Wednesday 21 February 2024

AI is changing how elections are fought, from deepfake endorsements to chatbot campaigners.

Extract from ABC News 

ABC News Homepage

Recent election campaigns in Pakistan, India and Indonesia offer a glimpse of how artificial intelligence (AI) is changing the way elections are fought, sometimes in surprising and unexpected ways.

From generative AI cartoon avatars of candidates, to political endorsements issued from beyond the grave, to sentiment-tracking chatbots writing stump speeches, parties are deploying the new technology in ways that its inventors may not have imagined.

And with almost half the world's population eligible to vote in elections this year, these AI companies are racing to keep up.

Here's how AI may be used to swing your vote next election.

A campaign team in a box

When Farah Puteri Nahlia travelled to a new part of her electorate during the recent Indonesian election campaign, she turned to AI to learn about her constituents.

The program, Pemilu.AI, sifted through a mountain of data, from the official government census to sentiment analysis of local social media groups, along with commodity prices and fine-grained voting data, to generate a complex picture of the lives of the residents.

Indonesian politician Farah Puteri
Farah Puteri has been youngest member of Indonesia's parliament for the past five years.(Supplied: Farah Puteri)

It told her how many finished school, how many were malnourished, what work was available in the area, what they complained about on social media, and how each polling booth voted in previous elections.

"AI really helps us to simplify everything," she said.

"It helps with the manpower in the field so much."

And if she asked it to, the program would do more than analyse the data. It would write her speeches too.

Powered by OpenAI's GPT-4 and 3.5, it could also generate social media posts and mock up designs for campaign merchandise.

Yose Rizal, a political consultant who developed Pemilu.AI, described the program as "your personal political consultant".

A screenshot from the Pemuli.AI campaign platform
A screenshot from the Pemuli.AI campaign platform.(Supplied: Pemuli.AI)

It performs the kind of detailed analysis, strategy work and copywriting that would ordinarily be done by a team of campaign staff, he said.

Indonesia is the world's third-largest democracy, with more than 200 million eligible voters and around 300,000 candidates competing for seats.

Mrs Farah's West Java electorate has more than 3 million voters.

Mr Rizal said AI helped even the playing field.

"Not all candidates can afford to pay for the consultants," he said.

He added that around 700 candidates used the tool in the recent campaign.

Such technology could be misused to spread AI-generated fake news, he acknowledged, but it had its upsides.

"It's good if you can understand everything about your voters.

"What is dangerous is if someone got elected and they didn't know anything about the voters."

From fearsome general to cuddly grandfather

Katie Harbath, who was until 2021 the top election policy official at US tech giant Meta, said Pemilu.AI was an example of the "unprecedented and groundbreaking" use of AI in Indonesia's recent election.

Another AI innovation from the Indonesian elections was the widespread use of generative AI artwork in campaign materials, Ms Harbath, who writes about the intersection of technology and democracy, said.

Prabowo Subianto, a former ex-special forces commander dogged by allegations of human rights abuses, was the focus of a slick makeover that presented him as a chubby cartoon avatar.

The use of the avatar was so widespread, appearing on everything from posters to TikTok videos, that it eclipsed Mr Prabowo's real image, some experts said.

Mr Prabowo claimed victory last week.

Campaigning from prison using deepfakes

AI also played an important role in the recent Pakistani elections, allowing former prime minister Imran Khan to campaign from prison.

Over two months Mr Khan's party, the PTI, published four deepfake videos in which a clearly labelled AI version of Mr Khan addressed supporters.

A portrait of Imran Khan is held up in a crow which is also holding flags
Supporters of Imran Khan's party the PTI protesting allegiations of vote rigging at the recent elections.(Reuters: Akhtar Soomro)

Jibran Ilyas, the party's social media lead, said the videos made a big impact.

"People were so pleasantly shocked they got to hear their leader from behind bars.

"We had never gone without an Imran Khan speech at any political rally for the last 26 years."

The videos, which were made using technology from US company ElevenLabs, were generated from archival footage of Mr Khan and notes he'd passed from prison, where he was barred from recording audio and video.

In February, PTI declared it had won in a surprise landslide. An AI Imran Khan delivered the victory speech.

"People were crying. It was very emotional," Mr Ilyas said.

The outcome of the election has since been mired in allegations of government-backed vote rigging.

An endorsement from beyond the grave

AI Imran Khan isn't the only example of political parties deploying AI deepfakes for their own gain.

In India, dead politicians have been resurrected to endorse the living.

Muthuvel Karunanidhi, an icon of Indian cinema and politics, has appeared at media events and book launches for the past six months, despite having died in 2018.

A screenshot from the deepfake video of Muthuvel Karunanidhi
A screenshot from the deepfake video of Muthuvel Karunanidhi speaking at a book launch.

The deepfake version, usually displayed on a screen above a stage, praises the leadership of his son, MK Stalin, chief minister of the state of Tamil Nadu.

Senthil Nayagam, founder of tech firm Muonium, which made the deepfake of Mr Karunanidhi using open source software, expects strong demand for these types of videos.

His firm has already cloned the voices of 45 current and former Indian political leaders, including Mahatma Gandhi, ahead of the Indian elections this year.

These clones can be used to help candidates communicate with constituents who speak other languages, he said. India is home to several hundred languages.

"We showed that Prime Minister Modi's speech can be translated into 30 languages," he said.

The other use of deepfakes, he expects, will be simple nostalgia. 

Parties could wheel out deepfakes of cherished former leaders, particularly where the current leaders aren't so popular.

"This is not about India alone," Mr Nayagam said.

"What if [former US president] Reagan comes and speaks in the US elections?"

So what's in store for 2024?

Ahead of this year US presidential election, concerns are running high about the potential for AI deepfakes to undermine the vote.

In January, OpenAI banned the developer of a bot that allowed voters to have conversations with an AI-generated version of Dean Phillips, who's running as a candidate for the Democratic Party presidential nomination.

That same month, a fake robocall claiming to be from US President Joe Biden urged voters to skip a primary election.

Last week, tech giants including Microsoft, Meta, Google, Amazon, X, OpenAI and TikTok signed a pact to voluntarily adopt "reasonable precautions" to prevent AI tools from being used to disrupt democratic elections around the world.

Specifically, the accord aims to counter "harmful AI-generated content meant to deceive voters" including audio, video and images that "deceptively fake" the appearance, voice or actions of candidates and "other key stakeholders".

Whether this applies to consensual deepfakes, like Mr Khan's, remains to be seen.

In Australia, the work of policing electoral communications falls partly on the Australian Electoral Commission (AEC).

A spokesperson said the AEC was "alert to AI and deepfakes and the risk they present as potential vectors of mis- or disinformation about electoral processes.

"While AI is a constantly evolving technology, the rules around the authorisation of electoral communication have not changed.

"For example, videos must include a written and spoken authorisation message at the end of the video, and telephone calls must include a spoken authorisation message at the beginning of the call.

"This remains true regardless of how the content was created."

For all the legitimate concerns about deepfakes undermining the US vote, some experts say overstating the risk can be detrimental to democracy.

Voters may end up not trusting any electoral information, Ms Harbath said.

"My biggest concern is actually the narrative around AI versus the actual use of it.

"I'm worried that if we aren't careful, people will think AI is more pervasive than what it is and that alone could reduce trust in the information environment."

In Indonesia, Mrs Farah believes there's no going back from the use of AI to help analyse electorates.

"It's not only useful for the elections, but after the elections," she said.

"We can really narrow down the issues that are happening in our electorate.

"Everyone's going to use it by the end of the day."

No comments:

Post a Comment