Bill to counter election deepfakes meets need to uphold trust in electoral process: Janil

SINGAPORE - A proposed law against AI-generated misinformation during Singapore’s election periods addresses the specific need to maintain public trust in the electoral process here.

This is in keeping with the Republic’s approach to regulating the use of artificial intelligence (AI), where legislation is put up to deal with specific challenges instead of having just a wide-ranging law on AI, said Senior Minister of State for Digital Development and Information Janil Puthucheary on Sept 20.

Speaking at a panel discussion on AI and governance at the Lee Kuan Yew School of Public Policy, Dr Janil said the Elections (Integrity of Online Advertising) (Amendment) Bill tabled in Parliament earlier in September addresses a very different need from the Protection from Online Falsehoods and Manipulation Act (Pofma), the Republic’s fake news law.

Under Pofma, there is a public interest test and a requirement that fact be established by an authoritative third party, he said.

But if an AI-driven tool is used to misrepresent a candidate’s words and actions, the only person who can verify this is the candidate, added Dr Janil. “The Government is in no position to do so – a third party, an academic, a professor, a research institution is in no position to do so.”

Ultimately, the aim is to shore up public trust during campaigning, and to prevent Singapore from going the way of countries where information provided to the electorate is being called into doubt, resulting in falling trust in both institutions and public discourse, he added.

The panel, moderated by Institute of Policy Studies principal research fellow Carol Soon, also included Google’s head of AI and emerging tech policy (Asia-Pacific) Eunice Huang, Straits Times editor Jaime Ho and OpenAI’s head of policy (Asia-Pacific region) Sandy Kunvatanagarn.

Kicking off the discussion, Dr Soon asked if governments and industry need to do more to safeguard election integrity amid the proliferation of AI tools.

Dr Janil responded that Singapore is privileged, in that there is generally high trust in elections here, but that the process of getting information to voters to make their choice “is ripe for manipulation with AI-driven tools”.

That is why the Republic has taken steps in recent years by passing laws such as Pofma, as well as others that tackle online criminal harms, foreign interference and harassment, with each undergoing fairly robust debate in the House.

The new Bill will undergo a similar process, added Dr Janil, who is also Minister-in-charge of the Government Technology Agency.

“The Bill hasn’t been debated, certainly hasn’t been passed, and so we will have to scrub through some of these details and give an accounting to the public of how we will do this,” he said.

Among its proposed safeguards is for digitally manipulated content – whether generated with AI or non-AI techniques like Photoshop and dubbing – to be illegal from the time the Writ of Election is issued to the close of polls.

If passed, election candidates will be able to ask the Returning Officer to review content that has misrepresented them, which could lead to corrective actions such as an order for the offending content to be taken down.

A false declaration of such misrepresentation is illegal, and could result in a fine or loss of a seat.

Other panellists said AI firms are also doing their part to prevent their products from being used to generate disinformation.

For instance, OpenAI has policies that disallow its tools from being used for campaigning or to create impersonation chatbots, said Ms Kunvatanagarn. There are also safeguards so that its image generator Dall-E is not used to create images of real people, she added.

Ms Huang also brought up Google’s SynthID, an invisible watermark put into text, video and pictures generated by its AI assistant Gemini, as well as ongoing efforts by industry to develop technical standards to trace the provenance of media content.

Panellists were also asked if AI could one day replace policymakers in analysing social issues and creating new policy.

Mr Ho said the imperative is to assume that society will get there, whether or not that eventually happens.

Doing so “forces us to reckon with it earlier, and to recognise that we do need to take this time to crystallise what it is that we do as humans, that value-adds”, he said.

He noted that newsrooms are grappling with similar concerns, and that a challenge as AI develops is to better explain what journalism is, what journalists do, and to show the value-add that they bring, whether in verifying facts or the kinds of stories pursued.

“It will be a kind of scary process, but at least it forces us back to really the basics of what we do. And I’m sure the policymakers are doing that as well,” he added.

Dr Janil said policymaking requires a certain sense of imagination. This entails not only intuiting how people will respond to a specific policy decision, but also imagining what society is ready for and taking the leap of faith in making new policies.

He noted that AI firms today are refining their models to eliminate “hallucinations”, which are incorrect or misleading results generated due to faulty training data or algorithms.

A time could come when AI could draw inferences and synthesise knowledge in a way that is trusted and “that we don’t think of derisively as a hallucination”.

“Then, you run the possibility of them finally closing the gap, (though) I don’t see that on the horizon,” he said.