On September 14, 2023 I spoke at Tūranga’s Election Special session “Is Artificial Intelligence Dangerous for Democracy?”. The event also featured Professor Donald Matheson and Dr Zita Joyce from UC Media and Communications and some of my former students from UC’s Master of Applied Data Science Programme. The students introduced generative AI for audience and provided examples of the technology.
In my talk I discussed the newest technologies we talk about as ‘AI’, what’s new and not new about them in relation to elections, and suggested some ways to cut through political messages coming our way during the election campaign.
To illustrate the potential of generative AI, I generated a fake news article, which proposed Jacinda Ardern had resigned politics to pursue a music career. The generated article, entitled “Ardern’s Rock n Roll Dream: New Zealand Prime Minister Resigns to Pursue Passion for Hard Rock”, featured text generated by ChatGPT and an image generated by MidJourney.
Here’s an excerpt (click the image to read the full story):
Wearing a leather jacket and a band t-shirt, she spoke passionately about her love for music.
“I’ve always had a burning desire to rock out, and I believe now is the time to chase that dream”
Generative AI can certainly be used to produce false or misleading political messages, including imagery designed to influence hearts and minds. I suggested concern about the use of AI to generate disinformation and misinformation required some perspective.
Producing misleading political messages does not require generative AI. For example, while generating imagery that aligns a political leader with Christian-nationalism is straightforward (as the examples below illustrate), this kind of imagery is already produced by political supporters and choreographed by politicians. Generative AI does certainly provide possibilities to scale up the volume of unique messages and add to the noise around elections, but noise is already a feature of online communication.
Established political actors already present their messages in selective and partial ways and obscure the origins of their political messages. Online communications designed to influence through deception are not new. One of the most glaring examples from recent elections was the UK Conservative party’s rebranding of their Twitter account during a 2019 debate to resemble a factchecking service.
Concerns about misleading messages should best be understood as a continuation of longer-term problems with social media platforms and online communication. It is possible to obscure who is behind a specific political message we encounter online. Is this “real” or is it “a bot”?
To respond to this, I proposed that one of the most profound things we can do in the context of an election is ignore the noise. The last billboard we passed, the last thing we heard a politician say, and the last social media post we read - these are the background noise of an election. As citizens we have agency and way to use that is to understand the ideas and interests driving New Zealand’s political parties and what they’ve done in the past when they’ve had power. This might mean visiting a library and reading some books! Understanding these ideas, interests and histories provides a way to tune out the perpetual hot takes of political commentators and trending social media posts. Knowledge about the ideas of political parties, their interests, and their histories provides a way to make sense of and question the messages coming our way.
To help reduce the noise, I suggested we should recognise when we are mindlessly repeating the attack lines of politicians and interest groups. Just as ChatGPT is trained to regurgitate text by ‘reading’ many examples, there is a risk we let ourselves be trained too. We’ve got a choice about whether to be a bot or an AI. We can engage and participate in an election without being unthinking political text regurgitators, perhaps even ready to question the parties and ideas we hold dear.