AI needed to combat disinformation as it moves to video: Taiwan expert

Taipei, March 29 (CNA) AI is needed more than ever to combat disinformation as it becomes more prevalent on video-based platforms such as YouTube and TikTok than in text-based formats, a Taiwanese expert warned during an international panel discussion on Thursday.

Ethan Tu (杜奕瑾), the founder of Taiwan AI Labs, said his company found a rise in troll accounts in the run-up to Taiwan's presidential and legislative elections in mid-January that disseminated false information using short videos and that AI tools were used to identify the accounts.

Tu was speaking at a livestream event held by the Institute of Global Politics at Columbia University's School of International and Public Affairs on how AI tools have changed the information ecosystem and how governments and the public will respond to new challenges.

The event was held as many countries, including India and the United States, prepare to hold key elections later this year in which online disinformation could be used to shape narratives, just as Tu said occurred during Taiwan's presidential vote.

He told the panel that his company observed a large amount of coordinated activity on online social media platforms such as Facebook, X (formerly known as Twitter), PTT (a popular online forum in Taiwan launched in 1995), and TikTok ahead of the election.

"We used AI to identify those people, who were actually not real humans," and found that they "appeared together and disappeared together," a clear sign of coordinated activity that is characteristic of troll groups, Tu said.

These troll accounts are "referencing short videos now," a different approach of spreading information from using text, which has been their preferred method in the past, he said.

The AI Labs founder said these were deepfake videos shown on platforms such as YouTube and TikTok that used different backgrounds and voices but trumpeted the same narratives.

"We used AI speech recognition and language understanding tools to identify the troll accounts and cluster the stories they were trying to spread," Tu said.

He said there have been peaks of troll activity when events with political significance take place, such as President Tsai Ing-wen's (蔡英文) visit to the U.S. in April 2023 and when U.S. President Joe Biden said the U.S. would defend Taiwan if China attacked.

Asked by moderator Anya Schiffrin, a senior lecturer on international affairs at Columbia University, whether China could be identified as the source of the troll activity, Tu said it could.

"By using AI and grouping the same narratives together, we found that the troll accounts on the social media platforms echoed the narratives of China's state-affiliated media," he added.

He also noted that many of the troll accounts "try to emphasize China's military strength," suggesting a link to China.

The panel discussion also touched on regulations in place to monitor disinformation on social media platforms and the difficulties that entailed.

In Taiwan's case, Tu said, "if we go to the direction of fact-checking and content moderation, there would be a lot of challenges because people would say that was against freedom of speech and there would be debate over what 'fact' is in fact-checking."

So instead of fact-checking or content moderation, Taiwan now focuses more on information manipulation disclosure, he said.

Other event participants included former U.S. Secretary of State Hillary Clinton.