In the intricate nexus of technology and humanity, the dynamic evolution of Artificial Intelligence (AI) raises significant concerns about the preservation and defence of core human values. AI is the theory and development of computer systems capable of performing tasks that historically required human intelligence, such as recognizing speech, making decisions, and identifying patterns. AI is an umbrella term that encompasses a wide variety of technologies, including machine learning, deep learning, and natural language processing. Despite AI promising numerous benefits, its extensive integration into the private lives of humans has led to the alarming infringement of fundamental human rights, particularly the right to privacy and dignity. This blog will analyse how cybercriminals maliciously use AI tools and algorithms to violate the right to privacy and dignity of others. It will further shed light on potential solutions for countering these violations.

International and Constitutional Protection

Article 12 of the Universal Declaration on Human Rights (UDHR) declares that people are not to be subjected to arbitrary interference with their privacy. The International Covenant on Civil and Political Rights (ICCPR) makes this an actionable and enforceable right under Article 17, creating a positive obligation on the State to ensure the respect and protection of a person’s ‘privacy, family, home or correspondence’ as well as protection against ‘unlawful attacks on his honour and reputation.’ This right is espoused domestically under Article 14 of the Constitution of Islamic Republic of Pakistan, 1973.

In the landmark judgment of Mohtarma Benazir Bhutto and Others. vs. President of Pakistan and Others (PLD 1998 SC 388) the Supreme Court defined the extended scope of right dignity and privacy in light of Article 14 of the Constitution. The Court emphasised that privacy extends beyond the boundaries of one’s home and includes public spaces. Any illegal intrusion or invasion should be forbidden to preserve the dignity of individuals (para 29). Thus, the right to privacy extends outside the home as well and can include privacy of communications.

Unfortunately, Pakistan has yet to implement any legislation specifically addressing Artificial Intelligence (AI) and its applications. However, a notable development occurred in May 2023 when the Ministry of Information Technology and Telecommunication drafted a National Artificial Intelligence Policy for Pakistan, which is currently published for review. The primary vision of this policy draft is to embrace AI by appreciating human intelligence and stimulating a hybrid intelligence ecosystem for equitable, responsible, and transparent use of AI. Furthermore, this policy stems from the ‘AI for Good’ initiative led by the International Telecommunication Union, alongside the Sustainable Development Goals (SDGs) established by the United Nations.

Despite being a worthy product of human innovation, the widespread integration of AI, notably into the private domain of human life, has created unforeseeable challenges and hurdles for individuals. One of these applications is AI voice cloning technology which is a novel method frequently used by hackers and other cybercrime actors who misuse the technology to fulfill some of their ulterior motives.

How AI Voice Cloning Works

AI-voice cloning operates on the basis of deep learning. Deep-learning is a method of machine learning that is based on artificial neural networks which allows AI to process data in a manner similar to the neurons in a human brain. That is to say, the more human-like AI becomes, the better it is at emulating human behavior. Through extensive deep-learning methodologies, these neural networks have the capability of replicating speech patterns and pitches of the human voice. The users of such networks simply have to input text, and the AI voice cloning system will generate the corresponding audio based on its learned speech behaviours. The more speech data they are exposed to, the more adept they become at emulating human speech. Due to relatively recent advancements in this technology, state-of-the-art text-to-speech software can essentially replicate the sounds it is fed.

For instance, while sharing videos or voice notes on social media platforms, one’s audio can be easily usurped by notorious scammers by using AI-driven voice cloning techniques. It becomes nearly impossible for one to distinguish the artificially produced voice from the victim’s real voice. These alarming expansions of AI tools infringe upon fundamental human rights, particularly the rights to privacy, dignity, reputation and, in many cases, access to accurate information.

It is evident that the sanctity of private data has been compromised in light of contemporary realities, wherein purportedly secure end-to-end encryption protocols on social media platforms are at risk of exploitation by cybercriminals. Communications transmitted through telephone, photographic images and audio recordings stored within digital recorders are fundamentally devoid of confidentiality and security reassurances. The act of carrying a mobile device within one’s possession parallels residing within a domicile devoid of boundaries, thereby rendering personal data vulnerable to unauthorised access facilitated by modern technological means.

Examples of AI Voice Cloning Technology Infringing on Honour and Reputation 

In Pakistani politics, a notable trend involving audio leaks has surfaced. These leaks allegedly capture conversations between various renowned figures and political leaders discussing sensitive and controversial issues. However, many of these were fake and considered to be generated by AI voice cloning technology.

In May 2023, a widely circulated audio recording allegedly features the voice of Behroze Sabzwari, a prominent Pakistani television actor, posing commentary on the Pakistani army. The alleged remarks incited strong condemnation from the public, casting a shadow of controversy over the actor’s reputation. The voice presumed to be Sabzwari’s can be heard saying, ‘Corp Commanders have put their foot down and [General] Asim Muneer is in Oman, he’s been stopped there. I think he’ll be court-martialed. Everything will be revealed in a few hours.’ Following the audio leak Sabzwari issued an official statement aimed to address the controversy to dispel the doubts. He vehemently denied the authenticity of the widely circulated audio, attributing it to being generated by AI.

In November 2021, a voice, allegedly of former Chief Justice of Pakistan Mr. Saqib Nisar, was heard talking to an unidentified person. He said that Ex-PM Nawaz Sharif and Maryam Nawaz should be punished to make space for Imran Khan to come into power. He told media persons over the telephone that the accusations levelled against him were ‘contrary to the facts’; therefore, he did not want to respond to the ‘plain lies.’

Similarly, there have been several instances internationally where audio clips allegedly containing prominent figures surfaced on social media, addressing controversial topics and sparking widespread debate among the public. However, the accused parties vehemently denied the authenticity of the audio claiming they were digitally manipulated fabrications and not their original voices. In 2023 fake audio recordings of Emma Watson, an actress best known for the Harry Potter series, reading Mein Kampf by Adolf Hitler were released. Watson later denied the authenticity of the recordings and claimed that the circulated audio was AI-generated by her haters to harm her repute among her fans.  Cybercriminals took the voice-cloning technology to create the audio files and posted them on the message board 4Chan.

In March 2023, a viral TikTok video caught the attention of The New York Times. In the video, famous podcaster Joe Rogan and Dr. Andrew Huberman, a frequent guest on The Joe Rogan Experience, were heard discussing a ‘libido-boosting‘ caffeine drink. The video made it appear as though both Rogan and Huberman were unequivocally endorsing the product. In reality, their voices were cloned using AI.

On March 15, 2023, a clip of a vintage tape recorder accompanied by a voice bearing a striking resemblance to President Joe Biden was circulated online. In this clip, ‘Biden’ talked about a Silicon Valley Bank collapse with an emphasis on calming the public. But later, experts showed that the audio was machine-generated and made to disturb the public order.

These various examples indicate the ability of AI voice cloning technology to harm one’s honour and reputation. With AI tools, it is now feasible to generate audio recordings that accurately mimic a specific individual’s voice tone, accent, pitch, and frequency. Through text-to-audio tools, one can generate a speech in the voice of any desired person and incorporate any desired content into the audio clip.

From the abovementioned examples, it becomes evident how AI-driven voice cloning techniques can be used to prejudice individuals’ reputations, potentially leading to grave outcomes for those trapped by such fabricated audio leaks. This constitutes a blatant infringement of the fundamental rights to dignity and privacy enshrined in our Constitution.

Recommendations

To address the increasing prevalence of AI-powered voice cloning tools and their adverse societal impacts, several recommendations are proposed. Firstly, comprehensive legislation must be enacted with the aim of advocating and specifically targeting the deliberate misuse of AI technologies. Additionally, the need for proficient implementation mechanisms is emphasised, as demonstrated by the inadequacies in the National Artificial Intelligence Policy draft. Incorporating established global standards and guidelines into national legislation, coupled with a stringent regulatory framework, is crucial to banning AI tools capable of generating fake audio and exploiting vulnerabilities. Collaborative initiatives with tech companies are recommended to develop software for identifying forged content while strengthening national privacy laws and enforcing strict punitive measures aimed at deterring offenders.

Furthermore, public awareness campaigns and promoting digital literacy to verify information authenticity are essential for educating individuals about AI-driven cybercrimes, including voice cloning. Establishing recourse mechanisms for victims of AI-enabled crimes, such as provincial complaint cells with streamlined legal procedures, would enhance access to justice and restitution. Finally, rigorous legal frameworks for companies overseeing voice cloning applications are proposed, mandating user authentication and clear purposes prior to providing cloned voices. These measures collectively aim to mitigate the detrimental impacts of AI voice cloning and uphold societal integrity and security.

Conclusion

In a nutshell, the rise in AI technology and its unbridled interference in the personal lives of humans have not only disturbed societal order but have also violated fundamental human rights, notably the right to privacy and dignity. This blog has explored how AI-voice cloning tools can be exploited for ulterior motives, highlighting the urgent need for solutions to mitigate these threats. Through legal regulations, technological innovations, and improved digital literacy, States can address the challenges posed by AI-driven risks and uphold individual rights and integrity.

Centre for Human Rights (CHR) blog