While discussing AI, the topic of fake (generated) news inevitably comes up.
I initially wanted to create an artwork to discuss fake news, bubbles and echo chambers, to see how convincing fake news generated by AI can be.
The idea of a fake news website that generated a headline, image and a complete article from real news headline came to me. So I made a design mockup of what this fake news website could look like.
After coming up with this idea I pitched it to the class, sort of like it was a business proposal. The feedback I received on this was that the truth is vague in this work And I could take more risks with that. Also it might be more interesting to choose on thing to focus on. For example; just the news titles.
For the next meeting we were tasked with creating a physical artwork to show.
I wanted to make a continuation of this artwork since it felt quite unfinished to me. I came up with the idea to create a fake newspaper completely filled with AI generated articles. Since one of the topics I wanted to touch upon is information bubbles; I also wanted to see if my work group would notice that the articles are all based upon sentences said within class. Our own personal echo chamber.
Rather than making a fake-news source and discussing it ourselves I came to the conclusion that it would be much more interesting to see what AI thinks about this topic. Having just one AI ‘speaker / actor’ however turned out to be a bit dull, and has already been done in class. If you have a problem, make it worse; so I chose to use 2 different computers with AI voices to have a conversation about the news paper among each other.
Sources, tools and Methods:
To write the articles and the discussion used by the AI voices I used GPT-J-6B. A version of GPT2 that performs really well in ‘keeping the conversation onto a single subject’.
For the TTS I used 15.ai with the Wheatley and GLaDOS voices from portal. 15.ai uses a modified version of the Tacotron speech synthesis algorithm. Which is great at turning a small amount of recorded voice into TTS.
To create the visuals for the robot voices I used geometry nodes in Blender.
I followed and modified this tutorial to get the desired look.