In 2019, a BBC drama entitled ‘The Capture’ premiered. It was the type of complex political thriller the Beeb delights in, involving high-level conspiracy, international espionage and, crucially, the deployment of ‘deepfakes’: hyper-realistic audio or visual media created using deep learning technology. It was generally well received, though some critics caveated their praise by noting the far-fetched nature of the tech-driven plot.
Five years hence, ‘The Capture’ seems decidedly less fantastical. Last year, both the Leader of the Opposition and the Mayor of London were subjects of deepfake audio clips that sought to defame their characters and inflame political tensions. Overseas, election campaigns in India, Argentina and Taiwan have experienced the increasing use of deepfake technology by both official campaigns and anonymous online sources. Most notably, last October’s Slovakian election saw the emergence of a deepfake audio of Michal Simecka, the election’s frontrunner, conspiring to rig the ballot, which occurred in the days leading up to his eventual and surprising narrow defeat. While these examples were rudimentary enough to be identifiable by authorities and tech platforms, there is concern that the increasing sophistication of AI tools will render identification much more difficult, if not impossible.
These events are widely seen as forerunners to a future where deepfake technology will have a larger impact on democracy. Here in the UK, there is a growing acceptance that deepfakes pose a national security threat, insofar as they enable external and internal actors to manipulate the democratic process, facilitate political polarization and undermine citizens’ trust in the state. There is a pressing need to develop a comprehensive strategy to address this challenge.
The Infopocalypse? The Threat of Deepfakes to Democracy
Deepfake technology has emerged at a critical juncture for electoral democracies globally. The sluggish economic performance of many established democracies, not least the UK, has fermented significant disillusionment with democratic institutions. Increasing numbers of populist movements, which valorise common sense and reject technocratic ‘elitism’, have rallied criticism against traditional informational gatekeepers – the mainstream media, the academy and established political parties. In turn, the ubiquity of social media platforms as a means of communication has democratised the process of information exchange, further weakening gatekeeper institutions, and creating new avenues for mis- and disinformation.
Professor Graham Meikle explains that deepfakes land in this environment like a ‘dirty bomb’ and have the potential to weaken the already precarious situation of many democracies globally. Democratic self-government requires populations to be able to collectively deliberate on political issues and make informed electoral choices based on their preferences. Liberal democracies also rely on robust human rights protection regimes and the ability of the judiciary to effectively determine facts and identify whether laws have been contravened. All three of these key democratic pillars – democracy, human rights and the rule of law – are threatened by the widespread deployment of deepfake technology.
It is not difficult, then, to understand the challenge of deepfakes to electoral democracy. If anyone with access to a computer and the internet can create fabricated audio clips, videos and images (using popular applications such as HeyGen, Synathaeasia or DeepSwap), and share them widely across social media, then there is a very real risk that our information ecosystem becomes degraded. Added to this is evidence of foreign interference in British politics, and the identifiable doctrine of enemy states (most notably Russia) to wage information warfare designed to accentuate social polarisation within their adversaries. In this context, there is good reason to expect deepfakes to form part of disinformation campaigns in the upcoming election.
Unfortunately, the media outlets responsible for accurately reporting political events may also be hamstrung by uncertainty over the provenance of audio and video clips, which are crucial to effective reporting. The proliferation of deepfakes also creates a dynamic which scholars have termed the ‘liar’s dividend’, in which political figures can deny accountability for actions depicted in voice recordings and videos by claiming they are synthetic (this tactic has already been deployed by high-profile figures such as Donald Trump and Elon Musk). All told, there is widespread concern that deepfakes could fundamentally diminish the digital public sphere, leading to what one analyst describes as an ‘infopocalypse’, in which effective democratic deliberation becomes impossible without shared facts.
The threat of deepfakes also extends to other democratic pillars. Legal experts are concerned about the potential of synthetic media to problematise the judiciary’s fact-finding mission and make litigation more costly, prolonged and contested. And there is undoubted potential for human rights abuses. While existing literature on deepfakes in the UK has correctly framed their deployment in national security terms, there exists the possibility that the state itself could deploy these technologies with the view to discriminate against or disempower individuals.
Cats and Mice: the Challenges of Defending Against Deepfakes
These democratic threats are difficult to defend against due to two interrelated ‘cat and mouse’ dynamics. The first pertains to the creation of deepfakes. Like other forms of cyber warfare, there’s a deepfake ‘arms race’ between offensive and defensive technology. As new technology for identifying deepfakes is developed, deepfake technology will adapt to become more realistic and harder to flag. One scholar has argued that, such is the pace of development of AI technology and the unreliability of existing detection mechanisms, those developing software to defend against deepfakes have to ‘run as fast as [they] can to stay in the same place’.
The second ‘cat and mouse’ scenario relates to the proliferation of deepfakes. The political economy of social media platforms, fuelled by algorithms designed to hook users, means that harmful deepfakes that contain volatile or controversial content are likely to travel at speed. Twitter initially refused to remove the aforementioned clip of Kier Starmer due to its inability to determine its provenance and it subsequently garnered over 1.5 million listens in the time it was online. This episode demonstrates that, even when deepfakes are crude enough to be eventually identified by software, they may still fool social media users and spread widely. Indeed, one recent study estimated that people could only identify 73% of deepfake audio, a number that will likely decrease as technology develops.
These dynamics mean there is a tentative consensus amongst experts that deepfakes are an ‘offence-dominant’ technology, which means that they are harder to defend against than they are to use offensively. However, offence-dominance may operate at different levels. While deployers of deepfake technology may have a tactical advantage, it is uncertain whether deepfakes can provide enemies of the UK’s democracy with an operational or strategic advantage. In other words, whilst it is clear that an actor could create confusion by deploying one or more deepfakes, it remains unclear whether deepfakes could be used to shape an electoral contest or generate fundamental instability within the UK polity.
Defending Democracy
We should be wary of claims that either the ‘infopocalypse’ is nigh and claims that we are an ‘information revolution species’ that will inevitably adapt to the advent of new technologies. There is nothing pre-ordained about the societal impact of deepfakes, much of which is contingent on policy responses by governments, tech companies and civil society.
Broadly speaking, there are two approaches to tackling the corrosive potential of deepfakes: techno-legal measures that seek to limit the supply of deepfake disinformation, and measures that seek to understand and eradicate societal demand for the type of malicious and conspiratorial content that deepfakes often carry. A national strategy to tackle the threat will inevitably require both.
Many of the supply-side solutions to tackling the deepfake threat are predicated on the development of technology that will allow for the rapid and accurate identification of synthetic media. The private sector is awash with promising AI-driven detection tools, including Intel’s high-profile ‘FakeCatcher’. Experts have also proposed the development of innovations such as ‘deepfake zoos’, depositories of known deepfakes that can be used for shared training of detection tools. Similarly, digital watermarks can now create ‘invisible labels…embedded in content before it is distributed’. The technological landscape on deepfake detection is fast moving but, given the dynamics described above, experts remain unsure whether rapid and reliable detection mechanisms are plausible solutions.
The development of detection and labelling tools opens the door for legislation that polices the creation and proliferation of deepfakes. The Government has shown a desire to legislate in this space – the Online Safety Act makes the sharing of deepfake pornography an offence and gives Ofcom the power to fine tech companies who don’t abide by their terms of service. Some commentators believe that legislative approaches must go further, with mandatory labelling of deepfakes and outright bans on the use of deepfakes for political campaigning examples of legislation that could deter their spread. However, enforcing these proposed laws relies on the development of consistent detection tools.
In any case, the development of techno-legal approaches to tackle deepfake disinformation is a long-term project, whose threat exists in the immediate term. This is why the development of demand-side policies is also important. The most commonly cited of these is the amplification of media literacy campaigns, which several experts across the academic, policy and private sectors cited as crucial in a recent parliamentary inquiry on disinformation. Many also see a heightened role for fact-checking organisations, such as FullFact and BBC Reality Check, which are currently swamped by the scale of their tasks. Another commonly cited policy is strengthening the conventional media ecosystem, especially the UK’s anaemic local journalism landscape. The idea behind such policies is that a more tech-savvy population with better media options will be less susceptible to political deepfakes.
However, such an approach only works insofar as people believe themselves to be media illiterate. There is a portion of the population which chooses to engage with conspiratorial deepfakes not out of ignorance but out of a sense of alienation and disillusionment with conventional politics. Understanding why such content is attractive to people across the country is a nebulous task but it is crucial that policymakers consider the social dynamics that determine the impact of political deepfakes.
Reasons for Optimism
Analysts are correct to point out that deepfakes have yet to truly bite democracy and that we should not ignore the wider disinformation woods for the deepfake trees. However, such are the democratic stakes of the widespread proliferation of deepfakes, it is only right that a forward-looking strategy is developed.
There is evidence that such a strategy is emerging. This week, the Conservatives announced a ban on the creation of deepfake pornography, a policy previously endorsed by Labour. Whilst the enforceability of such a policy is questionable, the existence of cross-party agreement on this issue is welcome. Moreover, there is widespread reporting on deepfakes across national media, significant private sector engagement in developing solutions, and a cross-departmental ‘Defending Democracy Taskforce’ that covers this area. This activity should engender some optimism that, when deepfakes inevitably become a more prevalent part of our democratic discourse, we have the policy tools in place to mitigate their damaging effects.
Steph Coulter.
Steph is a researcher at the Bennett Institute for Public Policy and a contributing writer for the Constitution Society.
The Constitution Society is committed to the promotion of informed debate and is politically impartial. Any views expressed in this article are the personal views of the author and not those of The Constitution Society.