AI fakes in campaign mix

IN a March interview with Newsday, cyber security expert Sean Hoyte urged voters to exercise caution in their consumption of media during the short April 28 general election campaign.

Mr Hoyte suggested that social media users should adopt the ABCs of investigation – Assume nothing, Believe no one, and Check everything.

Matthew Stone, an artificial intelligence researcher and graduate student of UWI’s Jamaica Faculty of Science and Technology warned in the Jamaica Observer in March of the potential for “irreparable harm” caused by deepfakes.

These are sensible warnings.

AI technologies applied to deepfake videos, which use existing footage to create invented visuals, have evolved beyond simply matching the movement of lips to match invented new words to creating persuasive clips of wholly invented scenarios. Abetted by improvements in voice-cloning technology, which creates new audio based on existing, readily available voice samples, the falsehoods can be persuasive.

This is a new frontier of digital fakery, as these new technologies increasingly rely on next generation Massive Machine Learning models, which are rapidly eliminating the indicators that once marked AI-generated images with excess fingers, videos with unnatural movement and the staccato robotic tones of voice cloning.

TT can learn from the experiences of countries which had to manage the growing use of AI in elections held in 2024.

In February 2024, the US Federal Communications Commission outlawed AI-driven robot calls that used voice cloning tools.

AI robocalls in January 2024 mimicking then-president Joe Biden’s voice discouraged people from voting in the New Hampshire primary, a critical bellwether caucus for nominees in the US voting race. The creator of that campaign was fined US$6 million by the FCC and indicted on criminal charges in New Hampshire.

The FCC’s position led executives from major US tech companies to sign an accord recognising the threat and vowing, at least on paper, to respond to deepfake technologies on their platforms.

In Indonesia, Suharto was reanimated by AI to endorse the candidates of the Golkar Party. The dictator died in 2008 at the age of 86. The Golkar Party won the election.

Beyond outright falsehoods, AI tends to be used for the most primal of political campaign goals, to make your candidate look good and your nemesis look evil, usually by underlining widespread rumours or concerns.

The PM and Opposition Leader made such claims against each other in their respective PNM and UNC public meetings on April 5.

What’s at stake here for everyone, particularly in TT, is an agreed on understanding of truth and falsehood.

TT has already seen in the US what blurring those lines can create.

Local media houses must continue to be diligent in their efforts to clear this underbrush of fakes, but tech-savvy citizens must also participate in the critically important mission of labelling and separating politically convenient AI lies from truth.