AI Gone Rogue - The Dangers Of Deepfake Technology In The Digital Age
Author: LawSphereX
Category: Technology, Cyber & Privacy
Published:
Updated:
AI GONE ROGUE: THE DANGERS OF DEEPFAKE
TECHNOLOGY IN THE DIGITAL AGE
Vikas Kabeer, LLM, MVN University, Palwal
ABSTRACT
Deepfake technology, which combines artificial intelligence (AI) and machine learning is an emerging technology that allows the creation of hyper realistic, fabricated audio, video and image content in the digital age. Deepfakes have the ability to provide creative opportunities in the entertainment and virtual reality fields, but they also pose enormous threats to political stability, personal privacy and economic security. This paper examines the perils of deepfake technology, such as privacy invasion, manipulation and economic fraud, and explores the evolution and mechanics of this technology in depth. The paper also conducts a comprehensive analysis of current and proposed regulatory responses that are intended to moderate these risks.
It also tackles the forthright issues of biased algorithm and accountability in terms of the social effect of biased deepfakes and ethic responsibilities of the AI creator. In an effort to contribute to the debate on how to mitigate the risks of AI driven media manipulation this paper, investigates the multifaceted risks of deepfakes and proposes ways to regulate and publicize about deepfake technology. Keywords: Deepfake Technology, Artificial Intelligence (AI), AI Bias and Accountability, Digital Privacy.
1. Introduction
Artificial intelligence (AI) and machine learning advancements have invented something new and profoundly changed [the way] digital media functions — the art of the deepfake. Deepfakes have originally been developed for entertainment and creative industries, but have since exploded, sparking an alarm over their potential misuse. 1 With such precision, the ability to manipulate media content also poses unprecedented risks, with respect to politics, personal privacy and economic security. First coined in 2017 is the term "deepfake" which combines "deep learning" and "fake" to indicate the underlining AI algorithms that are used to create these great fake fabrications.
2 The positive side of this technology, such as in film production or virtual reality, should go unnoticed, but the negative side of deepfakes – especially the usage of Deepfakes for propaganda and abuse – received so much attention. For instance, political deepfakes can be used to disrupt democratic processes through spreading misinformation and propaganda. 3 What’s more, the risks with deepfakes for personal privacy are scary. There have been huge cases of cases of non-consenting pornography, identity theft, and online harassment, which caused huge psychological and reputational damages to individuals. 4 The challenge of detecting and mitigating the growing sophistication of deepfakes is increasingly urgent, fueled by the increasingly sophisticated nature of the technology.
To address the dangers of deepfake technology a multi-pronged approach is needed that includes legal, technological, and educational approaches. In this paper we discuss the evolution and mechanics of deepfake technology, the various harms it depicts, and current and proposed regulations. Furthermore, it investigates the problems of bias and responsibility of deepfake technology, and calls for moral guidelines and civil awareness on this high-tech instrument.
2. The Rise and Perils of Deepfake Technology
The name, deepfake technology, refers to the confluence of the terms 'deep learning,' and 'fake,'
1 Chesney R and Citron DK, 'Deepfakes and the New Disinformation War
The Coming Age of Post-Truth Geopolitics' (2019) 98(1) Foreign Affairs 147 https://www.foreignaffairs.com/articles/world/2018-12- 11/deepfakes-and-new-disinformation-war accessed 10 November 2024.
2 Goodman B, 'Deepfake Threats to Democratic Elections
Digital Watermarking as a Defensive Measure' (2020) 5(1) Journal of Cyber Policy 55 https://www.tandfonline.com/doi/full/10.1080/23738871.2020.1722992 accessed 10 November 2024.
3 Fallis D, 'The Epistemic Threat of Deepfakes' (2020) 33 Philosophy & Technology 239
https://link. springer. com/article/10. 1007/s13347-019-00345-6. 4 Kietzmann J, Lee L, McCarthy IP and Kietzmann TC, 'Deepfakes: Trick or Treat? ' (2020) 63(2) Business Horizons 135 https://www. sciencedirect. com/science/article/pii/S000768131930182X accessed 10 November 2024. referring to its artificial nature of synthesizing realistic media. The technology gained prominence in 2017 with the advent of Generative Adversarial Networks (GANs), which consist of two neural networks: It consists of a generator that behaves as an engine that spits out fake content, and a discriminator that judges the authenticity of the content that is produced. The generator improves its output through iterative training, making deepfakes more difficult to detect, which keeps being more and more difficult to detect. 5 Deepfake technology first came about through computer graphics and AI.
Earlier works in the field of AI were concerned with creating realistic images and videos, but GANs, which were invented only recently, have turned everything upside down since they make it possible to generate very realistic counterfeit content. And because deepfakes have begun to transition from academic research to practical applications, that’s raised ethical and legal concerns. 6 It’s one of the most pressing concerns around deepfakes — there’s the potential for deepfakes to spread misinformation. Deepfakes can take advantage of video or audio recordings, creating entirely fabricated scenarios, which will then be published as false information. Deepfakes are not good in political settings, where you can use deepfakes to impersonate political figures, or fake speeches or actions from those politicians.
Indeed, deepfakes could be weaponized, for example, to discredit candidates, bend public perception, or trigger interference with democratic processes, during election cycles. 7 For a deepfake video, if in one infamous case, it could be used to broadcast fake statements or actions of a political leader with the intention of sowing doubt or misinformation amongst the population until they sway public opinion to the point of social unrest, or even conflict. For instance, in 2018 deepfake of U. S. former President Barack Obama had risen, making saws of deepfakes do deceiving activities to visitors. Political deepfakes have become a big issue as they can influence elections and popular opinion. One example: during 2024 U. S.
presidential primary, a robo call, put out as one from President Joe Biden, told people to hold onto their vote for the coming general election and in doing so, misled many. Likewise in the 2023 Nigerian elections, an audio clip, allegedly
5 Goodman B, 'Deepfake Threats to Democratic Elections
Digital Watermarking as a Defensive Measure' (2020) 5(1) Journal of Cyber Policy 55 https://www.tandfonline.com/doi/full/10.1080/23738871.2020.1722992 accessed 10 November 2024.
6 Chesney R and Citron DK, 'Deepfakes and the New Disinformation War
The Coming Age of Post-Truth Geopolitics' (2019) 98(1) Foreign Affairs 147 https://www.foreignaffairs.com/articles/world/2018-12- 11/deepfakes-and-new-disinformation-war accessed 10 November 2024.
7 West DM, 'The Case for Regulating Deepfakes' (2019) Brookings Institution
https://www. brookings. edu/research/the-case-for-regulating-deepfakes accessed 11 November 2024. manipulated by AI, has erroneously linked a presidential candidate to ballot manipulation. 8 They illustrate the deepfakes becoming a more ominous thorn in the side of democratic processes. In recent years, celebrity deepfakes have also been making headlines; media taking advantage of the deepfakes for entertainment, while others used them maliciously. For example, there is the deepfake videos of Tom Cruise on TikTok, which a Belgian digital artist and an impersonator virally spread—and was real. The next would-be David Beckham who, to promote a malaria awareness campaign used deepfake technology to deliver a message in nine different languages.
9 Some uses of celebrity likeness are benign or even beneficial, but otherwise the abuse of celebrity likeness can ruin reputations and engender legal battles. Financal scams have been exploited with deepfakes as well, a serious problem for individuals and institutions. Deepfake technology that uses artificial intelligence to create realistic videos or audio clips from human faces, like that of CEOs or other executives, has been used by fraudsters to force employees to transfer large amounts of money. 10 Earlier this month, the Reserve Bank of India warned its people to not fall for deepfake videos of its Governor endorsing fraudulent investment schemes, using such videos to confuse the public into making substandard financial decisions. 11 But these scams serve to highlight the need for advanced detection tools and public, now more than ever, awareness to combat deepfake related fraud.
This also poses a major threat for personal privacy. Digital video and sound of sufficient quantity helps you to make you virtually into videos you never had. It's been used in what are known as 'revenge porn', or 'harassment' cases where the likenesses of individuals are used to create explicit or defamatory content without their consent. 12 Deepfakes could be used for identity theft too, where an individual's 'lookalike' is used to pretend as that person in digital interactions disposing of their privacy and personal security. There have been falling efforts to address deepfakes dangers at various fronts, technological, regulatory, and educational. To 8 Mirza R, 'How AI Deepfakes Threaten the 2024 Elections' The Journalist's Resource (16 February 2024) https://journalistsresource. org/home/how-ai-deepfakes-threaten-the-2024-elections/ accessed 13 November 2024.
9 'Artificial Intelligence: Deepfakes in the Entertainment Industry' (n.d.) WIPO
https://www.wipo.int/wipo_magazine/en/2022/02/article_0003.html accessed 12 November 2024. 10 Kaushal TJ, 'Deepfake Scams and Financial Frauds: How to Protect Yourself?' Business Today (10 November 2024) https://www.businesstoday.in/personal-finance/news/story/deepfake-scams-and-financial-frauds-how-toprotect-yourself-453171-2024-11-10 accessed 12 November 2024.
11 Dube M, 'RBI Warns Against Deepfakes Showing Governor Das Endorsing Investment Schemes' NewsBytes
(19 November 2024) https://www. newsbytesapp. com/news/business/rbi-warns-against-deepfake-videospromoting-fraudulent-investment-schemes/story accessed 12 November 2024. 12 Franks MA, 'Deepfakes and the Law: The Ethics of Synthetic Media' (2019) 94(4) Indiana Law Journal 1057. detect deepfake content, AI researchers have been developing algorithms based on inconsistences — uneven blinking, nonnatural facial movements, and so on. 13 But as deepfake technology advances, so do the countermeasures. That means what makes a good defense today may be incomplete or even worse tomorrow, so that countermeasures are always behind.
To create deepfakes you will first need to train a network to identify and generate face features, generate deepfake content by putting a target’s face on a new body or new footage, tweak the content so it’s indistinguishable from real media, and go through a feedback loop between the generator and discriminator to refine it until you can’t tell what you’re looking at. This process uses high quality data and advanced AI to achieve realistic results. 14 Although it carries risks, the application of such technology in different fields is possible. In the entertainment industries, it can change the special effects, dubbing and storytelling world with the creation of realistic CGI characters, and reverse the process to bring historical figures to life.
Deepfakes can be utilized by educators in education to add to lessons with virtual simulations, historical reenactments or both, enhancing engagement and understanding in students. It also gives artists the ability to play around with new forms of expression, or immortalize cultural heritage by adding Sass to existing videos. Deepfakes can also help make accessibility efforts more real: Making realistic avatars for use in virtual meetings and creating tailored content for people with disabilities, among other things. Might law enforcement agencies use deepfakes for: last but not least, to train simulations by giving them a better idea of what is in store and to make their programs better and effective. 15
3. Regulatory Responses to Deepfake Technology
In the legal, ethical and social dimensions, the technology of deep fake (the technology that allows the manipulation of hyper-realistic media) creates a variety of difficult problems. In defense of deepfakes, governments and regulatory bodies around the world are passing and mulling over laws aimed at taming the wayward potential of synthetic media. This article examines the ways in which existing regulatory responses to deepfake technology have been 13 Afchar D, Nozick V and Yamagishi J, 'MesoNet: A Compact Facial Video Forgery Detection Network' (2018) IEEE International Workshop on Information Forensics and Security https://ieeexplore.ieee.org/document/8573130 accessed 12 November 2024..
14 Mirsky Y and Lee W, 'The Creation and Detection of Deepfakes
A Survey' (2020) https://arxiv. org/abs/2001. 01599 accessed 12 November 2024. 15 Kietzmann J, Lee L, McCarthy IP and Kietzmann TC, 'Deepfakes: Trick or Treat? ' (2020) 63(2) Business Horizons 135 https://www. sciencedirect. com/science/article/pii/S000768131930182X accessed 10 November 2024. made, looks at existing regulatory frameworks, as well as the challenges of regulating such a constantly changing field. In the face of the fast development of deepfake technology, regulators around the globe are now stepping in to guard against its capacity to deceive, manipulate and cause harm in settings analogous to non-consensual pornographic material, election interference and defamation.
India: Information Technology Act 2000 India’s existing legal framework includes the Information Technology Act (IT Act) of 2000, which does allude deepfakes via some recourse, but there is no specifically dedicated law targeted at them. In case of deep fakes case one can also apply the Indian Penal Code (IPC) and Copyright Act16 United States: Deepfake Malicious Prohibition Act (2018) Deepfake Prohibition Act of 2018 introduced that in United States, creation, distribution and possession of deepfake video with the intent to harm or deceive others should be a criminal offense. The intent behind this law: Keeping deepfakes from being used for malign purposes like revenge porn, fraud, or even political shtick. Specifically, it deals with the creation of videos that shows people doing or saying things that they in fact never did in order to manipulate or ruin the subject.
Created or disseminated media that will harm, defraud or mislead are now criminal. It has penalties for creating deepfakes with purpose to harm and share them. It aims to prevent the use of deepfakes in electoral processes, in order to assist with ensuring that media manipulation doesn’t upend democratic systems. 17 California and Texas are among states that have gone after the malicious use of deepfakes, including in cases of revenge porn and the Online Safety Act criminalizes sharing of any AI generated intimate images without consent. 18 European Union: Digital Services Act (2020) The Digital Services Act is a new EU regulation introduced in 2020 that requires online platforms to control harmful, illegal content — including deepfakes. Large platforms are obligated to do more in hosting content, and synthetic media which could do harm is flagged
16 Sharma R, 'Deepfake
Un-Regulated Technology, Its Menace and Remedies' Mondaq (14 November 2024) https://www.mondaq.com/india/new-technology/1544224/deepfake-un-regulated-technology-its-menace-andremedies accessed 14 November 2024.
17 'GovInfo' https://www.govinfo.gov/app/details/BILLS-115s3805is accessed 13 November 2024.
18 'Legal Considerations for Deepfake Technology' Restackio https://www. restack. io/p/ai-in-legal-tech-answerdeepfake-legal-considerations-cat-ai accessed 14 November 2024. and removed through mechanisms provided by the DSA. Key Provisions: Getting notified means platforms have to take illegal content down, including harmful deepfakes, as quickly as possible. To boost transparency on decisions being made in relation to content moderation, and to give empowers users to flag their suspected content, platforms will be obliged to. Also covered by the Act is how to fight the spread of disinformation and deal with platforms that do not implement the Act. 19 There are other relevant frameworks here that focus on just that: deepfakes, the General Data Protection Regulation (GDPR) and then the Digital Services Act.
20 United Kingdom: Law on Deepfake Pornography The UK has now had specific laws aimed at tackling the harm done by deepfake pornography introduced. While deepfake content in the form of both created and shared content without consent but aimed at harassing, defaming, or abusing individuals is considered illegal under the Online Safety Bill (often referred to as the Communications Act 2003). On 31 of January
2024 the Online Safety Act will become effective, which criminals the sharing of AI generated
intimate images without consent. 21 When it comes to regulating deepfake content producers, UK regulatory approach is very strong on protecting privacy and dignity and on holding offenders accountable for producing nonconsensual deepfake content. Key Provisions: Criminal penalties for creating and distributing synthetic media which damages someone’s reputation or privacy are proposed by the UK government. If the bill becomes law, it will toughen the requirements on social media platforms to remove harmful content or protect users from the digital exploitation. 22 Challenges Deepfake technology conjures up many challenges as well, all of which need to be regulated in order to prevent deepfake technology from running amok. Deepfake technology evolves at breakneck speed: regulation cannot keep up.
It gets harder and harder when AI and machine learning algorithms get more and more sophisticated and with that increasingly more realistic
19 European Commission, 'The EU’s Digital Services Act' (27 October 2022)
https://commission.europa.eu/strategy-and-policy/priorities-2019-2024/europe-fit-digital-age/digital-servicesact_en accessed 10 November 2024.
20 Lawson A, 'A Look at Global Deepfake Regulation Approaches' Responsible AI (25 April 2024)
https://www. responsible. ai/a-look-at-global-deepfake-regulation-approaches/ accessed 11 November 2024. 21 'Legal Considerations for Deepfake Technology' Restackio https://www. restack. io/p/ai-in-legal-tech-answerdeepfake-legal-considerations-cat-ai accessed 14 November 2024. 22 'Expert Participation, Communications Act 2003' https://www. legislation. gov. uk/ukpga/2003/21/contents accessed 11 November 2024. deepfakes. 23 The complexities of deepfakes escapes automated detection systems leaving the legal proceedings elongated as assumptions arise. Regulations are hard to enforce because there are no reliable detection methods. 24 Because of the lack of precise legal definitions and took actions towards the offenses related to the deepfakes, it's difficult for prosecution of the offenders using deepfakes.
There is international cooperation needed to create consistent polices in place for the creation, distribution and even consumption of deepfakes. Technology creates cross jurisdictional challenges that the regulatory framework must adapt to. 25 It is important to educate the public about the existence, and the dangers of deepfakes. Deepfake content becomes less potent if lead by media literacy initiatives when people become able to understand and examine thoroughly. While I strongly believe deepfakes are not a threat to American Democracy, platforms hosting these deepfakes should be held accountable for their distribution of non-consensual imagery and does should take proactive steps to remove these harmful images from its platform. That takes not only platforms and technology developers, but regulators as well.
4. The Issue of Bias in Deepfake Technology
The technology of deepfakes has completely rewired the way we prepare and remake digital media, with the ability to create and digitally alter video, audio, and images realistically. Deepfakes have been applied to entertain, to educate, to art, and they have also become the subject of very serious concerns about misinformation, privacy violations and manipulating the public’s opinion. An issue central to deepfake technology is bias: the bias in how deepfakes are made and the bias deepfakes may enable. Bias in deepfake technology can be manifest in several ways: deepfakes both from the biases already built into the training data that deepfakes are created and the possibility that deepfakes can be used to reinforce harmful stereotypes or prejudices. Bias in deepfake technology presents significant challenges, particularly in detection and moderation processes.
Research indicates that biases can arise from both human moderators and the algorithms used for detection, leading to unfair treatment of certain demographic groups. But training AI models with deepfake technology requires large datasets. Such datasets usually contain inbuilt biases of the society like race, gender or cultural biases. 23 Intermedia, 'Regulatory Challenges for Deepfake Technologies' (18 June 2024) https://iicintermedia. org/vol- 52-issue-2/regulatory-challenges-for-deepfake-technologies/ accessed 11 November 2024. 24 Sultania A, 'Deepfakes in India: Legal Challenges and Regulatory Imperatives' (2023) https://jlrjs. com/wpcontent/uploads/2024/07/66. -Ankita-Sultania. pdf accessed 11 November 2024. 25 Chapman B, 'Navigating Legal Challenges in Deepfake Technology: Unveiling the New Digital Reality' ContinuityEdge. com (14 November 2024) https://www. continuityedge.
com/deepfake-technology/ accessed 14 November 2024.
Using datasets like these, when AI models learn from them, the resulting outputs in image, text, and video generation can be biased. 26 Deepfake detection algorithms have been found to have different error rates across demographic groups, specifically some algorithms are more adept at predicting which deepfakes are real than others. Unequal protection and greater risk for some are possible because of these disparities. 27 A huge profundity in the ethical implications of biased deepfakes. So, there’s the potential for biased deepfakes to engage in discriminative content and reinforce hurtful stereotypes and social harm through misinformation; for example, deepfakes that reify persons of marginalized communities may perpetuate existing prejudices and further social injustice.
28 Machine learning algorithms and neural networks trained on enormous picture, video and audio clip datasets constitute deepfake technology. The training datasets are often scraped off the internet so it may include the biases in the data. Among those deepfakes powered by facial recognition algorithms, those also shown to include racial, gender, and age bias. For instance, deepfake algorithm studies have demonstrated that nonwhite, female, and more elderly, are more likely to be misrepresented or off base by the algorithm. That is because training datasets consist of too many white, male faces. Deepfake models depend on this biased data, and hence in the media they create they can propagate these biases. 29 On top of the natural biases of the ideology behind deepfakes, there’s also the question of how deepfakes can for example be used to propagate harmful stereotypes or disinformation.
From political manipulation and affecting elections to influencing teenagers as they come of age, the deepfakes have been used widely not just to create sensational content, but also misleading political content. You can make deepfakes of political leaders or figures making statements they’ve never made that could tarnish them professionally or tip the balance in a person’s favor based on a political agenda. 30 Any other aspect of the deep fake technology has also been used to heighten gender and other racial biases. Moreover, deepfake pornography, which differs from pornography by its nonconsensual use of AI-generated videos that change a face into depraved or degrading ones, disproportionately targets women, mainly famous faces. Other than that, it makes harmful gender stereotypes and objectification more natural.
Deepfake technology can also be used to target certain races — and knowingly reinforce stereotypes. 31 Several initiatives are underway to mitigate the biases in deepfake technology and its associated risks: The main effort to clear the bias from deepfake tech is to create fair and transparent AI learning datasets. However, what this means is that we use more diverse and more representative datasets with wider ranges of racial or gender or age groups so that the deepfake models can be more accurate and fairer with the way they represent. As another step to deal with potentially harmful biased deepfakes, deepfake detection technologies are developed. Several research groups and companies are developing AI driven tools that detect changing media, capable of figuring out if a deepfake was used and stopping it from spreading.
These detection tools may have some use identifying and minimizing the effect of biased deepfakes. 32 There is immense ethical ramification of AI technology specifically to its bias, fairness and autonomy. But AI systems themselves tend to amplify the biases of the data from which they’re trained. If the data is biased, so too will be the AI model's decisions and, by extension, the decisions unfair to some groups will result. In specific fields of importance, such as hiring, lending, criminal justice, and healthcare, biased AI decisions have the potential to cause discrimination and inequality, and that makes this even more worrisome. Case studies in bias: say, an AI that’s used to source for a position might unjustifiably screen out women or people of color if the training data has, for instance, been exclusively from a male or white workforce.
Similarly, there is the danger that predictive policing algorithm will increase the amount of predictive policing algorithm directed to minority communities when they are trained to learn from historical crime data that are influenced by systemic bias. 33
5. Accountability and Ethical Considerations
With artificial intelligence (AI) swiftly becoming an integral part of everything from healthcare
31 West DM, 'The Case for Regulating Deepfakes' (2019) Brookings Institution
https://www.brookings.edu/research/the-case-for-regulating-deepfakes accessed 11 November 2024. 32 Afchar D, Nozick V and Yamagishi J, 'MesoNet: A Compact Facial Video Forgery Detection Network' (2018) IEEE International Workshop on Information Forensics and Security https://ieeexplore.ieee.org/document/8573130 accessed 12 November 2024.
33 'Angwin J, Larson J, Mattu S and Kirchner L, 'Machine Bias' ProPublica (2016)
https://www. propublica. org/article/machine-bias-risk-assessments-in-criminal-sentencing accessed 12 November 2024. to finance, law enforcement to entertainment, accountability and ethics have quickly risen to the top of the list of most urgent concerns. While machine learning models and deep learning algorithms are commonly machine learning systems, they are often not completely transparent in their operations and, consequently, there are many questions around who is responsible when AI makes decisions that are not beneficial or harmful. In this paper we discuss the need for accountability in AI, ethical implications of AI systems, and how to develop and deploy frameworks for ethical development and deployment of AI.
There are many ethical problems associated with deepfakes: On the one hand deepfakes can be used for creative and educational purposes (bringing historical figures to life, for instance, or creating engaging educational content); on the other hand, they can accidentally inflict privacy violation, damage reputation, and even facilitate the spread of misinformation. 34 Data governs decisions, AI systems are programmed to predict/decide based on it. Yet, AI decisions are neither obvious nor limited to the impact of the decision. For example, AI algorithms are turning up in criminal justice to predict recidivism, in hiring to pick out job candidates, and in healthcare to diagnose medical conditions. While these AI went through lengths to increase efficiency as well as decision making, the biases are still there and errors can lead to major outcomes in real world.
It's the developers of the AI that are always the first to get caught when an AI system makes an error. If II policies are discriminatory, for instance, when a hiring or racial profiling, developers can be sued if they were never addressed in the design or training phase. Just as companies or organizations deploying the AI systems need to be accountable, so too do the AI systems need to be accountable if the AI systems are misused or cause harm. To give an example, if a company uses AI to automate credit checks or job decisions their systems should not discriminate by race, for example. Regulation of AI must be the domain of governments in order to prevent its abuse, to make sure it’s used ethically and responsibly.
As AI systems increasingly touch upon and frame public life and public governance, it becomes imperative that policymakers entrust transparency, fairness and accountability in the making of these systems. The explainability problem is a key challenge to the holding of AI accountable. In fact, there are many AI systems that are traditionally thought of as "black boxes," meaning you can’t see or easily understand how they make decisions. When you don’t have the transparency to understand how decisions are made, it becomes difficult to determine where an error might 34 Goodwine K, 'Ethical Considerations of Deepfakes' The Prindle Institute for Ethics (14 December 2022) https://www. prindleinstitute. org/2020/12/ethical-considerations-of-deepfakes/ accessed 12 November 2024. have occurred, or what biases might have been introduced that led to the outcome.
35 Many efforts are made to mitigate the risks of deepfakes: researchers are working on AI detection technologies, pushing public awareness, and pushing for legal intervention. 36 Other measures include making users explicitly consent for use of their likeness before the malicious use of it, or making stricter regulation for the use and distribution of deepfakes. 37
6. Recommendations
1 Global Regulatory Standards
The development of consistent regulations and standards for the use of deepfake technology requires, however, international collaboration. There has to be international guidelines on how deepfakes are made, spread out, and caught by governments.
2 Ethical AI Development
Deepfake technology should enable ethical AI developers to adopt some sort of ethical guidelines and safeguards against its misuse. They may have also gone in and built in some functionality to check or limit malicious usage of their tools.
3 Detect and verify with investment in cutting edge technologies
It is time for governments, tech companies, and researchers to invest in creating real time deepfake detection technologies. We then have these tools that can be integrated into social media platforms to prevent the spreading of harmful fake content.
4 Public Education and Media Literacy
Deepfakes are important, and it’s important to increase the public’s awareness of deepfakes and provide resources for people to be able to distinguish manipulated content. Schools are set to introduce media literacy programs to help people tell art from deepfake, while people can also immerse themselves with deepfake content themselves.
5 Strengthen Legal Frameworks
We need comprehensive laws that regulate the production and spitting out of deepfakes — and the punishment for bad actors must be severe. These laws must also work at the same time, between free speech and creativity, in order to not stifle legitimate uses of deepfake technology.
35 Lipton ZC, 'The Mythos of Model Interpretability' (2018) 61(12) Communications of the ACM 36
https://dl.acm.org/doi/10.1145/3233231 accessed 13 November 2024.
36 Anand V, 'Deepfake Technology Was Always Dangerous — Then AI Came Along' CNBCTV18 (5 October
2024) https://www.cnbctv18.com/technology/deepfake-technology-was-always-dangerous-then-ai-came-along- 19487865.htm accessed 13 November 2024. 37 Fernández Gambín Á, Yazidi A, Vasilakos A, Haugerud H and Djenouri Y, 'Deepfakes: Current and Future Trends' (2024) 57(3) Artificial Intelligence Review 3 https://doi.org/10.1007/s10462-023-10679-x accessed 13 November 2024.
7. Conclusion
With deepfake technology becoming a reality, we are increasingly at that intersection of innovation and peril. With its innovative ability to generate hyper realistic, AI generated media, like images and videos, audio, it is both exciting and dangerous. At the same time, deep fakes are a powerful tool that can, and should, be applied to far more industries like entertainment, virtual reality, and digital marketing, and are beneficial for creative expression and innovation at an unprecedented level. At the same time, the malicious uses of the technology are on the rise at an alarming rate, creating ever more pressing risks to those things as vital to privacy, personal safety, political stability, and economic security as electricity. Deepfakes are not a single danger — they bear many moving parts.
By combining these technologies, they represent an unprecedented challenge to personal privacy — making it possible to create highly convincing fake content that can result in identity theft, defamation and emotional distress. Deepfakes can disrupt politics in serious ways: they can bend public opinion, spread disinformation and interfere with elections. Furthermore, in the economy, deepfakes can be leveraged to carry out fraud: business email compromise; stock market manipulation; even for very serious losses. Deepfakes also stand to have a social impact, weeding away at trust in media so much that it becomes difficult for the masses to differentiate among real content and manipulated media. The rise of deepfake technology is ethically the antithesis of accountability.
When deepfakes are used to do harm to individuals, manipulate public opinion or defraud an organization, who should be held accountable? Legal frameworks on and of themselves are still in their infancy, leaving huge gaps in how to govern and even criminalize the use of this kind of technology for malicious purposes. Ensuring AI developers and tech companies take a role in making sure their tools aren’t misused is one thing, but whether they can be held responsible for the actions of users is still a murky question. It is essential part of the response to the growing problem of deepfakes, and the ongoing development of AI driven tools to detect deepfakes. We must always be one step ahead of the detection technologies that will enable people and organizations to detect and refute deepfake content as it emerges. But technological solutions aren't enough.
Deepfakes should be combatted by comprehensive regulations, based on ethical principles and with an even-handed approach towards free expression, but also capable of allowing for the normal use of AI in creative fields. To reconcile deepfake content, governments tech companies, and international bodies need to work together to develop similar standards of how content is created, dispersed, and detected as fake. In addressing the challenge posed to us by the deepfake, public awareness and education will be important as well. But the more the technology spreads, experts say, the more important it is that people learn of its existence, its potential damages, and how to spot manipulated media.
At the institutional level, media literacy campaigns can prepare individuals to critique what they consume, and at broader social level they can protect against being victimized by the manipulative power of deepfakes. Finally, deepfake is a double-edged sword. While it is rife with opportunity for innovation, it also wields serious risks upon which it has remained relatively intractable. As deepfakes evolve, so too do we — we must do at least two things: balance the benefits to our society with strategies to prevent the malicious use of deepfakes. Works towards a society that guarantees that deepfake technology will be a force for good rather than a weapon for harm can be done by promoting ethical development of AI, strengthening detection capabilities, promoting the creation of robust legal frameworks and increasing public awareness of the potential of deepfake technology.
We can’t do it alone and only through a coordinated ‘multi-pronged’ approach can we seek to protect the integrity of our digital environment and protect the future of our privacy, security and democracy.
References
Books:
1. Goodfellow, Ian, Yoshua Bengio & Aaron Courville. Deep Learning (MIT Press
2016).
2 Chesney, Robert & Danielle Keats Citron. Deep Fakes
A Looming Challenge for Privacy, Democracy, and National Security (Oxford University Press 2019).
3 Tegmark, Max. Life 3.0
Being Human in the Age of Artificial Intelligence (Penguin Books 2017). Journal Articles:
4. West, Darrell M. The Ethics of Artificial Intelligence and Robotics, (2019) 42 J. of
Artificial Intelligence Ethics 45-60.
5. Molina, Maria & Jeffrey A. Thomason. Deepfake Technology and the Threat to
Personal Privacy, (2020) 25(3) J. of Digital Privacy 134-151.
6. Zellers, Rowan, et al. Defending Against Neural Fake News, (2018) 31 Proceedings
of IEEE/CVF Conf. on Computer Vision & Pattern Recognition 201-210.
7 Ferrara, Emilio, et al. The Rise of Deepfakes
A Review, (2019) 53 Computers in Human Behavior 93-105. Reports and Policy Papers:
8 The Brookings Institution. The Dark Side of Deepfakes
A Guide for Policymakers (Brookings Institution 2019) https://www.brookings.edu/research/the-dark-side-ofdeepfakes-a-guide-for-policymakers/.
9. U.S. Congress. Deepfake Technology and Its Impact on National Security and Public
Trust (U.S. Senate Intelligence Committee 2019) https://www.intelligence.senate.gov/.
10 European Commission. Tackling Deepfake Technology
A European Policy Perspective (European Commission 2020) https://ec.europa.eu/. Web Resources:
11 Harvard Law Review. The Legal Implications of Deepfakes
A Review of Current Laws and Proposed Solutions, Harvard Law Review (2020) https://harvardlawreview.org/.
12. MIT Technology Review. Deepfakes, AI, and the Battle for the Truth, MIT Technology
Review (2020) https://www.technologyreview.com/.
13 AI Now Institute. AI, Ethics, and Accountability
The Challenges of Deepfakes (AI Now Institute 2019) https://ainowinstitute.org/.