How Generative AI Is Transforming Journalism: Development, Application and Ethics

Abstract

Generative artificial intelligence (GAI) is a technology based on algorithms, models, etc., that creates content such as text, audio, images, videos, and code. GAI is deeply integrated into journalism as tools, platforms and systems. However, GAI’s role in journalism dilutes the power of media professionals, changes traditional news production and poses ethical questions. This study attempts to systematically answer these ethical questions in specific journalistic practices from the perspectives of journalistic professionalism and epistemology. Building on the review of GAI’s development and application, this study identifies the responsibilities of news organizations, journalists and audiences, ensuring that they realize the potential of GAI while adhering to journalism professionalism and universal human values to avoid negative technological effects.

1. Introduction

Generative artificial intelligence (GAI) is a technology system based on algorithms, models, etc., designed to create text, audio, images, videos, and code. GAI includes various technologies and architectures to learn the underlying patterns and correlations in data, and subsequently generate content based on this acquired knowledge (Jovanovic and Campbell 2022). A research report published by Accenture in March 2023 shows that GAI ushers in a bold new future for science, business and society, with a profoundly positive impact on human creativity and productivity. Among business leaders, 98% of respondents agree AI foundation models will play an important role in their organization’s strategies over the next three to five years. Furthermore, as much as 40% of all working hours will be supported or augmented by language-based AI, such as GPT-4 (Harper et al. 2023).

GAI is currently used to assist and accelerate particular procedural tasks, such as content creation, fact-checking, data processing, image generation, speech conversion and translation, reducing the burden on human and increasing efficiency (Caswell et al. 2021; Nishal and Diakopoulos 2024). With the widespread adoption of GAI in news production, academia has begun to scrutinize GAI news production from three perspectives. The first line of study is comparing GAI-assisted journalism and traditional journalism (Hong and Tewksbury 2024; Wang et al. 2023). The second is to examine how GAI is used in journalism of specific countries or regions. For example, Gondwe (2023) explored how journalists in sub-Saharan Africa use GAI, while Pinto (2024) investigated GAI in the Brazilian news industry. The third is reflecting on ethical challenges and calling for upholding journalistic professionalism and its core values. Some scholars argue that journalistic professionalism is compromised by GAI, championing “Without journalists, there is no journalism” (Fernández et al. 2023). Some scholars express concern that the growing integration of GAI into news production may have a profound impact on public opinion and even influence the future of democracy (Spennemann 2023; Arguedas and Simon 2023). These three lines of study, whether emphasizing operational aspects or ethical concerns, isolate the examination of the ethical challenges posed by GAI in journalism and fail to contextualize these challenges within specific stages of news production for comprehensive investigation.

Therefore, this study attempts to discuss ethical challenges under the context of specific stages of news production. In addition to scrutinizing existing technological implementations, its goal is to offer theoretical perspectives that can guide researchers, professionals, and policymakers in leveraging the capabilities of GAI responsibly while upholding journalistic integrity and universal human values, thus mitigating adverse technological impacts.

2. Development of GAI in Journalism

AI has been used in the news industry for the past decade. The Associated Press (AP) was one of the earliest news media to integrate AI into news production (Radcliffe 2023). In 2014, the AP began using AI to process reports on corporate earnings. Before using AI, AP editors and journalists spent countless resources creating financial reports, which drew their focus away from news of greater significance. Despite considerable investment, the AP could only produce 300 financial reports per quarter, leaving thousands of potential corporate earnings reports unwritten. With the help of Wordsmith created by Automated Insights, now they are able to convert earnings data into publishable news stories within seconds; this is nearly 15 times more efficient than the traditional way (Miller 2015). Other media organizations including Bloomberg, Reuters, Forbes, The New York Times, The Washington Post, and the BBC also use AI for news production. The primary applications of AI in these large media organizations involve news gathering, production, and dissemination. In China, Tencent was the first to use AI for news writing. In September 2015, Tencent’s financial channel used a news-writing robot known as “Dreamwriter” for a news report titled “August CPI rose 2.0% year-on-year, hitting a new high in 12 months”. The content mainly consisted of data analysis and expert commentary on the data. Dreamwriter continued to release more reports, reaching 40,000 in the first three quarters of 2016 (Zhou 2015). On 18 November 2015, ** the roles and responsibilities of journalists and other media professionals. For example, traditionally, the information gathered by journalists should possess societal significance and humanistic considerations, reflecting the intricate dynamics between individuals and society, while also adhering to principles of fairness, objectivity, and rationality. If journalists are entirely replaced by machines, and it is left to machines to decide what information to gather as news material, there is a risk that machines may indiscriminately collect personal information or user data without human constraints, thus infringing on individual privacy. As a result, news production may violate basic human rights as journalism forsakes its fundamental purpose of serving society and instead takes on the role of mass surveillance (Dilmaghani et al. 2019; Andrew and Huang 2023; Guembe et al. 2022). With intelligent systems in place, users’ privacy is exposed, and the prerequisite for accessing intelligent services is the relinquishment of personal information.

Last year, June 29 witnessed sixteen individuals in the United States filing a lawsuit against ChatGPT. They accused the platform of collecting and revealing their personal information, such as account details, login credentials, emails, payment information, browsing history, social media interactions, chat logs, and other online activities, without adequately informing or obtaining consent from the users (Mauran 2023). Additionally, the restricted accuracy of GAI, particularly in comprehending intricate emotions, results in a growing prevalence of standardized news devoid of human touch in its delivery.

4.2. Content Production: Credibility

Credibility serves as not only a foundational principle in journalism but also a fundamental aspect of journalistic ethics. However, with the advent of the digital age, both scholars and professionals are reexamining the concept of “news credibility”. Alongside the digital transformation of the news industry, discussions surrounding the credibility of news have become increasingly prevalent. For instance, within the realm of digital news, discussions delve into nuanced concepts such as “experiential truth” within media technology, “perceived truth” within cognitive psychology, and “negotiated truth” concerning power dynamics. Despite the evolving definitions and discussions surrounding news credibility, its significance remains paramount in defining what qualifies as “news”.

In fact, truth is not absolute but falls within a measurable range. Thanks to technological advancements, journalism is now closer to truth than ever, sometimes even overly detailed and accurate, appearing clearer than reality itself. For example, the fact-checking ability of GAI is highly accurate—for example, ChatGPT achieves an accuracy rate of up to 68.79% (Hoes and Bermeo 2023). However, GAI’s usefulness in making fabrications convincingly authentic is concealing its detrimental risks of blurring the boundaries of reality. GAI, characterized by technical rationale, stealthily integrates mechanistic cognition rooted in technical logic and preconceived judgments into news production. Extreme technology supporters become adherents of technological superstition (Shipley and Williams 2023). For example, GAI has caused deepfake challenges, where Generative Adversarial Networks (GANs) are used for face swap**, lip synchronization, facial re-enactment, motion transfer, and more. GAI’s ability to generate fake content of individuals expressing opinions with specific emotions and even dancing has led to an upheaval in the news industry, making it more difficult for media professionals to ensure the authenticity of their content (Marconi and Daldrup 2018).

Moreover, the decreasing cost associated with producing deepfakes via GAI is exacerbated by the adoption of the Platform as a Service (PaaS) model in cloud services and cloud terminals. This trend intensifies the challenges posed by deepfake technology, amplifying its potential negative political ramifications. For example, a video features a two-minute-long conversation in which the leader of Progresívne Slovakia, Mr Michal Simecka, appears to discuss buying votes from the Roma minority with a journalist. AFP fact-checkers consulted several experts who concluded that the audio was synthesized by an AI tool trained on real samples of the speakers’ voices, and several copies are available on social media without a label marking them as misleading (Solon 2023). As Thomson et al. (2022) observe, “while the creation of synthetic media is not inherently problematic (for example, in the case of art, satire, or parody), issues emerge when those media are presented without transparency and masquerade as reality”. Indeed, outlets like The New York Times have begun labeling AI-generated images as such to highlight their provenance.

4.3. Customization and Dissemination: Value Orientation

News values encompass the qualities that enhance the significance or appeal of a story to the public. Consistent factors of news values include credibility and timeliness, while variables are prominence, impact, proximity, and interest. The depth and diversity of values embedded within news facts directly correlate with the overall value of the news report. Traditional media pursue truth and fulfill social responsibilities. Their news values include timeliness, relevance, and prominence. Caring about what the audience want, they also produce interesting news or news that matter to a given audience. With the development of GAI, power replaces news values, which means that whoever possesses the technology and controls the resource decides what is newsworthy. Deploying GAI requires substantial resources such as digital infrastructure and training corpora that are controlled by countries and enterprises with the most resources and power, leading to their monopoly. Consequently, the AI models prioritize their worldviews and perspectives, while their technological regulations and standards establish global norms, and their value systems become universal, ultimately causing a shift in news values towards power (McIntosh 2018; Acemoglu 2021). New Zealand’s data scientist David Rozado created a GAI model called RightWingGPT to express US right-wing political views. Research indicates that language models can subtly influence users’ values, highlighting the potential serious consequences of political biases in GAI models (Knight 2023). Other researchers use DALL-E and Stable Diffusion to convert text into images, revealing that these models perpetuate common stereotypes. For instance, prompts related to cleaners consistently generate images depicting women (Bianchi et al. 2023).

The inherent flaws in large language models directly affect the output of GAI, which relies on machine learning to mimic human behavior and generate content. This situation leads to systemic biases, value conflicts, cultural hegemony, stereotypes, and misinformation (Zhou et al. 2024). In addition, GAI models lack common sense reasoning, making it difficult to comprehend complex issues and distinguish nuances in tone and subtle emotions. Issues like AI hallucinations emerge, where generated content may seem plausible but contradicts real-world knowledge or existing data (Salvagno et al. 2023; Alkaissi and McFarlane 2023). As a result, GAI values shape the customization and dissemination of news, hindering comprehension of events, narrowing perspectives, intensifying polarized discourse, and eroding public discourse.

To facilitate an understanding of the study, a figure is shown here to depict the application of GAI in news production and the associated ethical challenges (Figure 1).

Lastly, it is imperative to discuss the legal ramifications stemming from the integration of GAI in news production, particularly concerning potential violations of individual privacy rights and copyrights. On 15 February 2023, Francesco Marconi, a Wall Street Journal reporter, publicly accused OpenAI of unauthorized use of content from prominent foreign media outlets such as Reuters, The New York Times, The Guardian, and the BBC to train its ChatGPT model, without compensating the original sources. Presently, prevailing legal frameworks predominantly attribute responsibility to individuals, which may prove insufficient in addressing GAI as digital entities. Legal accountability demands the establishment of a direct causal link between the AI operator and any infringement, with penalties commensurate with the actual harm inflicted. However, existing legal standards lack the necessary mechanisms to precisely identify and attribute damages, as well as engage in ongoing assessment and risk mitigation protocols to minimize adverse consequences. Moreover, the incorporation of GAI poses significant challenges to the protection of intellectual property. The question of whether AI-generated entities are entitled to intellectual property rights remains unresolved, although disputes over the rights of AI-generated entities have already arisen. Notably, in China, a defendant faced litigation for employing AI-generated images without the plaintiff’s authorization as illustrations for articles. The Bei**g Internet Court held that AI-generated images demonstrating human-like “originality” and intellectual input could be protected under copyright law as manifestations of creative intellectual endeavors.