1. Deepfakes and Deep-learning

Image manipulation has a rich history, pre-dating the advent of the internet and digital media. Under Stalinist rule, for example, political opponents and adversaries were not just executed but routinely removed from official photographs, deleting any trace of their existence from the public eye. The classic example is that of Nikola Yezhov – at one point, Stalin's right-hand-man and secret police official – who was notoriously removed from a photograph of Stalin and himself walking along the Moscow canal after he was assassinated. More recently, the advent of computing technology has ushered in programmes like Adobe Photoshop and CGI, which allows users to creatively edit images and video footage, inserting people and objects into places they never originally were. This has led to the creation of slightly more sophisticated versions of shallow-fakes called cheap-fakes, which rely on ‘cheap, accessible software’ to manipulate media (Paris and Donovan, Reference Paris and Donovan2019, p. 2). Given the extensive use (and misuse) of Photoshop and CGI, one might question whether deepfakes do pose a distinctive threat. That is, haven't people been able to subvert the truth with deepfakes and cheap-fakes alike (Paris and Donovan, Reference Paris and Donovan2019: 6)? Keith Raymond Harris (Reference Harris2021) has recently questioned whether the pessimism surrounding deepfakes is ‘overblown’. To an extent, Harris is right. Currently, many deepfakes are blurry, costly to produce, and easily identifiable. However, this should not fool us into thinking that this will continue to be the norm. As I mentioned in the introduction, part of the name ‘deepfake’ derives from deep learning, and it is this technology that arguably sets deepfakes apart from their predecessors.

At its most basic, deep learning works by computing a process and then attempting to carry it out. If the process fails, the technology revises it and tries again until it is successful. In this way, deep learning operates using a similar ‘trial and error’ method by which the human brain learns. It is for this reason that the deep learning techniques behind many of the emerging deepfakes today – including that of Queen Elizabeth II – are called deep neural networks. Roughly, these can be understood as algorithmic models that predicts and generate an output – in our case a video – based on an input of images or video content. As the cyber-security researchers Yisroel Mirsky and Wenke Lee (Reference Mirksy and Lee2020) explain, the neural networks often used to make deepfakes are so-called Generative Adversarial Networks (GANs). GANs consist of two algorithms – one called the ‘generator’ and the other called the ‘discriminator’ – which compete against each other to manufacture highly authentic videos. After being fed the same input of images or recordings, the generator creates new samples based on the input that are good enough to trick the discriminator, while the discriminator works to identify which samples are fake. Eventually, through trial and error, the GAN produces a highly sophisticated video that is capable of manipulating the speech inflections and facial movements of a target person doing and saying things they never did.

In this process, GANs are able to manipulate the attributes on a target's face – adding or removing features, lightening or darkening the skin colour, or swapping certain expressions on one person and superimposing them onto another. By doing this, deepfake purveyors can quite literally swap the identity of one person with another, and in some cases, even generate entirely new, non-existent faces from scratch (Tolosana et al., Reference Tolosana, Vera-Rodriguez, Fierrez, Aythami and Ortega-Garcia2020).Footnote ² This sort of deep learning manipulation is popular in the video game and film industries, but it has also been exploited to create faces to match fake online profiles on social media platforms from Facebook to Twitter. In the early days of GAN technology, the photographic and video content required to create a compelling deepfake was often in the hundreds or thousands, but recently Samsung has developed a technology that allows purveyors to do this with only a handful of photos of videos (Zakharov et al., Reference Zakharov, Shysheya, Burkov and Lempitsky2019). As the authors of the study report, a reasonably sophisticated deepfake could be created using only a single image. Alongside this, researchers at Princeton University have collaborated with Adobe to develop ‘VoCo’ technology, which allows video creators to modify the audio of a recording by typing the desired words into a transcript. The authors of the research observed how the algorithm was able to quickly analyse voice samples from a target, then synthesise what that person's voice would sound like, were they to say the words typed into the transcript (Jin et al., Reference Jin, Mysore, Diverdi, Lu and Finkelstein2017). As the lead author, Zeyu Jin, (Reference Jin, Mysore, Diverdi, Lu and Finkelstein2017) reports, when the synthesised words were inserted in the context of a spoken sentence, the modified sentence was ‘often perceived as indistinguishable from other sentences spoken in the same voice’.

Together, the use of VoCo technology and GANs marks a watershed moment not only in deepfake technology, but in media manipulation more broadly. For one, use of GANs has meant that purveyors can efficiently produce deepfakes that are increasingly indiscernible from authentic video footage. As the technology learns to improve, and the amount of video input required to produce a sophisticated fake decreases, deepfakes will inevitably become more widespread. This is observable, for example, in the rapidly growing number of deepfake apps and programmes available including Zao, Reface, Wombo, and DeepFace Lab, to name just a few.Footnote ³ As a result, deepfake production no longer lies necessarily in the hands of specialists as with CGI or Photoshop, but in the hands of a broader, non-specialist audience. Moreover, when VoCo technology is factored into this assessment, deepfakes become more credible than older forms of media manipulation because purveyors can design deepfakes with an accompanying, indistinguishable voice. In turn, we are more likely to accept the videos’ testimony unless there are significant reasons not to. Considering these developments, its unsurprising that the pioneer and computer scientist Professor Hao Li (Stankiewicz, Reference Stankiewicz2019) warns that we're going to get to a point where there is ‘no way that we can actually detect them [deepfakes] anymore’.

2. Vice Epistemology and Epistemic Corruption

If there is any truth to the above statement, then it's understandable why philosophers would worry that these videos could jeopardise our trust in videos (Rini, Reference Rini2020) or the information we gain from such a ubiquitous source (Fallis, Reference Fallis2020). My aim is not to press these issues any further but to address a currently unexplored problem that deepfakes risk generating. This problem is exposed by offering a corruptionist critique of deepfakes. Ordinarily, the verbal meaning of ‘corrupt’ refers to way in which the character of something or someone is made worse or degraded – like when we say that somebody has been ‘corrupted by power’. When we talk of people becoming corrupted, we often invoke the language of virtue and vice, specifically that a person has lost certain virtues or gained particular vices in their place. This is the function of a corruptionist criticism: to demonstrate how an environment causes one to acquire vices. To get a better sense of the kind of vices that I'm interested in here, we need to say something about vice epistemology.

Epistemology is the branch of philosophy concerned with knowledge, beliefs, and thinking, while vices are often considered to be defects, faults, or failures. Together, it is the job of the vice epistemologist to study what makes one's beliefs and thinking bad or faulty. Generally, when vice epistemologists approach this task, they tend to focus on failures of intellectual character – that is, the traits, attitudes, and habits of attention that hinder good thinking or knowledge, and which reflect badly on us. Some classic examples include close-mindedness, arrogance, and snobbery. Intellectual character failings such as these are what Quassim Cassam (Reference Cassam2016; Reference Cassam2019) calls the ‘vices of the mind’, or simply intellectual vices. These are contrasted with what Linda Zagzebski (Reference Zagzebski1996) calls the ‘virtues of the mind’, such as open-mindedness, intellectual humility, and creativity. Intellectual vices are bad for several reasons. First, they often reflect poor or deficient motivations towards engaging in knowledge-related practices. Call this the motivational account (Tanesini, Reference Tanesini2018; Reference Tanesini2021). Second, they systematically obstruct responsible inquiry and the transmission of knowledge. Call this the obstructivist account (Cassam, Reference Cassam2016; Reference Cassam2019), Third, they reflect entrenched patterns of bad judgement. Call this the judgement account (Baehr, 2020; Crerar, Reference Crerar2018).

Regardless of how one's vices play out in the actual world, they're usually the product of what we might call our intellectual environments. Think of these as physical or virtual spaces where epistemic practices take place. Amongst other things, these include practices such as inquiry, deliberation, educating, and evidence-sharing. They are ‘epistemic’ in the sense that they usually aim at discovering the truth of some matter. While some of us are lucky enough to engage with relatively healthy intellectual environments – where these practices are encouraged and facilitated – it's more likely that many of us will inhabit less than healthy intellectual environments; one's where inquiry is facilitated up to an extent then shut down after a certain point, where questioning authority in deliberations is a risky business, and where the evidence available to us can come from dubious or untrustworthy sources. In a world of disinformation campaigns, ‘fake news’ and ‘alternative facts’, it's hard not to find oneself in steadily deteriorating intellectual environments.

Ian James Kidd (2019b; Reference Kidd, Kidd, Battaly and Cassam2020) introduces the concept of epistemic corruption to draw attention to the way in which our intellectual environments can encourage the development and exercise of intellectual vices, or lead to the loss of intellectual virtues. Those environments which fail to nurture intellectual virtues are what Kidd calls ‘passively corrupting’, while environments that encourage the exercise of intellectual vices are ‘actively corrupting’. Importantly, this distinction is not clear-cut and the two will typically reinforce each other in a mutual fashion. For example, suppose you attend an elite debating club that encourages and rewards ‘kill or be killed’ argumentative styles. Within that environment, arrogant or dismissive tendencies are likely to become entrenched in one's intellectual character. Simultaneously, the longer one is exposed to such conditions and styles, the harder it will become to cultivate humble or attentive behaviour and traits towards others. Of course, not all environments are equally corrupting.Footnote ⁴ Some might contain structures that are conducive to the development of many vices or lead to the loss of numerous virtues. At other times, the environment might only feed a handful of vices in a small number of domains. My participation in a football fan club might cause me to develop a close-mindedness towards other football clubs but leave my political or music tastes wholly unscathed.

How might these environments contribute to the corruption of intellectual character? Kidd says that epistemic corruption can occur in multiple ways. A particular environment might contain conditions that cause one to acquire certain traits or habits that one didn't previously possess. Only after joining and engaging in the activities of the football fan club, for example, might I realise that I have acquired an intellectual vice – assuming I'm even in a position to recognise it in the first place! (Cassam, Reference Cassam2019). Other environments might draw out and ignite one's previously managed intellectual vice(s). Suppose that you recognise your arrogant tendencies but nonetheless keep them contained. By participating in the debating club, the norms, rules, and structures present could unlock your arrogance and allow it to flourish. Furthermore, the same environments could well amplify, stabilise, or intensify particular intellectual vices, thereby corrupting one's intellectual character from many different angles (Kidd, Reference Kidd, Kidd, Battaly and Cassam2020, p. 72). In short, epistemic corruption is a complex phenomenon with real implications for our character and conduct.

3. Intellectual Cynics

Using the concept of epistemic corruption, I now want to argue that deepfakes encourage us to develop the epistemic vice of intellectual cynicism. To see just how deepfakes might put us in this position, it will be useful to consider Rini and Fallis’ concerns in more detail. According to Fallis (Reference Fallis2020), deepfakes are epistemically harmful because they risk undermining the information videos carry to viewers. On Fallis’ view, a video carries information if there is a low probability of watching a false positive, i.e., a video that looks and sounds just like a genuine video but turns out to be fake. If this probability is low, a video will accordingly carry more information. However, Fallis worries that if deepfakes become sufficiently sophisticated and wide-spread, they will increase the incidence rate of false positives. As such, the information depicted in a genuine video will be undercut by the deepfake, increasing the likelihood of us suspending judgement about a video or simply accumulating false beliefs.

For Rini, meanwhile, the worry with deepfakes is that they will cause us to ‘reflexively distrust all recordings’ (Reference Rini2020, p. 8, my emphasis). This starts with the concept of an ‘epistemic backstop’, the idea that videos acutely correct other people's testimony and passively regulate what people say in public. After all, the possibility of being recorded tends to encourage people to be sincere and competent testifiers. However, Rini worries that as deepfakes become more sophisticated and widespread, they will lead to a sense of ‘displaced epistemic reality’ (Reference Rini2020, p. 8), where the very possibility of an authentic video being a deepfake lingers in people's minds. In turn, deepfakes could undercut the corrective function of videos by eroding their perceived evidential status. If deepfakes make it more difficult to ascertain the veracity of a recording, Rini believes that people will have less incentive to testify sincerely and competently if they can quite easily ‘cry deepfake’ on a video.Footnote ⁵ As a result, people will plausibly invest less trust in what they hear or see from videos, and thus videos will lose the ability to passively regulate what people say in public. For these reasons, Rini believes that deepfakes will usher in ‘backstop crises’ and erode the epistemic credentials of all recordings (Reference Rini2020, p. 8).

While we needn't endorse the full extent of Fallis and Rini's arguments here, I do want to press the idea that deepfakes could plausibly lead us to increasingly distrust videos or cause us to suspend judgement about what a given video depicts. In particular, I want to draw parallels with this behaviour and the distrust and disengagement that is central to what Samantha Vice (Reference Vice2011) calls moral cynicism. What makes cynicism specifically ‘moral’ is that it primarily concerns our attitudes and beliefs towards other human beings and ourselves. For Vice (Reference Vice2011, p. 173), moral cynicism can be thought of as a habitual and cultivated attitude of ‘scepticism or distrust regarding people's professed values and motivations’ or their ‘strength and nobility of character’. Underpinning this distrust is the cynic's core belief that humans are ultimately self-interested or of little worth. In this way, cynicism ‘structures our perception, interpretation, evaluation, and expectations of others’, while also affecting our own actions. At its darkest and broadest, Vice claims, cynicism can even lead one to disengage from humanity and its institutions altogether, due to a habitual wariness of what drives people (Reference Vice2011, p. 172). Those who adopt this stance are what Vice calls ‘pure cynics’, someone whose ‘pervading and fundamental character is cynical’ (Reference Vice2011, p. 174).

If deepfakes are able to spark distrust towards online videos, thereby prompting us to suspend judgement more readily about their depictions, then we can begin to see a kind of cynicism on the horizon. This kind of cynicism, I contend, exists presently but will be exacerbated by deepfakes. In particular, it concerns our behaviour towards intellectual or epistemic matters as opposed to people and human relations. Call this intellectual cynicism. Much like its moral counterpart, we can think intellectual cynicism as a negative stance or cluster of attitudes towards a particular epistemic object. Here, the object in question needn't necessarily be epistemic goods such as truth, knowledge, or understanding, but rather the source of these goods themselves. In the past, this source has typically been directed at testimony. Think of those, for example, who adopt negative attitudes towards Facebook's recent claims and pledges to improve online safety. What makes these cynical attitudes ‘intellectual’ as opposed to ‘moral’, I suggest, is that they are grounded in a belief about the epistemic status of Facebook's pledges and claims qua propositions, that it whether their statements are true or false, rather than a belief about Facebook as a corporation or agent itself.Footnote ⁶ Adopting this kind of moderate cynicism might therefore be warranted, so long as one is receptive to evidence to the contrary.

Given this, how does intellectual cynicism factor into the spread of deepfakes? As the world has ‘moved’ online during and after the Covid-19 pandemic, the source of many of our epistemic goods has increasingly shifted to videos. We watch, listen, and rely on videos for work, education, and information. However, suppose at some point you find out that they've recently spoken to a deepfake or been tricked into believing a deepfake's content, or worse. As deepfakes grow more sophisticated, a natural reaction might be to adopt a cynical attitude towards online, unregulated videos. At first blush, this move would bear similarity to the cynical attitudes adopted towards Facebook's safety pledge. Yet in both cases, I want to suggest that there is a problematic element to this. Just as Vice carves off ‘pure cynicism’ as a particularly acute kind of moral cynicism, so too can we draw attention to a similarly damaging kind of intellectual cynicism. This involves distinguishing intellectually cynical attitudes from the trait of intellectual cynicism. As deepfakes become more sophisticated, it's plausible to think that people will come to adopt cynical attitudes towards unregulated, online videos more and more, and through doing so have these attitudes congeal in their intellectual character. Why is this a bad thing? Well, take one familiar feature of character traits: they necessarily involve being disposed to think, feel, and act in particular ways. Those who cultivate the trait of intellectual cynicism, I contend, will come to habitually view videos in a negative light, which will in turn generate corresponding habits of thought and action towards them. In particular, the trait will dispose its possessors to make inferential leaps that all unregulated, digital, audio-visual information online is potentially deepfaked.Footnote ⁷

How can we make sense of this inferential leap? In habituating certain traits, people also develop corresponding sensibilities. One way to think about sensibilities is in terms of one's characteristic patterns of seeing the world. For instance, a distinct feature of those who cultivate the trait of attentiveness is that they are more likely to notice certain behaviours, actions, or people than those without this trait. A similar story holds for the trait of intellectual cynicism. In fact, Vice alludes to this when she observes that moral cynicism ‘structures our perception, interpretation, evaluation, and expectation of others’ (Reference Vice2011, p. 172). By the same token, the trait of intellectual cynicism will manifest cynical sensibilities that infect one's intellectual character at large, leaving one with a ‘‘corrupt sense of value’ (Vice, Reference Vice2011, p. 175). Specifically, those who habituate this trait will be driven by a belief in the sub-optimality of online, unregulated videos as a source of epistemic goods as a whole, much in the same that the moral cynic views humanity. Consequently, the intellectual cynic is unable to attend to the possibility that such content possibly aims at the truth or other epistemic goods. Instead, their sensibilities are too highly attuned to the foregone conclusion that these videos never serve intrinsically good purposes or are instrumental to the success of a dubious agenda. To put this point another way, we can think of the intellectual cynic as somebody who habitually fails to consider the intellectual merits of online, unregulated videos. In this way, the trait of intellectual cynicism is self-sufficient: the more one habituates it, the more pronounced the accompanying sensibilities are at structuring what the intellectual cynic sees and evaluates – or perhaps more apt, what they fail to see and misevaluate. It is for this reason that the intellectual cynic makes such large inferential leaps.

If this is the case, then we can say that deepfakes are actively corrupting: they risk encouraging consumers of digital media to develop the epistemic vice of intellectual cynicism. We can make good on this claim by reference to the three accounts of epistemic vice canvassed above. First, those who habituate the trait intellectual cynicism might be said to possess bad motivations towards wanting knowledge if they are disposed to distrust and associate videos with suspicion. Second, in cultivating this trait, the intellectual cynic obstructs the transmission of knowledge and information. Third, by developing the cynical sensibilities associated with this trait, the intellectual cynic makes poorly judged decisions about the merit of videos as a source of epistemic goods because they perceive them in such a skewed light. These three manifestations will, of course, overlap and be borne out more strongly in some than others. Regardless, though, by habitually distrusting and disengaging from epistemic practices, the intellectual cynic never engages with the very sources and practices that could potentially shake their cynicism. If this is the case, deepfakes have the potential to encourage an ultimately damaging intellectual character trait that deprives us of the knowledge and information needed for a flourishing intellectual life. It's beyond the scope of this paper to give a full account of the nature of intellectual cynicism, but I hope it's at least clear what both the attitude and trait consists of, and how deepfakes might encourage this intellectual vice.

4. Cultivating Digital Sensibility

In the previous section, I argued that deepfakes are epistemically corrupting because they risk encouraging consumers of digital media to develop the vice of intellectual cynicism. However, as Kidd rightly emphasises, identifying corrupting environments – or in our case technologies – enables us to work out what sorts of corrective measures to put in place (Reference Kidd2019, p. 227). The corrective approach I want to advance is somewhat analogous to our everyday treatment of the testimony we hear from informants. The motivation for this comes from Rini's worry above. Part of the reason why videos play such an important role in her idea of an ‘epistemic backstop’ is because of their justificatory or evidential status. It is widely acknowledged that videos and photographs provide us with perceptual evidence of states of affairs (Cavedon-Taylor, Reference Cavedon-Taylor2013; Hopkins, Reference Hopkins2012), which is more authoritative than, say, testimonial evidence. Rini worries that by eroding our trust in videos, deepfakes will relegate the justificatory status we assign to recordings from perceptual evidence to that of testimonial evidence. In other words, we will take videos to be less authoritative of the states of affairs they depict.

If Rini's concerns are correct, then why don't we treat videos like testimony? So, just as we deal with liars and bullshitters (Frankfurt, Reference Frankfurt2005) by exercising particular caution, hesitance, and critical evaluation in our credibility assessments of what they say, I suggest that we deal with the threat of deepfakes and intellectual cynicism by cultivating a countervailing sensibility, what I will call a digital sensibility. The basis of this sensibility is rooted in Miranda Fricker's (Reference Fricker2007) idea of a testimonial sensibility. This a carefully cultivated way of experiencing and evaluating other people's testimonies, informed by an awareness that our ability to properly listen to others is corrupted by all sorts of biases and prejudices. It involves training oneself and gaining more understanding of social prejudices, which is no mean feat. By cultivating this sensibility, one comes to gradually acknowledge and, to an extent, nullify the prejudices that interfere with one's credibility assessments of others (Reference Fricker2007, pp. 72–81). For Fricker, there are two components of this sensibility that stand out for our discussion: its perceptual nature and its affective dimension.Footnote ⁸ I will look briefly at each as a way of modelling the digital sensibility required for mitigating the intellectual cynicism promoted by deepfakes.

Fricker's starting point is an idea dating back to Aristotle, for whom the good or virtuous person is one whose ‘moral perception’ is drawn towards ethically salient aspects of a situation. This perception is carefully cultivated and trained to be reflexive and spontaneous, so that upon being confronted with a morally charged scenario, the good person comes to ‘just see’ the situation in a certain light and is able to pick out the best course of action to help. Fricker suggests that we do something similar with what she calls an ‘epistemic perception’. Similarly, this is a calibrated sensitivity to salient aspects of a speaker's situation and is informed by a number of background assumptions about the speaker's trustworthiness, including cues relating to their sincerity and competence on the matter at hand (Reference Fricker2007, p. 72). Importantly, the work of calibration should also be an interpersonal project, rather than an individual one, though obvious care should be taken in deciding whom to trust – something that gradually improves with practice. With the help of a conducive intellectual environment – one that encourages and emphasises the virtues of the mind – people can develop an ability to see the world in a certain light, enabling them to assess their interlocutors in a fair, open-minded, and non-inferential manner.

This perceptual sensibility is also supported by an affective dimension. In order to properly judge the competency and sincerity of a speaker, one needs to be able to sufficiently empathise with them. Only by invoking a degree of emotional engagement with a speaker can one fully appreciate the socially situated factors that might be in play when it comes to ‘reading’ another's body language or the competence one displays in their testimony (Reference Fricker2007, pp. 79–80). In order to better determine the level of trust one ought to place in another's word, Fricker observes that good advice is ‘to listen to one's emotions’ insofar as the virtuous person's emotional response to different speakers across different contexts is only improved by training and experience (Reference Fricker2007, p. 80). By exposing ourselves to different speakers in different contexts, we can build up a sophisticated emotional radar for detecting trustworthiness in speakers. In doing so, we not only cultivate a trained testimonial sensibility, but we also manifest the virtue of ‘testimonial justice’ – a disposition to reliably neutralise prejudice and bias from our credibility assessment of speakers (Reference Fricker2007, p. 92).

Taking these points into consideration, we can develop a model for digital sensibility. For starters, this will be a reflective and spontaneous sensitivity towards online content, so that when we're confronted with certain videos, images, or website we learn to ‘just see’ them as trustworthy. Again, this will turn on features like the content's sincerity and competence. For instance, it is common nowadays to stumble across an online video or see adverts on websites and immediately withhold judgement about the claims made, or outright flag them as deceptive. One reason for this hesitance is that the content seems insincere: it makes claims that do not accord with the background assumptions against which we ordinarily evaluate videos or other media. Another reason is that the content appears incompetent: its presentation and configuration fail to conform to established norms of publishing and communication. A further factor in our assessments of trustworthiness might also include our worries about clicking on videos and developing computer viruses, being scammed, or losing personal information.

This last factor draws out the affective dimension of a digital sensibility. While Fricker takes empathy to be an important emotion in judging the sincerity of an interlocutor, I think that past, present, and future experience with our digital environments will require the kind of ‘healthy scepticism’ that, as Pierre Le Morvan (Reference Le Morvan2011, p. 98; Reference Le Morvan2019) puts it, ‘guards’ against both gullibility and close-mindedness’. Fine-tuning this, he claims, bolsters our ‘doxastic immune system’ from the threat of false beliefs (Reference Le Morvan2011, p. 91). Successfully exercising this healthy scepticism will partly rely on us listening to our ‘gut feelings’ when they flag a video as unscrupulous or dubious, the effectiveness of which will come from prior experience and exposure to a range of claims, depictions, and sources of digital content. An important part of this fine-tuning, then, will be the perseverance to fact-check such content, and a degree of humility in recognising that our credibility assessments will not always be watertight. Over time, this healthy scepticism will inform our emotional radar and help us to judge certain content as more trustworthy – established news media and websites that adhere to communicative norms – than other content which fail to adhere to such standards or are at odds with the catalogue of past experiences we have built up. In this way, we continually fine-tune the ‘radar’ that comes with cultivating a digital sensibility. Eventually, a trained digital sensibility, like its testimonial counterpart, will develop to be an active, automatic, and critical alertness towards online content.

Crucially, though, the development of a digital sensibility is as much about individual caution and critical awareness as it is about developing a collective safeguard against deceptive online media. As Fricker makes clear, the formation of our testimonial sensibilities is partly individual and partly collective (Reference Fricker2007, p. 82). Accordingly, the cultivation, exercise, and regulation of a trained digital sensibility relies on cooperating with our peers across digital environments. Those who are more technologically savvy will have duties to flag potentially untrustworthy content, allowing those less well-versed in digital matters to observe how to decide how much trust to place in websites and online media. Gradually, those with less experience can learn from the ‘digital exemplars’ in their community and slowly fine-tune their perceptual and emotional radar towards trustworthy online content. This collective dimension not only illustrates the educational role that training one's digital sensibility entails, but it also emphasises the socially-scaffolded nature on which the effectiveness of this sensitivity is premised. This is particularly salient when it comes to regulating the affective dimension of one's digital sensibility. It's not hard to imagine somebody who exercises a deficiency or excess of scepticism towards digital content, perhaps sympathising with the claims made by a video or website or being overly sceptical of authentic content. To ensure that one's healthy scepticism is kept in check, one's peers will need to hold them to account, warning them when their emotional radar fails to track the merits or demerits of digital content.

However, even a collective effort to cultivate a trained digital sensibility might not be sufficient to fend off the cynicism that deepfakes and emerging technologies might encourage. In addition to individual and collective practice and cultivation, a well-trained digital sensibility requires institutional and structural scaffolding (Anderson, Reference Anderson2012; Medina, Reference Medina2013). This will require social media companies and news corporations to cultivate the virtue of collective responsibility (Astola, Reference Astola2022). In our case, this responsibility will manifest in the companies flagging deepfakes, correcting mis- and disinformation, and raising awareness of other untrustworthy sources of information. For a long time, Facebook allowed users to post ‘parody’ deepfakes on its website, which could be exploited by malicious actors to indirectly spread misinformation and sow distrust (Shead, Reference Shead2020). If efforts to cultivate a trained digital sensibility are to succeed in the long-term, institutions with regulatory powers like Facebook, Google, Microsoft, and Twitter will need to work in tandem with their user-base. Only then, will a trained digital sensibility be sufficiently socially scaffolded and effective.

5. From Digital Sensibility to Techno-Intellectual Virtue

In a world increasingly characterised by deepfakes, fake-news, and emerging technologies, how can we best flourish? I want to briefly conclude by reflecting on this question. According to the philosopher, Shannon Vallor (Reference Vallor2016), the cultivation of what she calls techno-moral virtues will help us achieve this end. Techno-moral virtues are an extension of our existing moral capacities or traits, adapted to a rapidly changing techno-socially opaque world, which require a collective moral wisdom on a global scale (Reference Vallor2016, p. 10). Some notable examples include techno-moral courage – ‘a reliable disposition toward intelligent fear and hope with respect to moral and material dangers and opportunities presented by emerging technologies’ (Reference Vallor2016, p. 131), and humility – ‘a renunciation of the blind faith that new technologies inevitably lead to human mastery and control of our environment’ (ibid, pp. 126–27). In the face of emerging technologies, Vallor rightly stresses the need to cultivate these virtues as a means of offsetting any potential harm to our ethical character.

While Vallor is certainly right to draw attention to this prospect, my aim in this paper has been to demonstrate just how deepfakes can degrade our intellectual character by encouraging us to develop the vice of intellectual cynicism. In light of this, I proposed a way to fortify our intellectual character against the potentially corrupting nature of deepfakes. I believe that developing a well-trained digital sensibility will be a major step towards doing this. This will not simply lead those wishing to navigate online spaces to acquire a number of ‘virtues of inquiry’ towards digital content, but to specifically cultivate the techno-intellectual virtue of digital attentiveness, a corrective disposition to carefully attend to the merits and demerits of digital media. Just as cultivating a trained testimonial sensibility helps one nullify prejudices and biases against speakers, a trained digital sensibility enables one to offset the vice of intellectual cynicism and fortify one's intellectual character described above. For those wishing to flourish in a rapidly changing world, a trained digital sensibility and the correlative techno-intellectual virtue of attentiveness will help to make this a reality.Footnote ⁹