly be held responsible for what we do? If your actions are fully determined by antecedent causes, what role do you really play? Additionally, reaching a consensus on what exactly free will entails is notoriously difficult. Is it enough to have some kind of “control”? Must the world be indeterministic? Does our best science exclude the possibility of free will? And, perhaps most provocatively, perhaps we don’t have free will at all, and that it doesn’t actually matter!

What I want to do here is take a somewhat different approach to the problem of free will. Instead of trying to figure out what exactly free will is or whether we need it, I want to start with a commonly accepted intuition: most of us, at least some of the time, feel as though we are free. From a first-person perspective, it really does (at least to me) feel as though we are in control of what we do, and that there is some central “willer” behind our actions. Moreover, whether we or not we really “believe” in free will or not, it seems this feeling of freedom will not go away.

So, even if we don’t have free will or cannot agree on its metaphysical content, we at the very least have a sense of freedom. Of course, whether this feeling is itself “free will” in some stricter sense is contentious. But my point is not to give a strict definition of free will, but rather to focus on a specific aspect of our believing that we are free in some broad sense.

Most discussions of this sense of freedom focus on the various pragmatic benefits that such a belief confers: if we believe ourselves and others to be free, we can justifiably hold one another morally responsible. If agents are not in control or the source of their actions, it would be odd to blame (or praise) them for what they do. Hence, belief in free will can facilitate the just attribution of moral responsibility (as accountability). Moreover, our entire criminal justice system seems to be predicated on the idea that people are free, and so widespread belief in our lack of freedom might be socially disastrous. There are many more pragmatic benefits, and here I have just mentioned a few, because my main point is that what has not been explored with respect to free will are the potential epistemic benefits of this belief.

While pragmatic considerations are traditionally focused on the useful consequences of a specific belief (things going well), epistemic considerations are much narrower in their scope, and home in on what implications certain beliefs could have on our ability to acquire knowledge about the world (getting things right). Moreover, even if a belief is false, it might still confer some epistemic benefits. Thus it need not be epistemically good overall. So even if those who argue that free will is an illusion or does not exist could still follow this kind of argument.

So, how might belief in free will confer such benefits? I will defend two claims. First, that belief in free will can fill an important gap in the explanatory process. Second, that it facilitates the process of reason giving by endowing us with a sense of authorship for our actions.

We do not always have direct access to the underlying causes of our actions, and so when we are questioned as to why we acted a certain way we tend to respond with post-hoc rationalizations. A troubling example of this occurs in the self-reports of people who on the whole think they’re pretty decent human beings. When assessing application documents of students applying for a lab position, faculty members consistently rated male applicant’s CVs as better suited to the job than their female counterparts. The confabulation becomes clear when the applicants’ supporting documents are controlled for: there is no other distinguishing factor between applicants besides their gender, and so this gender-bias seems to have been operative in faculty member’s decisions. However, the reasons given were that the male candidates were “more competent and hireable”. These reasons are of course false given the setup of the experiment, and so we here we have an example in which people really don’t know what’s going on inside their heads.

However, believing that we are free might allow us to fill a gap in this explanatory process: precluded from having direct access to our neurophysiology, our neural architecture (itself a product of our evolutionary history) has graciously provided us with a mechanism by which we are able to show the illusion of competence (a particularly widespread tactic among politicians). This user-illusion is best understood as an emergent property heavily dependent on our unique ability to use language and develop culture. It makes sense, in a way, for a creature with linguistic competencies to have a “self” that it can refer to when addressing others. This prevents us from having to claim ignorance with regards to our own actions: we feel that we are in control, that we are the authors of what we do, giving us the confidence to go out into the world and potentially form more true beliefs over time. Of course, it also means that we are often at a loss as to the true causes of our intuitions and actions, but a belief in the fact that these in some sense belong to us might help us weed out problematic beliefs. This links to my second point.

When we try and give reasons for our beliefs and actions, we are almost inevitably engaging in imperfect reasoning strategies (as noted above), as we cannot have all the required evidence to make perfect decisions, nor do we even fully understand how our own minds work (both neuroscientifically and psychologically). However, we can engage in practices in which we question the intentions that others may have when performing or not performing certain acts (“why don’t you donate some of your salary to charity?”, “why do you still eat factory farmed meat?”, etc.). Despite the epistemically faulty nature of our decision-making procedure, however, the process of reason giving itself might confer epistemic benefits. The way this comes about is that through reason giving we might come to better understand ourselves and make what are perhaps implicit beliefs explicit. Once explicit, these commitments can be challenged and revised if shown to be false when faced with evidence to the contrary. The act of speaking itself (as opposed to merely just thinking) can be thought of as providing a new medium which we can use to better understand ourselves, and the success of talk therapy lends credence to such a view. In this way belief in free will, belief that we really are the authors of our actions, opens up the possibility of a “peer-review” for our beliefs.

What I have argued is that regardless of the truth or falsity of our belief in free will, it confers not only certain pragmatic benefits, but also some epistemic benefits. Some implications of this are that perhaps the recurring discussions of determinism versus indeterminism that sometimes dominate discussions of free will are perhaps unnecessary. No matter the truth of either metaphysical thesis, we still feel free, and thus the belief in our freedom seems unlikely to change. In other words, it’s not just that we feel free, it is also that it seems we cannot help but act as if we are free, independently of some grand speculative metaphysics. With this in mind, it seems we should at least take the time to think through the epistemic benefits that this belief confers.

One of the cornerstones good of science is that its results furnish us with an objective understanding of the world. That is, science, when done correctly, tells us how the world is, independently of how we might feel the world to be (based, for example, on our values or commitments). It is thus central to science, and its claims to objectivity, that values do not override facts. An important feature of this view of science is the distinction between epistemic and non-epistemic values. Simply, epistemic values are those which would seem to make for good science: external coherence, explanatory power, parsimony, etc. Non-epistemic values, on the other hand, concern things like our value judgements, biases, and preferences. In order for science to work well, so the story goes, it should only be epistemic values that come to matter when we assess the legitimacy of a given scientific theory (this is often termed the “value-free ideal”). Thus, a central presupposition underpinning this value-free ideal is that we can in fact mark a distinction between epistemic and non-epistemic values Unfortunately, as with most things in philosophy, things are not that simple.

The first thing to note are the various ways that the value-free ideal plays out in the context of discovery, justification, and application. With respect to the context of discovery, it doesn’t seem to matter if we find that non-epistemic values are operative. While decisions about funding lines, the significance we attach to various theories, and the choice of questions we might want to investigate are all important insofar as they influence where we might choose to look for evidence, they do not determine whether the theories we come up with are valid or not.

Similarly, in the context of application, we could invoke the age-old is-ought distinction: scientific theories cannot justify value-laden beliefs. For example, even if research shows that taller people are more intelligent, it would not follow that taller people are more valuable than shorter people. Such a claim would depend on the value that one ascribes to intelligence beforehand. Therefore, how we go about applying scientific theories is influenced by non-epistemic values, and this is not necessarily problematic.

Thus, in both the context of validation and the context of discovery, we find non-epistemic values to be operative. This, however, is not seen as much of a problem, so long as these values do not “leak” into the context of justification, as it is here that science’s claims to objectivity are preserved. Is this really possible in practice though?

It seems both natural and acceptable that scientists choose certain research topics because they find them interesting, and that institutions fund and promote research in fields that are deemed important for political or social reasons. The value-laden nature of these contexts is all very well, again, so long as they do not come to bear on the context of justification. Justification here refers to how we assess scientific theories, and one way of defining it would be to say that it is based on empirical evidence in conjunction with various epistemic values. Together, these two elements, however they are understood, combine to aid us in assessing the scientific merit of a given theory. One problem with this, and the problem I will focus on, concerns the influence of non-epistemic values in the context of discovery, and the fact that these non-epistemic values amount to more than just the selection of a research topic. In this way there is not such a clear distinction between epistemic and non-epistemic values, which would undermine the value free ideal.

Non-epistemic values not only impact the selection of research topics, but also influence how this research is in fact carried out. How a research question is framed, and which aspects of it are deemed significant, are value-judgements. “Significance” is not something we find out in the world but rather something we attribute to things in the world, and the reasons for such an attribution could be social, financial, religious, etc. The point is, it need not be (in many cases is not) epistemic (although of course it could be). This becomes especially salient in the era of Big Science, where scientists are not limited to working in siloed academic laboratories but sometimes find themselves in large-scale organizations with political backing (and political agendas).

And this feature of science, well, is significant. This is because scientists cannot disinterestedly investigate every aspect of the world and must be by necessity picky. And so, which hypotheses are developed and tested, and which data are generated, are all influenced by these non-epistemic values. This is not to say that scientific theories are determined by these values, but rather to emphasize how these values can come to guide and shape research programs and the eventual support of specific theories.

Going back to the context of justification, which you will recall, was concerned with the evidential support for a theory and various epistemic values. However, if, as I have argued above, the data that are generated and the research directions that are pursued (and hence the available theories) involve non-epistemic values, then the context of justification also seems to involve such values. Essentially, we will always have “blind spots” when deciding on the best theory that explains some phenomena. The value-laden nature of the context of discovery prohibits us from coming up with theories that might have been viable alternatives to the one chosen, and so there is really no “value free” part of scientific investigation. Thus we see how a central assumption underpinning the value free ideal (that of an independence of justification from discovery) is not maintained in practice.

This, however, does not undermine science’s objectivity. I have not dealt with the empirical adequacy claim in the context of justification, and there is much to be said about how this constraint supports claims to objectivity in science. However, what this argument does do is shine light on how the idea of a “pure” science free of values is untenable. Even if you are unconvinced by the argument wholesale, it seems right (at least socially and politically) that we do not constrain our choices of research lines on epistemic grounds alone. Such a constraint seems far from what actually occurs in science, and, moreover, could be seen as ethically indefensible as new social issues arise and demand scientific investigation.

Human beings are rather silly creatures. Some of us cheer billionaires into space while our planet burns. Some of us think vaccines cause autism, that the earth is flat, that anthropogenic climate change is not real, that COVID-19 is a hoax, and that diamonds have intrinsic value. Many of us believe things that are not fully justified, and we continue to believe these things even in the face of new evidence that goes against our position. This is to say, many people are woefully irrational. However, what makes this state of affairs perhaps even more depressing is that even if you think you are a reasonably well-informed person, you are still far from being fully rational. Decades of research in social psychology and behavioural economics has shown that not only are we horrific decision makers, we are also consistently horrific. This makes sense: we all have fairly similar ‘hardware’ (in the form of brains, guts, and butts) and thus it follows that there would be widely shared inconsistencies in our reasoning abilities.

This is all to say, in a very roundabout way, we get things wrong. We elect the wrong leaders, we believe the wrong theories, and we act in the wrong ways. All of this becomes especially disastrous in the case of climate change. But what if there was a way to escape this tragic epistemic situation? What if, with the use of an AI-powered surveillance state, we could simply make it impossible for us to do the ‘wrong’ things? As Ivan Karamazov notes in the tale of The Grand Inquisitor (in The Brothers Karamzov by Dostoevsky), the Catholic Church should be praised because it has “vanquished freedom… to make men happy”. By doing so it has “satisfied the universal and everlasting craving of humanity – to find someone to worship”. Human beings are incapable of managing their own freedom. We crave someone else to tell us what to do, and, so the argument goes, it would be in our best interest to have an authority (such as the Catholic Church, as in the original story) with absolute power ruling over us. This, however, contrasts sharply with liberal-democratic norms. My goal is to show that we can address the issues raised by climate change without reinventing the liberal-democratic wheel. That is, we can avoid the kind of authoritarianism dreamed up by Ivan Karamazov.

It is widely accepted in the scientific and political communities that we are on the precipice of a climate catastrophe. Through our own actions (and increasingly, inactions) we have plundered our planet and entered what some have called the ‘Anthropocene’: an age in which we have the agential powers equivalent to geological forces. This has given us the capacity to dominate the earth and spread the environmental disaster that is human civilization all over the planet. One good feature of civilization, however, has been the establishment (and defence) of liberal democracy. Such participatory democracies (at least in theory) encourage robust debate, diversity, and inclusivity. They also preserve individual liberty: freedom is an important political value (although of course not the only one that matters), and manipulation and coercion would be seen as unjustifiable in the pursuit of some or other political objective (even if this objective is climate change mitigation).

This brings us to the problem I introduced at the beginning of this piece: our liberal-democratic framework, by giving us the ‘freedom to choose’, has resulted in human stupidity being the guiding logic in many attempts to ameliorate the climate crisis. Might we be better off giving up our freedom and establishing an authoritarian state that could force compliance with global climate goals? The dilemma is as follows: Is there a way to deal with an existential threat such as climate change whilst also preserving liberal democracy? My own view (and that of Mark Coeckelbergh in his recent book on the topic) is that liberal democracy is worth fighting for, and it might just be possible for us to preserve it and not go extinct. To do so, however, we have to take seriously the authoritarian threat posed by AI powered systems as ‘solutions’ to the climate crisis.

Advanced AI systems are ubiquitous. From social media timelines, video recommendations on YouTube, and the kinds of adverts we see online, AI, in a very real sense, filters the world we see. Whether that world is rose tinted or stained red ultimately depends on what the algorithm thinks will maximise revenue for its owner in the best possible way. These systems also play on our emotions: we are fed information that confirms our biases, exposes us to a very narrow range of interests, and promotes extreme views. This content is often inflammatory, divisive, and polarizing. Combine this with work done on nudging in social psychology and behavioural economics, and what you get is the potential for an AI-driven authoritarian monster. Proponents of nudge theory argue that due to how easily ‘hackable’ the human mind is, we ought to design our world in such a way that makes it easier for us to ‘do the right thing’. For example, we might stop placing sugary treats at children’s eye level in stores in the hope that they might not see this form of legal poison, thus making it easier to lower their refined sugar intake. Termed ‘libertarian paternalism’ by its defenders, it is claimed that this approach does not necessarily infringe on the freedom of individuals.

However, it is difficult to see how having ‘nudgers’ decide on our behalf what they think is good for us ‘nudgees’ is not an infringement on our freedom. Nudgers decide for us. Why? Because we are not to be trusted. And here, I think, is where we see the guiding presupposition that has been operative in my discussion thus far: that we cannot trust human beings. If left to our own devices, we will inevitably muck things up (evidence for this can be found by either looking around or reading a book). Thus, our freedom can be curtailed justifiably because when given freedom we do Very Bad Things with it. And if data-driven AI systems operated by surveillance states are the way to implement such restrictions, and they mitigate the climate crisis, then so be it (call this the Hobbesian solution, named after Thomas Hobbes and his view of human nature).

Nonetheless, this view operates with an impoverished understanding of freedom. More precisely, we can see freedom or liberty in both positive and negative terms. On the mistrustful view of human nature outlined above, freedom is framed as purely negative. Negative freedom is our freedom ‘from’ things: freedom from interference by the state or other persons. Positive freedom, on the other hand, is about our freedom ‘to’. What we can do with our freedom, what its limits and possibilities are. On the Hobbesian solution, regulations and restrictions are justified, but are nonetheless framed as infringing on our negative liberties. What such an approach is silent on, however, is positive liberty. What if we are not inherently awful, what about those times when we are rational, and what about the potential we have to do Very Good Things? What if we can be trusted?

From a democratic perspective this would be preferable. Trust in human nature means that we can trust our leaders to get it right when it matters, and that we can trust citizens to vote for virtuous leaders. Of course, in practice, this is very difficult to achieve. However, when we see our political options through the lens of positive freedom, we begin to see how it might be possible. For example, one approach could be to enhance certain capabilities.  In this way we can increase our positive freedom by creating conditions and policies that enhance individual and collective well-being. Moreover, the means by which we achieve this should be democratic. Following Dewey, we can see such democratic processes as ‘organized intelligence’, where we can bring conflicts to the table, air them out, potentially solve them, and make progress. The goal of our liberal-democratic political institutions should therefore be to create the conditions under which human beings have their capability to increase overall welfare enhanced.

In this way it is possible to move beyond the Hobbesian solution and to place trust in our democratic institutions. Additionally, we can see how constraints and regulations to mitigate the climate crisis are not simply infringements on our negative liberties. Rather, they can be viewed as enhancing positive liberty by safeguarding our biosphere for both currently existing humans and future generations. This is not to say that we should limit our political discussions to anthropocentric perspectives. Human caused climate change is of course also devastating to animal and plant life, and it is essential that we include these in our political discussions. An important part of dealing with the problems posed by climate change will involve establishing non-anthropocentric political discourse. This would involve protecting the rights of nature, and the interests of animals. With this in mind, it is no use to blindly defend negative liberty when such a defense results in the destruction of our environment. Dealing with the climate crisis will require us to go beyond such narrow framings of our freedom.

What does any of this have to do with AI? To enhance positive liberty with respect to climate change we need regulation and oversight, and the same is true of AI. While many tout the wonderful things that will be made possible by new technologies, we must also remain cognizant of how such technologies can reinforce existing social inequalities. Not just that, but overreliance on algorithmic systems can blind us to (social) problems that might not have technical solutions. The climate crisis is not a problem that we can solve with science alone: it is also a deeply political issue, a kind of ‘hybrid’, and so if we want to use AI as a means to ameliorate it, we cannot ignore such political questions. The notion of positive liberty provides us with a framework which we can use to both evaluate attempts to align AI with human values and interventions in the service of our biosphere. Instead of a focus negative liberty, we should also be cognizant of the ways that restrictions, regulations, etc. might also enhance positive liberty.

What unites both the climate crisis and emerging technologies like AI is that they call into question various liberal-democratic norms. In both cases, the tug of authoritarianism on the one side, and unfettered libertarianism on the other, put strain on our political institutions. What I have hoped to do here is to show that we can escape this binary way of framing things, while still preserving that which is good about democratic participation.

This article originally appeared in the Monday Magazine at 3 Quarks Daily (see here).

When thinking about the claims made by scientists you would be forgiven for assuming that such claims ought to be true, justified, or at the very least believed by the scientists themselves. When scientists make assertions about the way they think the world is, we expect these assertions to be, on the balance of things, backed up by the local evidence in that field.

The general aim of scientific investigations is that we uncover the truth of the matter: in physics, this might involve discovering a new particle, or realizing that what we once thought was a particle is in fact a wave, for example. This process, however, is a collective one. Scientists are not lone wolves who isolate themselves from other researchers. Rather, they work in coordinated teams, which are embedded in institutions, which have a specific operative logic. Thus, when an individual scientist “puts forward” a claim, they are making this claim to a collection of scientists, those being other experts in their field. These are the kinds of assertions that Haixin Dang and Liam Kofi Bright deal with in a recent publication: what are the norms that govern inter-scientific claims (that is, claims between scientists). When scientists assert that they have made a discovery they are making a public avowal: these are “utterances made by scientists aimed at informing the wider scientific community of some results obtained”. The “rules of the game” when it comes to these public avowals (such as the process of peer-review) presuppose that there is indeed a fact of the matter concerning which kinds of claims are worthy of being brought to the collective attention of scientists. Some assertions are proper and others improper, and there are various norms within scientific discourse that help us make such a determination.

According to Dang and Bright we can distinguish three clusters of norms when it comes to norms of assertions more generally. First, we have factive norms, the most famous of which is the knowledge norm, which essentially holds that assertions are only proper if they are true. Second, we have justification norms, which focus on the reason-responsiveness of agents. That is, can the agent provide reasons for believing their assertion. Last, there are belief norms. Belief norms suggest that for an assertion to be proper it simply has to be the case that the speaker sincerely believes in their assertion. Each norm corresponds to one of the conditions introduced at the beginning of this article and it seems to naturally support the view that scientists should maintain at least one (if not all) of these norms when making assertions in their research papers. The purpose of Dang and Bright’s paper, however, is to show that each of these norms are inappropriate in the case of inter-scientific claims.

Nevertheless, we will argue that the authors have strong readings of each of the norms, and that such interpretations are neither necessary nor desirable. We claim there are good reasons not to read the norms in their strongest senses. Specifically, that the examples they use for each norm do not seem to capture a more benevolent meaning of that norm. Below we will attempt to reconstruct the authors’ argument, following which we will present our criticism.

To make their case the authors use the example of William Henry Bragg, an early 20th century researcher in the field of radioactivity. In the early days of research into radioactivity, physicists had not yet settled on whether X-rays were in fact “rays” at all. That is, there was a debate as to the fundamental nature of X-rays: were they waves or particles? Bragg developed a particle account of radioactivity where he argued against the dominant wave theory at the time. His proposal was that X-rays are in fact made up of two oppositely charged particles, a positive α-particle, and a negative β-particle, which formed a neutral pair. He disputed the popular wave theory (or “aether pulse theory”), and it was his contention that such theories could not fully account for the results he had observed.

Bragg was therefore advancing a particle theory of X-rays, which he made clear, in the form of public avowals, in many publications and letters to Nature from 1907 to 1912. What happens when we evaluate this example in light of the norms of assertion outlined earlier? First, factive norms: were his assertions true? As it turns out, Bragg was wrong: X-rays are not material particles and there are no neutral pairs, and today we know that X-rays have wave-particle duality and are composed of photons.

What about justification norms? Well, it seems that Bragg was aware that his theories were unlikely to be corroborated. Even at the time Bragg was writing, his theory was far from the scientific consensus in the physics community, and many believed his theories to be improbable. There was strong evidence for the “pulse theory”, and Bragg understood that he was going against what was widely accepted in physics at the time.

Lastly, according to Dang and Bright, Bragg did not even believe his own theory. In personal letters it appears that Bragg was quite circumspect regarding the tenability of his hypotheses. For example, in a letter to Arnold Sommerfeld, Bragg stated that “I do not wish to press this unduly or be dogmatic about it. It seems to me to be the best model to be devised at present, and I have no right to claim more”.

On the basis of this, Dang and Bright conclude that while Bragg’s public avowals “were false, unjustified on total evidence, and were not believed, we want to maintain that these avowals did not violate the norms of inter-scientific communication”. Instead of viewing the Bragg case as indicative of various epistemic issues in scientific communication, the authors argue that this is in fact a feature and not a bug of scientific public avowals. While we believe that Dang and Bright’s argument is an intellectually stimulating one, we have a general issue with regards to their operationalization of norms of assertion, and then some specific issues with the three norms they focus on. We first present our more general criticism and then follow this up with critiques of the specific norms.

To recap: The authors claim that Bragg, in his continuous defence of the neutral material particle theory, had made public avowals that he did not believe, turned out to be false, and were not properly justified given the available evidence.

At the general level, however, the authors seem to apply significantly different norms in the example than they have laid out in their introduction. For example, if one was trying to apply the strong versions of each of the norms to any inter-scientific public avowal (or any assertion made by scientists, for that matter) few scientists, if any, would actually be adhering to them. Nonadherence here would not mean one adopts a position of dogmatic scepticism, but arguably reflects a well-tempered epistemic humility. Such humility would acknowledge that current theories and assertions may turn out to be wrong in the light of further evidence in the future. On the other hand, if one takes a milder framing of the norm, e.g., that one ought to retract assertions that turn out to be false, Bragg’s gradual abandoning of a dogmatic defence of his position may be read in his favour. Especially given the fact that the alternative aether pulse theory also turned out to be false, to simply capture whole theories as either true or false, misses the point. Bragg could reasonably assert, if anything, that his theory was arguably as true or as false as the available alternatives until 1912. Interestingly, after 1912 his defence seemingly stopped. We will now supplement this general critique with some specific remarks about each norm, starting with factive norms.

In talking about factive norms of assertion Dang and Bright fail to acknowledge a distinction between saying something that one knows to be false and saying something that is proven to be false without knowing it is false. In the former case one lies, but in the latter one doesn’t. It seems what the norms of assertion guard against is lying or asserting misleading information, if that is the case, then it is possible to accommodate norms of assertion in inter-scientific communication, at the very least, to guard against lying to colleagues. In most cases when scientific claims are proven to be false we do not conclude that their proponents were liars, because we believe they set out to communicate something they believed to be true or justifiably supported by the breadth of evidence at their disposal.

Moreover, one may argue that even if the primary target of scientific publications is the wider scientific community, it cannot be denied that non-scientists also have access to these publications, which serve as a means of public science education. In fact, according to Elizabeth Anderson, one way laypersons can determine whether there is scientific consensus on a certain topic is to assess surveys, reviews or meta-analyses of the peer-reviewed literature. If laypersons have to consult scientific publications to determine scientific consensus on an issue of societal concern, then it becomes important that scientists adhere to some norms of assertion adequate for expert-layperson communication.

Second, we have belief norms. We argue that the authors either presuppose an exaggerated understanding of the norm or have misunderstood it. The authors opt for a quote from the ICMJE’s Recommendations for the Conduct, Reporting, Editing and Publication of Scholarly Work in Medical Journals, which in the original document serves to clarify the question of (co-)authorship. The paragraph in the ICMJE addresses responsibilities that come with (co-)authorship. Dang and Bright, however, suggest that the paragraph’s original meaning contributes to defining authorship within a group of multiple scholars. The authors refer exclusively to the part of the sentence that states that the researchers involved should have confidence in the work done by their colleagues (in the accuracy and integrity of their work). Dang and Bright use this detail to criticize a disproportionate expectation of compliance with the belief norm. What Dang and Bright leave out at this point, however, is that trust in the accuracy and integrity of colleagues’ work is not identical to faith in the resulting claims. There is a subtle but essential difference between these two. The citation chosen by the authors does not preclude researchers involved in a paper from believing the claims. Instead, the point here is to emphasize that researchers should have confidence that their colleagues have worked according to good scientific standards. This is what is emphasized by the emphasis on accuracy and integrity.

Notwithstanding this, abandonment of the belief norm can lead to two scenarios. In the first instance a researcher could fabricate data and therefore know that their claim cannot be valid (implying that they do not believe it). This point links up with our critique of factive norms above, where we argued for a distinction between instances of lying versus genuine ignorance.

Second, a researcher who has worked to the best of her knowledge and is confident that her colleagues have done the same may yet not believe in her claim because she is well aware it might be wrong. If this is the kind of scenario Dang and Bright had in mind with their critique of the belief norm, then their paper would come across as less provocative than its current reading suggests. However, their wholesale criticism of the belief norm, instead of narrowing its scope, leaves out considerations of scientific misconduct. It would have been easy to avoid this pitfall by interpreting the ICMJE recommendations in its entirety.

Finally, we come to the cluster of justification norms. In contrast to the authors, we choose a different order of the clusters. Our order sorts the norms according to their degree of indispensability, starting with the cluster of most minor relevance. The factual norms represent the weakest cluster because both lack of knowledge and truth must not be (especially in the humanities) an exclusion criterion for public avowals (excluding the case of agnotology). Provided that the research was conducted properly, the norms of belief can apply in such cases. Therefore, the cluster of belief norms follows in second place. Even if both clusters of norms cannot be fulfilled, their absence may be compensated by the cluster of justification norms. One can assert X without knowing whether it is accurate or believing it to be true — as long as one can justifiably explain (according to the rules of scientific reasoning) why one is asserting X. Or can’t one? Here a weakness is revealed, which will show that all clusters (but especially the justification norms) should be maintained. The assumption above may be sound under the conditions that all participants in scientific discourse are trying to produce knowledge. However, it ignores the fact that supposedly inter-scientific statements can also be made by representatives engaging in pseudoscientific, non-scientific or fraudulent conduct (who could forget Andrew Wakefield?). A contribution to the scientific discourse that does not take any of the cluster of norms into account loses, as it were, essential demarcation criteria that distinguish the contribution from pseudoscientific claims. If looked at another way: is it not the case that pseudoscientific claims at least play to the norms of belief? This makes it all the more important for scientists to draw a clear line between their contributions and non-scientific or pseudoscientific claims through norms of justification (if not through factive norms).

With respect to Dang and Bright’s specific argument for justification norms, they claim that Bragg’s theory was not justified by the total evidence available. Here they choose to apply arguably the strongest framing of the norm possible (as noted in our general critique above), instead of the version they introduced earlier: that assertions must be discursively justified. It seems Bragg did just that, when he defended his theory against his rivals, by arguing how his account was allegedly better able to explain certain observations than the prevalent alternative. If instead one sets out to apply a stronger framing of justification norms (qua requiring them to be in line with the total evidence available), arguably few, if any, scientific theory and assertions thereof would ever satisfy the norm. The authors seem to be somewhat aware of this as they occasionally restrict the requirement to available evidence to the person asserting at a specific point in time. Not doing so, would mean, that one may adhere to the norm at time t, but could retroactively fail to have adhered to it, once what the norm refers to as “the available evidence” changes over time.

Based on the aforementioned, it is indeed possible to claim that Bragg violated the norms of assertion. This is only possible, however, if one frames them in a way, that few, if any, scientific theories and assertions thereof would be able to abide by. If, on the other hand, one takes much less radical framings of said norms, Bragg did not only adhere to them, his adherence also does much to explain why his input, even if strictly false and unjustified, was valuable.

The norms of assertion in their weaker versions, may arguably make the division of labour in science more efficient, contrary to the claim by the authors. If Bragg would have chosen to abide by no norms of assertion, not even these weaker ones, he may have not been honest about the limitations of his theory and dogmatically stuck by it, thereby risking others to follow suit on a hopeless avenue of inquiry. One does well to remember that division of labour explains why a competition of ideas helps organize inquiry efficiently by helping divide up scientists into a number of potentially fruitful approaches, but coordination does not run contrary to that, if it allows scientists to help others avoid dead-ends. That is where moderate versions of norms of assertion may come into play. At the very least, then, scientific claims should be justified. Dang and Bright seem to be aware of this and therefore propose a contextualist justificatory norm for inter-scientific communication which is distinct from the norm of justification in the analytic epistemology literature. Contrary to Dang and Bright, we hold that a more charitable reading of the norms of assertion (more importantly, the norm of justification) in analytic epistemology leaves room for the kind of justification which Dang and Bright propose for inter-scientific communication.

*All the authors are funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) — Project 254954344/GRK2073, “Integrating Ethics and Epistemology of Scientific Research“.