Our Correspondent, Noah Aderoju writes on how journalists can check AI bias
I once asked a journalist during a webinar if journalists need to be concerned about AI bias when using generative AI in their workflow. He answered that AI bias shouldn’t be a concern for journalists.
His reason was that whether the generated content is biased or not, a proper journalist’s human input will eliminate it and ensure it conforms to all necessary journalistic and ethical standards.
This is true for professional journalists who adhere to their ethical standards, unlike random content creators without such scruples. However, even if AI-generated content only forms the first draft of reports, and every misrepresentative or biased element is caught by a journalist’s keen eye and the rigorous editing process typical of an ideal newsroom, should journalists not be concerned with the AI bias phenomenon?
Should the presence of polluted, polarized, stereotyped, and unrepresentative information, which biased AI-generated content can contribute to, concern journalists? I believe the answer is a resounding YES. This is because journalists are tasked with providing factual and representative information for societal development, which includes debunking falsehoods in the form of misinformation and disinformation.
Journalists have long been information education and entertainment professionals, writers of the first draft of history and accountability agents in society, making them key custodians of the information ecosystem. This role predates the internet, citizen journalism, and social media content creation.
Now that generative AI is part of the information ecosystem, producing and generating definitive content to meet people’s information needs, its potential for generating biased content that entrenches stereotypes and misrepresentations of marginalized groups is a serious concern. Such bias undermines the work of journalists to maintain a responsible, factual, representative, and unbiased information ecosystem.
Although a proper journalist’s work won’t be affected by any bias in AI-generated content in the immediate sense, the broader ecosystem, influenced by unchecked AI-generated content shared by people who may not be journalists, can be compromised. This is similar to how misinformation and disinformation currently threaten factual information in our society. Generative AI content bias could become the next challenge for journalists to combat.
Generative AI is one of the most accessible and widely adopted AI tools today. Google uses generative AI to answer search queries, integrating its Gemini AI into Google Assistant on Android phones. Meta AI is available on WhatsApp, providing answers and generating content for millions of users. OpenAI’s ChatGPT, the most popular of them all, is now more sophisticated and accessible with its new GPT-4o model. These models are built on billions of pieces of content that aren’t transparently vetted, carrying inherent biases that manifest in various ways.
How can journalists help reduce the bias in generative AI content? The solution begins with understanding the origin of the problem. AI bias originates from the training data of the model, composed of billions of unclassified pieces of content from various genres of information, without regard for fact or fiction, authority or speculation, or outright fabrication. The algorithm responsible for ranking content for the model to learn from, which could check bias, remains opaque and unknown to outsiders.
There are several ways to mitigate AI bias, including using diverse and representative datasets, comprehensive testing, ensuring transparency and accountability in engineering factors like coding and algorithm development, and constant monitoring for necessary corrections.
Journalists can contribute to these solutions through public education, advocacy for ethical best practices, and the development of policies for accountability. Additionally, journalists can actively facilitate the provision of diverse and representative datasets and ensure transparency and accountability in AI models’ algorithms through watchdog journalism.
As content producers, more than ever, journalists now need to focus on creating more representative content compatible with digital media to train AI models. They should spotlight underreported issues to produce more representative content online, take more representative pictures and videos to provide factual, unbiased, and empirical content, and adequately describe and meta-tag audio and video content for easy machine and algorithm ranking and understanding.
Media groups can even build factual, unbiased, representative, and responsible datasets for AI model developers to use. This can even be a source of revenue for the media sector. The years of detailed coverage of a subject matter or a community with impeccable standards that a media house has can beat many subpar datasets that AI model builders are using. Investigative reporting on AI model development can help spotlight their faults and scrutinize the lack of transparency and responsibility by their developers.
This is why journalism, especially in this part of the world, should go beyond reporting announcements, press releases, and expert opinions about AI capabilities. Journalists should investigate the processes, data sources, and methodologies used by companies developing AI systems, exposing biases in existing AI systems and their impacts on different communities.