This tool detects media bias using OpenAI's GPT-4

Knowledge at Wharton

Ask about Knowledge at Wharton

Episode Description

Concerns about news coverage around events like an election often focus on false information. But what about news that is factually accurate, yet biased?

Enter the Media Bias Detector, the latest product of Wharton professor Duncan Watts' years of research into media coverage and consumption, created to address one of Watts' key findings: that biased news can be more dangerous than "fake news."

The free tool tracks and analyzes articles from top publishers in close to real time, providing, in Watts' words, "regular people with a powerful, useful resource to better understand how major events, like this election, are being reported on."

The popularization of @OpenAI's ChatGPT provided a breakthrough moment for the CSSLab at Penn – a joint venture between Wharton, Penn Engineering and the Annenberg School for Communication – to develop and launch the Detector, as the lab's researchers designed it around OpenAI's GPT infrastructure.

Learn more and explore the tool: http://whr.tn/mediabiasdetector

Read the full transcriptHide the full transcript
  1. [Music] everybody knows that misinformation is a big problem in the world right now but
  2. when we talk about misinformation we almost always mean false information our work leads us to believe that in fact
  3. bias can be a much bigger problem so that's what this project is really intended to uncover developed by the CSS
  4. Lab at Penn a joint venture between Warden pen engineering and the annenburg school for community ation the media
  5. bias detector tracks and classifies the top stories published by 10 prominent Publishers spending the political
  6. Spectrum in close to real time articles are fed into GPT 4 the large language model developed by open AI that
  7. underlies Chad GPT and classified by topics and events news type tone and political lean every week research
  8. assistants then monitor a random sample of the gpt's labels to ensure the model's accuracy remains high and
  9. identify any necessary adjustments there are two highlevel ways in which uh we think bias can manifest
  10. itself in journalism one is selection which stories journalists choose to write about which facts they choose to
  11. highlight which people they choose to quote and also framing how they spin the evidence that they present to you you
  12. can make your tone positive you can make your tone negative and both these forms
  13. of bias can have lot of impact on the way the story reads and on the takeaway that the reader gets from the story even
  14. to the point where one can get a highly misleading or even false impression of what is going on even though every
  15. sentence in the story might be correct this year's US presidential election presents a real-time high stakes
  16. environment to observe and analyze the choices media Publications are making every day in their coverage the media
  17. bias detector is a free resource designed to help you as a consumer better understand those choices and the
  18. biases that exist behind them [Music]

Brief Summary

The episode covers media bias and the Media Bias Detector project at Penn.

Key points

  • Understanding Media Bias. This project uncovers how bias manifests in journalism, beyond just misinformation.
  • The Media Bias Detector. A tool to help consumers understand media choices and biases in real-time.

Related episodes