Text by Nicholas Diakopoulos
Have you ever toyed with generative AI models like ChatGPT, GPT-3, or DALL-E and wondered whether they could actually help you do some newswork? The technology is poised to disrupt many aspects of media and communication industries by making content — not only text, but also visual imagery — easier to create. For instance, these tools can be used to illustrate news stories with images or videos, extract structured data from unstructured documents, or do a range of writing tasks from simplifying text for different audiences to summarizing documents, writing potential headlines, and brainstorming angles or potential directions for reporting, like in the following request to ChatGPT:
The Challenge
I am a researcher interested in understanding the challenges facing journalism practitioners in implementing generative AI in the newsroom. If you have a creative idea for how to apply a generative AI model (e.g. ChatGPT, GPT-3, DALL-E, or something similar) to help with a journalistic task (e.g. related to editing, reporting, curation, distribution, etc.), please send me a short pitch of 200–300 words describing the idea. I will select a number of pitches and will offer to work with the selected projects over the period of one month as a technical consultant to help think through how to make the idea work and to evaluate it.
The pitch should motivate and describe the specific journalistic task you think generative AI can help with, why and how you think it will help, and how you’re going to test it to see if it works well enough that you could actually use it in practice. If you think there are particular challenges, ethical concerns, or other aspects you’re not sure about please describe these too. As you work on your pitch you should sign up for a free account with OpenAI’s ChatGPT, GPT-3, or DALL-E tools so you can test the initial feasibility of your idea. OpenAI makes available extensive documentation on their technologies or you can check out the Notebook I created which can help you understand the capabilities, limitations, and potential use cases. Originality, feasibility, and your plan for testing the reliability of your approach will be the key selection criteria for pitches.
Once selected you will have about a month to implement and evaluate the idea as a pilot / prototype using, for example, the free interfaces from OpenAI linked above. Coding is not required, but if you do code you can also implement your idea using available APIs as well as other open-sourced or fine-tuned models. As part of this work you will also write a ~1000 word article based on your experience and on the outcome of your effort to describe what you achieved, how you did it, what kinds of prompts you needed to write to control the AI, what you learned that might help others follow in your footsteps, and any cautions or other reflections you may have. Once complete and accepted, I will publish the articles on a blog for the project and Northwestern University will pay you $250 for your time and effort.
You can email pitches to me, Nick Diakopoulos, at nad@northwestern.edu with the subject “Generative AI in the Newsroom Challenge”. Pitches will be evaluated on a rolling basis until March 15, 2023.