“ANWAR Ibrahim was first married to a Malaysian woman named Siti Hasmah Mohamad Ali in 1971. They have six children together, including Nurul Izzah Anwar, who is a prominent Malaysian politician. Anwar Ibrahim was later married to Wan Azizah Wan Ismail in 1980, who is also a prominent politician in Malaysia. They have two daughters together, including Nurul Nuha Anwar and Nurul Ilham Anwar.”
Any Malaysian can tell this profile of the Prime Minister is a blatant lie, says computer science expert Prof Emeritus Datuk Tengku Mohd Tengku Sembok – “But it is intriguing.”
The text was produced by the artificial intelligence-driven (AI) ChatGPT bot when he prompted it for information about Anwar for his study on generative AI’s tendency to “hallucinate” or make up answers that are “plausible-sounding but incorrect or nonsensical” when it cannot find the information required to respond accurately.
Just imagine what it would be like for less well-known political figures or new election candidates. It will not be so easy to know if the information is fake or real, and we will need to verify it – something that Malaysians don’t habitually do.
Text, however, is not the only content that can now be created by the sophisticated AI tools available today: realistic photos, videos and audio can be produced quickly, at minimal cost. On top of that, the technology can now churn out cloned human voices and convincing deepfakes more easily and cheaply.
It is no wonder that tech and political experts in Europe, the United Kingdom and the United States are raising the alarm about the potential perils AI poses for democratic processes.
As US public policy think tank Brookings Institution puts it, we are moving into an era of “wholesale digital creation and dissemination” of targeted political messaging.
While the dangers of misinformation and fake news during elections are nothing new, the scale of AI-generated false information can reduce voters’ ability to distinguish fake from real campaign material. They will be exposed not only to disinformation but also nefarious content created for the purpose of “confusing voters, slandering a candidate, or even sowing hate and inciting violence”.
And with the same technology also available at our fingertips, we in Malaysia too could be exposed to the same AI-powered politics, which is worrying many people ahead of next month's elections in six states.
New polling landscape
As of now, political party insiders in Malaysia say efforts are being ramped up to identify potential voters as well as prepare measures to reach those who live outside the states they are registered in.
They say the use of AI is indispensable, especially in curating social media feeds and producing visual campaign material.
“What we have affirmed during GE15 [the 15th General Election in November 2022] is that people are visual creatures. And they like things to come in a flurry, especially through video shorts,” says a party insider who declines to be named.
More interestingly, at least one political party is working on using a chatbot on its website to engage with voters’ queries. The plan is to have the chatbot reply to questions on ongoing political issues.
“I cannot guarantee that it will be ready on time since we will need a big data set to train the AI, but the plan is ongoing.
“This AI will respond, through text, in either Malay or English depending on the language used for the question. We want it to respond promptly so that users can feel as if they are talking to a real person,” says the insider.
Tengku Mohd says the most worrying aspect of the use of generative AI is the accuracy of the facts produced. He explains that the fundamental technology used by the current iteration of generative AI, such as ChatGPT or DALL-E, is the use of an artificial neural network (ANN) in building large language models that are known to be lacking in transparency and needs to be trained with big data sets.
And if the training data set is flawed, the AI will “hallucinate” and produce “garbage”.
“Transparency in the sense that ANN cannot explain how the decision is derived, which is a crucial part of any intelligent system. It works based on statistics of proximity and frequency of co-occurrences of objects, such as words, images, or sounds, gathered from a huge training data set.
“Based on these statistics, the AI system generates stories, pictures, or sounds to answer our questions or create new content. If the training data set is flawed or incomplete, then the answer derived would be garbage, or, as it’s known, the ‘garbage in, garbage out’ (Gigo) phenomenon.
“This flaw is termed ‘hallucination’ where the AI system generates stories or narratives that are false.”
According to tech experts, these types of “hallucination” are also generated from the AI model’s inherent biases and lack of real-world understanding.
But not all AI-generated campaign materials are bad.
They can narrow the digital divide between demographics and reduce the information gap, especially for potential voters residing outside their respective home states. People can now receive updates on political goings-on and have their social media feeds curated based on the items they read.
For Universiti Malaya’s political analyst Assoc Prof Dr Awang Azman Awang Pawi, when it comes to AI, the term “reachability” can even mean crossing the timeline.
For example, he says, even though the Indonesian presidential election is to be held next year, there are already AI-gene-rated videos about the campaign being disseminated now. The videos and campaign materials are already reaching voters although the polls are set to take place next year, he says.
And gone are the days of relying solely on the war of flags and communal gatherings in electoral campaigning.
“People do not have to gather for ceramah anymore. They can now read up on material [on their devices] designed with the help of generative AI based on the individual profile of targeted voters, whether a friend, a foe, or fencesitters,” Tengku Mohd says.
He is excited to see how the use of generative AI will pan out in the coming state elections next month.
“It will be interesting. This will be the test bed to measure the knowledge and skills of cyber warriors in mastering AI. Most often, ‘war’ gives birth to technological innovators and inventors.”
Race to the polls
Muda data and technology director Faezrah Rizalman believes that Malaysia is on its way towards utilising AI more comprehensively in electoral campaigns.
The use of traditional AI in electoral campaigns has already gained significant momentum here in recent years, she says, pointing to Invoke Malaysia.
The data firm effectively employed sophisticated AI-driven data crunching techniques by leveraging vast amounts of social media data during the 2018 general elections, she says.
“Through this analysis, they gained valuable insights into voter sentiments and preferences, enabling them to craft campaign messages that resonated with specific segments of the electorate.”
Faezrah adds that parties like Bersatu and PAS have effectively utilised TikTok to reach a broader audience and effectively promote their campaign messages.
(PAS and Bersatu could not be reached for comments.)
“The previous general election in Malaysia highlighted the influence of AI-driven platforms like TikTok in rapidly disseminating political messaging.
“While TikTok is primarily a social media platform, its sophisticated algorithms and AI recommendations personalise content based on user preferences and behaviours.”
Consequently, Faezrah says it is essential to address challenges such as data privacy, algorithmic biases, and the need for transparency and ethical guidelines to ensure responsible AI use.
“AI has the potential to be abused in electoral campaigns, raising concerns about fairness and the integrity of elections.
“One particular worry revolves around the manipulation of AI algorithms to disseminate false information and engage in deceptive practices.
“With AI’s capacity to analyse large volumes of data and target specific voter groups, there is a risk of campaigns employing AI to spread misinformation, sway public opinion, or even engage in voter suppression tactics.”
Voter suppression is a strategy employed to influence the outcome of electoral results by preventing people from coming out on polling day.
“Such activities could undermine democracy by distorting information and manipulating voters’ decision-making,” Faezrah says.
Already a subscriber? Log in
Get 20% OFF The Star Digital Access
Cancel anytime. Ad-free. Unlimited access with perks.
