We’ve just had a great roundtable of our Meta Content Library community of practice on this Saturday morning, so we’re already down to the final paper session of what has been another fabulous Association of Internet Researchers conference in Mexico City. The first presenter in this session on journalism and its challenges is Gabriel Ponniah, whose focus is on US journalism’s reporting on generative AI during 2022-23.
ChatGPT burst into public consciousness during this time, and journalistic framing of emerging technologies has a significant impact of public understanding of such technologies; so, how did US journalism do? This project scraped news content from various tech journalism sites from 1 Nov. 2022 to 31 Oct. 2023, resulting in a dataset of some 2,000 articles that directly dealt with AI.
Key themes in this were economics, arts and entertainment, US politics, the tech industry, geopolitics, regulation, and education; some of this was also driven by external factors such as the Hollywood strike related to AI use in the industry – but environmental impacts of AI were notably absent from the coverage.
Major frames in such coverage, each divided into positive and negative subframes, included cultural generativity (relating to creative opportunities, copyright, and IP), labour, information use, surveillance, and the human condition (from transcendence and evolution to an existential crisis). Some of this was coded using AI tools, but an in-depth manual analysis of over 300 articles was also undertaken.
Cultural generativity frames ranged from the dazzling opportunities in using generative AI to the threat posed by AI to human creators; labour implications (which were most prevalent) ranged from improved productivity to (more prominently) the devaluation of human labour; information use ranged from opportunities for information synthesis and accessibility to AI errors and misleading and disinformative uses; surveillance frames were less prominent, and ranged from better personalisation and customisation of technology use to AI undermining privacy; and human condition frames (also less prominent) included flowery language about how AI might enable humanity to transcend and evolve, but also AI as an existential threat to humanity.
Environment and climate debates were largely absent from press coverage. For all frames, the negative subframe tended to be more prevalent, but both tended to be present – this points to the deep-set bothsidesism in such coverage. Coverage was top-level and audience-centred, with systemic issues (design, climate, finance, etc.) largely ignored in this first year of the coverage of generative AI. Discussions of AI regulation were also absent.
Coding this content using the AI system DeepSeek was not successful, incidentally – it failed to reach sufficient intercoder agreement.
