Building tools for journalism and civic information by prototyping with open source AI
By Burt Herman / April 17, 2024
Updates from the open source AI hackathon
More than 100 journalists, coders, and product designers spent a weekend together in New York earlier in April to participate in the first of a series of hackathons convened designed to push the limits of the technology for journalism.
To kick off the series, hackathon participants spent three days exploring how AI might enable new, personalized experiences for journalism audiences, how to train large language models to increase accuracy and trust, and how to design platforms that ensure a human is always kept in the loop.
Organized by Hacks/Hackers, together with Brown Institute for Media Innovation at Columbia University, and with support from Hugging Face and Codingscape, the three-day event focused on experimenting and prototyping with open source AI to build tools for journalism and civic information.
Starting with a pitch session and mixer, hackathon participants collaborated for the rest of the weekend before giving demos of what they built.
Ultimately, the goal of the weekend hackathon was to give journalists and technologists a space to experiment with AI without any of the usual pressure to launch a fine-tuned product. Instead, hackathon participants were encouraged to push the boundaries of a new technology to find new ways for journalists to reach audiences.
Exploring and experimenting with how AI tools can deliver journalism to audiences in new ways
The hundred or so attendees split up into an eventual 12 teams to work on prototypes and ideas for how to combine AI with journalism.
As Yona TR Goldberg reported in Columbia Journalism Review, there were generally two categories or themes of hackathon projects over the course of the weekend.
One general approach adopted by hackathon participants over the course of the weekend explored how generative AI might be used as a journalistic or investigative tool. A second general theme of the hackathon was experimenting with using these new tools to deliver media to audiences in new ways.
For example, one team that flew in all the way from The Telegraph in London to participate in the Hackathon worked on using language models to more creatively tag articles in ways that could help power new news products. The idea was to build new content feeds that would give people new things to explore, something to learn, and something to do.
Another team came up with an idea called ProxiEdit to have AI offer different perspectives on a story before it’s published to help writers get more feedback and improve their work. Yet another team worked on an AI-infused editor interface that would pull in relevant parts of interviews to help enrich a story. Prompts and Provenance, another project, focused on image generation models in seeking to help create images more authentic to their actual geographic location. Meanwhile, Source Detector used AI to analyze sources cited by journalists in stories to detect potential bias in articles.
An opportunity to experiment and figure out what works, and what doesn’t
A weekend is both a long time and not very much time to build a product. While there has been much speculation about the potential of AI, the best way to see what’s possible is by using the technology in different ways and seeing what works and what doesn’t.
Ultimately the hackathon gave journalists and technologists the space to collaborate and experiment with AI without any of the usual pressure or complications of launching a finished product, while at the same time pushing the boundaries of the technology.
