How will AI shape our upcoming presidential election? Can AI-generated fake audio mislead voters? Is AI a threat to election integrity?
In an article in the Sunday Times, Sri Lanka, LIRNEasia’s Research Manager and Team Lead: Data, Algorithms, & Policy Merl Chandana highlighted these concerns and emphasised the need to address AI’s potential to create misleading and harmful content.
He told the Sunday Times: “AI can generate text, video, audio, and photos. However, text generation is not advanced for local languages, especially Sinhala. Tamil might see some progress due to developers in South India, making it a bit more advanced than Sinhala. There’s no need to worry about AI’s text-generating capabilities in Sinhala, but someone could take 15 minutes of a candidate’s speech, train AI to mimic their voice, and then produce fake audio. This fake audio could be circulated on platforms like WhatsApp, making it seem like the candidate said something they didn’t.
“This is very problematic with WhatsApp circulations, as it’s difficult to trace the origin, but it sounds like the person actually said it. Although this hasn’t been observed here yet, especially around elections, it has been an issue in Pakistan and India in recent elections. While video manipulation is more advanced and harder to debunk, audio is more easily spreadable.”
