Policymakers worldwide are grappling with escalating concerns about AI-generated disinformation influencing voters and exacerbating divisions in the lead-up to major elections in the coming year.
This issue has already become a reality in Bangladesh as it gears up for an early January general election, the British daily Financial Times reports.
Over the past few months, some pro-government media outlets and influencers in the country have actively promoted disinformation created using low-cost tools offered by AI startups, the report says.
In one instance, an AI-generated news clip featured an anchor criticizing the United States, a nation the ruling Awami League has lambasted in the pre-election period.
Another deepfake video, later removed, depicted an opposition leader wavering in support of Gazans–a potentially damaging stance in the Muslim-majority country with significant public sympathy for Palestinians.
Role of global tech giants
As global attention focuses on combating misleading AI content ahead of major elections next year, including those in the US, the UK, India and Indonesia, efforts by tech giants Google and Meta to enforce disclosure of digitally altered political adverts have been announced.
However, the situation in Bangladesh underscores the challenges in controlling the use of AI tools in smaller markets, often overlooked by major American tech companies.
Miraj Ahmed Chowdhury, managing director of the Bangladesh-based media research firm Digitally Right, emphasized that while AI-generated disinformation was still in its experimental stage, the potential threat becomes substantial when these technologies enable the mass production of misinformation.
Miraj said a platform's attention to a particular jurisdiction depended on its importance as a market.
Heightened concerns
The rise of tools like OpenAI's ChatGPT and AI video generators has heightened global concerns about the creation of misleading or false political content.
A dystopian future under President Joe Biden was depicted in an attack advertisement released by the US Republican National Committee earlier this year. Meanwhile, YouTube suspended several accounts in Venezuela that used AI-generated news anchors to spread misinformation in support of President Nicolás Maduro's administration.
Instances of AI-generated disinformation in Bangladesh include a video posted on the online news outlet BD Politico, where an AI-generated news anchor accused US diplomats of interfering in Bangladesh’s electoral process, blaming them for political violence.
The video was produced using HeyGen, a Los Angeles-based AI video generator available for as little as $24 a month.
The same anchor, known as Edward, appears as one of the many avatars–all created from actual actors–in HeyGen's promotional material.
The Financial Times said neither HeyGen, X nor BD Politico had responded to its requests for comment.
Similar examples on Facebook include deepfake videos targeting the BNP, which prompted the party to request content removal from Meta.
Similar instances include anti-opposition deepfakes shared on Facebook, such as one that appears to show BNP acting chairman Tarique Rahman, who is currently in exile in the UK, advising the party to "keep quiet" about Gaza to appease the United States. The media non-profit Witness and the think tank Tech Global Institute came to the same conclusion: the fake video was probably created using AI.
BNP official AKM Wahiduzzaman claimed that although his party had urged Meta to take down such content, "most of the time they don't bother to reply."
The video was later taken down by Meta following a request for comment from the Financial Times.
Threat to public trust
According to Sabhanaz Rashid Diya, founder of Tech Global Institute and a former Meta executive, one of the main obstacles to recognizing such misinformation is the dearth of trustworthy AI-detection technologies, with off-the-shelf products being especially bad at identifying content not written in English.
Diya went on to say that in nations like Bangladesh where political advertisements played a minor role in political discourse, the solutions proposed by major tech companies, who had concentrated on controlling AI in political ads, would have little impact.
The Financial Times said the issue was made worse by the absence of regulations or the authorities' selective enforcement of them. For instance, the Cyber Security Act had drawn criticism for granting the government harsh authority to suppress online opposition, it added.
Diya said the potential for politicians to dismiss inconvenient information by labelling it as deepfakes or AI-generated added an additional layer of complexity to the issue.
The issue facing the Global South would be how AI-generated content was weaponized to distort people's perceptions of what was genuine and incorrect, she added.