KU researcher proposes techniques for communicators to recognize, avoid inherent biases in AI

LAWRENCE — People worldwide are learning to use and interact with artificial intelligence in their lives and work. This emerging technology has deeply affected several domains, including mass communications. 

Today, generative AI is used by some professionals to create first drafts of news stories, corporate communications, marketing materials and other works. A University of Kansas scholar has written a chapter on how communications professionals can recognize AI’s inherent biases and avoid perpetuating them in their work.

AI tools, more specifically large language models, are trained on human work — but not the work of everyone. Mugur Geana, professor of journalism & mass communications, contributed a chapter on what information AI is trained on and what is left out for the book “Sustainable AI — AI for Sustainability: A Reader for PR, Communications, Public Affairs & Sustainability Professionals,” published by Quadriga University of Applied Sciences in Berlin, Germany. The book is free to download.

“What is epistemic bias, and how is that reflected in the way AI is trained? Some communities’ stories, specific currents, ideological or philosophical human experiences are omitted because they have never been digitized,” Geana said. “The corpus that we use to train AI either does not have access to all these sources of knowledge or decides to ignore them. I think that’s a significant problem that AI has today, from multiple perspectives that I talk about in this chapter.”

Geana uses University of London scholar Miranda Fricker’s account of epistemic injustice to analyze how AI, driven by potential training fallacies, may enable and perpetuate harm. Acknowledging that knowledge gaps exist in AI models is key, as is recognizing that they can be spotted and should be addressed; Geana argues this and provides a checklist for communicators to add to any AI output to help them avoid bias traps. 

The checklist contains four dimensions: 

• Testimonial credibility
• Interpretive completeness
• Representation (visual and narrative)
• Institutional accountability. 

It includes questions communicators can ask to identify potential biases, warning signs and actions to avoid them.

The book, “Sustainable AI – AI for Sustainability,” examines the technology as a deeply complex problem that affects people and societies differently, with no clear solution in a time of crisis. AI emerged as the world experienced war, polarized politics, social and economic division, inequality and other national and international challenges. The authors provide guidelines for consideration and inquiry.

“We try to address the topic of sustainable AI and AI for sustainability without expecting to propose an ultimate solution. That really attracted me to the book because it offers a series of perspectives on what AI is and its impact on strategic communication. Concomitantly, we ask what we can do to certify the epistemic veracity of its outputs and how we can address some of the multiple challenges that communicators face when using AI,” Geana said. “At the KU School of Journalism & Mass Communications, at least for now, we embrace the approach of human first, human last and that AI should not be anything more than a tool. It learns from human output, which is flawed, so everything that AI outputs has to be critically checked, because it’s a synthetic entity that’s ultimately not responsible for its actions.”

The book, edited by Ana Adi, Alexander Morton and Thomas Stoeckle, features contributions from authors from around the world and is touted as an AI experiment in publishing. The editors created a standardized rubric for disclosing AI use. Some authors used the tech as a research assistant, others used it to translate from their first language into UK English, and others asked AI to draft a first version of the text before editing it themselves.

The text is not intended to cheerlead for or warn against AI but to guide communications and sustainability practitioners on how to use it responsibly as a tool, according to the authors.

For his part, Geana said he feels the technology is progressing in a positive way to address epistemic and systemic biases, writing in his chapter’s conclusion, “The task for communicators is not to abandon the tools but to stop treating their fluency as evidence of their accuracy, and to build the habit of asking, systematically, whose voice and knowledge made it onto the page.”