Botcast
Audio Installation at ZKM’s Biomedia Exhibition
Developed by Yannick Hoffman & Yasha Jain.
An AI-generated speculative radio project
Botcasts – a portmanteau of “bot” and “podcast” – is an experimental radio format that transforms real-world headlines into AI-generated speculative fiction. Drawing from current news articles spanning politics, culture, and entertainment, the system creates fictional audio broadcasts that extend or distort reality, exploring the unsettling boundary between information, disinformation, and imagination.
Concept
In an era where so-called “alternative facts” have become tools of disinformation and political manipulation, and where over half of global internet traffic is generated by bots, Botcasts investigates the disruptive and infiltrative nature of automated media. The work poses questions about the future of journalism, authorship, and trust in an increasingly synthetic information ecosystem.
By converting legitimate news into speculative narratives, Botcasts simulates a media landscape where fiction and fact become indistinguishable—highlighting how easily perception can be shaped by algorithms.
Technical Write-Up
Workflow:
Botcast interacts with current events in real time. The workflow is:
- Browse Google news for trending news today.
- Generate a list and summary of the top trending articles of the day.
- These headlines are converted into a news broadcast using AI-generated speculative narratives.
- The text is converted into a news broadcast style text and spoken in the voice of Yasha Jain and Yannuck Hoffman using AI voice generation.
This process allows Botcast to blur the line between real-world information and AI-generated fiction, creating a constantly evolving, immersive media experience.
News Processing:
Articles are automatically scraped and summarized using the newspaper3k Python module.
Text Generation:
The summaries are then used as prompts for the GPT-Neo model, which generates fictional news stories.
Text To Speech:
We generated speech using Descript since the quality of voice cloning is not very good at emoting.
Model Choice:
GPT-Neo was selected for its open-source nature and its ability to produce coherent, contextually rich output. Other models like aitextgen were tested but produced less satisfactory results.
The generated content is formatted as a podcast-style radio broadcast, simulating the tone and cadence of traditional news media while presenting entirely fabricated narratives.