As in many other industries, the role that AI technology will play in the future of journalism is uncertain. In the past year, several news organizations have been caught using generative AI in questionable ways, and generative AI routinely produces subpar results when given basic reporting prompts, such as summarizing sporting events using box score data.
Yet some of the most reputable news outlets and journalism schools have also produced thoughtful explorations of the ways in which technology can be employed responsibly to assist in the production of news. So while it’s unlikely that hyperbolic predictions about reporters being fully replaced by machines are likely to materialize in the near (or even distant) future, it is worth trying to sketch out some of the key factors that the next generation of journalists currently being trained in communications programs will have to consider regarding the role of AI in their jobs.
Some recent unflattering headlines have certainly provided reason for skepticism about the incorporation of AI in journalism. The behemoth publisher Gannett (which owns, among many other publications, USA Today), for instance, ended up suspending its use of AI-generated sports recaps after receiving widespread ridicule for their low quality.
In a more problematic development, company was then suspected of attempting to disguise its use of generative AI technology to write reviews on Reviewed, its publication that focuses on product reviews. The real human staffers at the website began noticing published articles from authors whose names they did not recognize, and upon investigation, they found no other indication that the people in question existed.
(Gannett, for the record, has denied fabricating the authors or even using generative AI for the posts, insisting simply that the copy supplied by freelancers hired by a marketing company that Gannett had contracted with “did not meet our editorial standards.”)
The problems with these endeavors are threefold. First, in the case of something like product reviews, while some might argue that even fake reviews can still put forth informational content, the distinct value of this genre of writing arguably lies in the notion that they represent the authentic sentiments of an actual human.
Second, in this case at least, the deception went one step further, with an attempt to obscure the genesis of the reviews by attributing them to real “journalists.” While caught in this instance, such deception underscores how even the appearance of “humanness” can be simulated online. While a typical media literacy recommendation would be to look for the attribution to a human journalist, we therefore must now dig deeper into the online footprint of any such person referenced in order to be sure.
Finally, beyond the issues of deception, critics highlighted the compromising of the content itself that resulted from the use of generative AI. the reviews were simply “sterile and unengaging,” and displayed a kind of formulaic structure and cadence that has been identified in a variety of LLM-generated output, not to mention now-infamous words like “delve” that have become Chat GPT cliches.
When these kinds of slapdash efforts come to light, it can leave one with the impression that AI technology is an inherently negative force for journalism — simply another means for corporations to cut costs and extract higher profits while delivering an inferior product, as has been a trend for some time in the digital age. But leading news publications and journalism schools have also been thinking seriously about the productive ways in which AI technology can be incorporated — or already is — in responsible journalism.
A recent report from the London School of Economics “JournalismAI” initiative details how newsrooms already are using AI technology to assist in the gathering, production, and distribution of news. For the newsgathering process, for instance, newsrooms surveyed by the LSE researchers described using automated transcription technologies like OtterAI to accelerate the processing of interview data. They also already use software like Crowdtangle and Rapidminer to sift through a high volume of social media discussion to detect trends and determine directions for reporting.
For news production, the LSE report notes that such tools can be useful for fact-checking purposes, as they can be trained to analyze language usage and detect when a factual claim has been made in a piece of writing that could warrant checking. Additionally, while wholesale replacement of human story production has proven ill-advised so far, a number of respondents in the LSE survey noted that generative AI tools can be helpful in producing code to expedite the creation of interactive features (such as games or quizzes) that augment a piece of otherwise original reporting.
Finally, respondents also mentioned several ways in which even the kinds of generative text approaches that appeared problematic in the Gannett example above can be harnessed for good. For instance, Bloomberg has created “BloombergGPT,” a large language model (LLM) trained on financial data that can summarize financial documents, and the Washington Post has an application called “Heliograf,” which similarly produces summaries from “structured data” such as financial reports.
As the report stresses, however, these tools are not meant to be a replacement for human authors. They simply allow the human authors to spend their time on more substantive analysis and generally “focus more on in-depth reporting.”
In all likelihood, then, the young journalists currently being trained in communications schools across the world can expect to use AI tools in their work, if they do not already. But it will be imperative for them to be discerning about what these tools can and cannot do. If the above examples are any indication, LLMs and other AI tools can absolutely be helpful in obviating various rote tasks and sifting through large amounts of information quickly. What they cannot and will not be, however, is a substitute for human judgment about the meaning of this information and how to responsibly and coherently present it to the public.
Benjamin A. Medeiros 