The problem here is not just that AI is shifting public opinion. A fifth of claims fed to users were rated predominantly inaccurate. So, pairing the automated models’ misinformation, their sheer persuasiveness, and the fact that 44% of US adults are already using tools like ChatGPT regularly, we must ponder some key points: how does this shape future elections, how can we control deliberate misinformation, and what if unscrupulous actors find a way to exploit these systems?
Shifting voters’ preferences by between two and fifteen percentage points is an effect so large it could flip almost any modern election. These models don’t rely on emotional manipulation. Instead, they bombard users with confident streams of information – some accurate, but much of it not. In fact, the most persuasive systems were also the least truthful.
In the US, when Trump supporters engaged with a pro-Harris chatbot, their support shifted 3.9 percentage points towards Harris. Harris supporters exposed to a pro-Trump model moved 2.3 points in the opposite direction. These swings are extraordinary in magnitude considering that most political advertisements and campaigns achieve far less than a one-point shift – often statistically indistinguishable from zero.
The 3.9-point shift in favour of Harris is four times greater than the measured effect of all political ads during the 2016 and 2020 elections....<<<Read More>>>...
