ARTIFICIAL GOSSIP by Vilma Tiainen

If AI was asked what it thinks about you, what would it say? If different AIs had a conversation about you, what would the result be?

“Artificial gossip” showcases the communication between AIs which are having a “conversation” about me. It starts with one AI, the Google Ad personalisation AI, interpreting my behaviour and summarizing it into short assumptions about me. These were available through my Google account. I wrote a simple Scala program to choose three sets of three assumptions randomly.

Each set of assumptions went through the following process:

The assumptions in the set were first given as input to an AI text-to-image tool, generating an image. Then, the image was given as input to Google’s image labeling AI, which produced a list of possible labels (words that explain what is in the image) for the image. Three of the labels were randomly chosen with another Scala program I wrote. These chosen labels were given as input to the same text-to-image tool as before, generating another image.

This process created a back and forth between the two Google AIs and the text-to-image AI, which resembles a conversation. The three ”conversations” generated with this process from the three initial sets of assumptions are what constitutes “Artificial gossip”.

First conversation

Google ads AI
Construction & Maintenance, Marital Status: In a relationship, Shopping

Text-to-image AI