22 November 2024
Dealing with the Deepfake Dilemma
NUS Alumni Relations co-organised a spirited discussion centred on how society should navigate deepfakes while embracing innovation.
The discussion touched on legislation and regulation, public trust and transparency, as well as the ethical considerations around the use of AI and generative technologies.
Early in September 2024, the Singapore Government issued a clarion call against deepfakes, tabling a Bill in Parliament to tackle the scourge. The proposed Elections (Integrity of Online Advertising) (Amendment) Bill aims to counter deepfakes during elections.
The law comes as deepfakes increasingly become a part of everyday life — just last year, the Internet was abuzz with deepfakes of then-Prime Minister Lee Hsien Loong and then-Deputy Prime Minister Lawrence Wong promoting investment platforms. The videos used artificial intelligence (AI) to mimic their voices and facial expressions. Deepfakes sometimes also have less malicious intent. For example, de-ageing filters that enhance one’s looks, can — in the view of some academics — also be considered a form of deepfakes. But is this too extreme a position?
To help alumni cut through the noise surrounding the issue, the National University of Singapore (NUS) Alumni Relations partnered the Institute of Systems Science (NUS-ISS) to organise a panel discussion featuring academics, regulators and industry leaders. Themed The Rising Threat of Deepfakes – Understanding the Surge and Its Implications, the session, which was held on 3 September 2024 at the Shaw Foundation Alumni House, attracted over 180 attendees, ranging from those whose work revolved around deepfakes to those who were just learning about it. Mr Martin Khoo, Principal Lecturer & Consultant, Digital Strategy & Leadership Practice, NUS-ISS, moderated the session.
A central tenet of the discussion was the importance of considering the interests of various stakeholders while not stifling technological progress when it comes to the use of AI.
BALANCING INNOVATION AND REGULATION
Dr Saurabh Lal (MTech SE ’08) set the context for the evening’s discussion by providing a broad overview of deepfakes and explaining how they are created. Drawing on his experience as President of cybersecurity provider CYFIRMA, he articulated the importance of clear legislation that defines the scope and impact of deepfakes, with a focus on transparency, enforcement mechanisms and victim protection. He also stressed that regulations must consider not only the publisher and the victim, but also the intermediaries involved in the distribution of deepfake content. These include social media platforms, such as Facebook, Instagram and TikTok.
Ms Lim Ee Lin (MTech DL ’24), who serves as the Deputy Director, CISO & Governance, Home Team Science & Technology Agency, pointed out the importance of a balanced approach. “There are expectations for the Government to introduce more regulations and guidelines to address the deepfake issue,” she explained. “We are developing regulations, guidelines and standards that can be adopted by industry. The goal is to ensure that there is a baseline alignment across the ecosystem, where everyone is on the same page about the expectations and requirements.”
However, Ms Lim acknowledged that the Government has to contextualise regulations for specific use cases across industries. She highlighted that the Government also has a duty to protect the public interest. “The Government has to balance these different stakeholder interests when developing guidelines,” she said.
We cannot simply say that because open-source tools make these technologies easily accessible, we should prohibit any further developments. That would be akin to stifling innovation itself.
Mr Kenneth Zhu
INDUSTRY IMPACT
Mr Kenneth Zhu, Executive Director of Data Science & AI at OCBC Bank, shared the banking sector’s perspective. He highlighted the need to balance innovation with regulation, noting that open-source tools — which are often associated with the creation of deepfakes — can be both beneficial and risky. However, he cautioned against outright prohibitions of these tools. “We cannot simply say that because open-source tools make these technologies easily accessible, we should prohibit any further developments,” he said. “That would be akin to stifling innovation itself.”
Mr Zhu also stressed the importance of responsible development and use, suggesting that the banking sector work together to establish policies and guidelines. He highlighted the sector’s focus on protecting client data and payments, noting that even as tech evolves, the core objective of safeguarding customer interests must remain a priority.
To this, Dr Lal supported the idea of self-regulation, adding that industry players could collaborate to create associations or working groups to establish guidelines and standards. This, he argued, would help ensure that the development and use of these technologies do not harm society or unfairly target individuals or organisations.
The Q&A session provided audience members with an opportunity to share their experiences of deepfakes.
THE AUDIENCE WEIGHS IN
The session concluded with questions and comments from the audience, who shared personal experiences of deepfakes. They also raised concerns about privacy risks from sharing data with such apps and worries about the potential misuse of AI-generated faces or voices to compromise their children’s safety. Panellists suggested mitigating these risks, for instance, by training children on security protocols and understanding platform terms of use. “As a society, we cannot run away from the need to understand these technologies,” said Mr Khoo, pointing out various NUS-ISS courses that were ideal for those looking to deepen their understanding of deepfakes.
DEEPFAKES AT THE POLLS
Election Safeguards:
Singapore’s new Bill proposes fines of up to S$1 million for social media platforms that fail to comply with directives to remove deepfake content misrepresenting political candidates during elections.
New Detection Arsenal:
Singapore is investing S$50 million to develop advanced detection tools as part of a broader initiative to tackle deepfakes and safeguard online trust and safety.
Global Impact:
Deepfakes have disrupted elections worldwide, with countries like South Korea and Brazil implementing bans or restrictions on AI-generated content to protect electoral integrity.
First AI Election:
The US election in November 2024 is widely regarded as the world’s first ‘AI election’ due to the proliferation of deepfake videos featuring political candidates.
Text by Keenan Pereira. Photos by Mark Lee.