In the space of less than 2 years, an artificial intelligence (AI) revolution has swept across multiple sectors and industries, radically transforming how people work in a way that is comparable to the impact of personal computers or the internet.
Like most areas of media, the news business is adapting to a world in which generative AI can perform many of the basic tasks of the job, leaving many concerned about the technology’s effect on truthfulness and journalistic integrity.
Not so long ago, the idea of AI models writing news reports would have sounded like science fiction. But in May 2023, NewsGuard identified 49 news and information sites that “appear to be almost entirely written by artificial intelligence software.” By December, that number had ballooned to 687.
The trend is moving in the direction of more AI-generated content in news media and beyond. In 2022, Futurism reported that “experts” had estimated that as much as 90% of online content may be artificially generated by 2026. The statistic was erroneously attributed to Europol, but in an ironic demonstration of just how easily fake news can spread around the internet, has since been parroted by mainstream media outlets including Forbes and Wired .
Regardless of the veracity of Futurism’s claim, it appeals to real fears over AI’s effect on news reporting. A recent survey found that 79% of journalists expressed concerns about the impacts of AI on the quality of journalism, a widespread anxiety that is compounded by the rise of artificially generated fake news.
Some AI-generated news is clearly propaganda. The use of deepfakes and social media bots to further Chinese, Iranian and Russian state interests has been well-documented. What’s more, in the age-old arena of information warfare, the latest weapons have been applied at all levels of political rivalry.
In the context of an American democracy already suffering from an erosion of trust in politicians and media outlets, the threat has proved alarming enough to instigate a string of bills attempting to curb the use of generative AI in electioneering.
Journalists and policymakers alike are understandably concerned about disinformation campaigns powered by artificially generated content. But other instances of Large Language Model (LLM) inaccuracies emerge without malicious intent.
In a 2023 experimen t, the researcher Matthew Hillier prompted ChatGPT to produce an output containing APA references. The chatbot completed the task with 6 citations. But of these, Hillier found that 5 were inaccurate, mixing up authors, page numbers and dates, or referring to material that didn’t exist at all.
For news writers, the preponderance of such “hallucinations” continues to hold back AI’s usefulness as a research tool.
As the computer scientist Daniel Huynh has observed , “models producing wrong outputs, are the biggest obstacle to LLM adoption.”
“Humans will not be able to consistently offload work to AI systems until they are able to trust the output of the LLM to be correct,” he added.
But while LLMs have some way to go before they can be reliably used as a general-purpose research tool, they can still be a powerful aide for journalists.
When AI assistants are directed toward strictly defined and well-organized data sets, the rate of inaccuracies is greatly reduced.
It is now easier than ever to integrate AI into everyday workflows thanks to the rise of dedicated chatbot plugins for popular software platforms like Microsoft Office and Google Workspace. Using these and other integration techniques, prompting LLMs to sort, search and filter data is one of the most powerful journalistic use cases for the technology developed so far.
Yet, according to one survey, only 1 in 4 journalists have used generative AI in their work. Podcasters were found to have the highest rate of usage, with 38% reporting that they had done so. Television journalists reported the lowest AI use, with just 12% of respondents having done so.
Among those who have embraced the technology, 16% said they had used it for research, 13% said they had used it to generate headline ideas and 10% said they had used AI to summarize or edit content. Only 2% reported that their newsroom was using LLMs to generate full articles.
Besides transforming the way journalists work, AI is also impacting how people access news. Design choices made during model training impact how much LLM’s learn from various news outlets. In turn, this shapes what information chatbot users receive.
Some content creators, however, have claimed that AI developers’ use of their material as training data amounts to intellectual property theft.
In January, the New York Times filed a lawsuit against OpenAI alleging copyright infringement based on outputs generated by ChatGPT.
According to the complaint, OpenAI’s LLMs were trained using millions of the Times’ articles without the publisher’s consent. As a result, ChatGPT outputs often provide users with “near-verbatim” excerpts from articles, the newspaper claimed.
The firm rejected the claims, accusing the Times of using manipulative prompts to generate the alleged AI plagiarism.
Regardless of which direction it ultimately goes, as the case unfolds, it could set an important precedent in the nascent field of AI copyright law, establishing the respective rights of journalists and AI developers for years to come.