One month before local office elections and referendums on November 28, the Taiwan Fact-Checking Education Foundation (Taiwan Fact-Checking Center) colluded with MyGoPen and Cofacts The three major verification organizations jointly issued a self-discipline convention on "Maintaining Election Fairness and Rejecting AI Deep Forgery" today (29th), calling on political parties, candidates and campaign teams to promise to abide by the principle of "no two" and not to produce or disseminate political deepfakes that can be mistakenly believed to be true; if mistakenly believed and forwarded, they should be removed immediately and publicly corrected with the same visibility.
In the face of the rapid evolution of AI technology, there have been many cases at home and abroad of using AI deepfake technology to fake the words and deeds of politicians in an attempt to influence elections. This self-discipline convention to reject deep fakery has been jointly responded to by 12 non-governmental organizations spanning the fields of law, media supervision, Internet governance, technology industry, and youth.
Huang Zhaohui, chairman of the Taiwan Fact-Checking Education Foundation, said that generative AI has brought the threshold for fraud close to zero. Deep fake videos that look fake and real can be generated with one click in a few seconds. However, fact-checking cannot determine authenticity with just one-click detection.
She said that behind each verification report, the team was not afraid of political pressure from all parties, tracing back the original images, comparing the audio tracks and context, and consulting experts. It took hours or even days to lay out the evidence layer by layer in the sun. In the past three years, more than 50 political and election-related AI deepfake verification reports have been accumulated. From this, we can see that Taiwan, regardless of the blue, green, and white camps, has fallen victim to it.
Huang Zhaohui pointed out that although the fact-checking team pursues the truth, the speed of refuting rumors afterwards can never keep up with the viral spread of algorithms, and judicial prosecution is even more slow in an election campaign where every second counts. The deeper crisis is the "liar dividend". When deep fakes are rampant, politicians who actually make mistakes or break the law in the future can easily blame AI for smearing, eventually leading society to fall into the abyss of zero trust where "nothing is believed."
She said that by drawing on the experience of European partners who are also members of the International Fact-Checking Alliance, the line of defense has been moved from "post-facto verification" to "source prevention." In democratic elections, we can have fierce debates on ideas and policies, but we must not be deceived by AI deepfakes, and we must not let fake images and fake voices steal real votes.
The principle of "two musts" is a self-discipline convention that calls on all political parties, candidates and campaign teams to abide by it. "Two Nos" do not produce or disseminate AI deepfake videos and images that "allow political opponents and their camps to say things they have never said and do things they have never done."
The second requirement is to remove it immediately and correct it with equal visibility. If you mistakenly believe and forward deepfakes content, you should remove it immediately after discovering the error, and make public corrections with equal visibility through the same public channel.
