Between time savings and additional effort: Generative AI in the workplace 

AI-based chatbots are transforming the workplace significantly. They assist in information retrieval, analysis, and text creation, promising substantial productivity gains. Consequently, generative AI is often portrayed in the media as a catalyst for employee productivity. In practice, however, employees experience it in contradictory ways. While ChatGPT saves time on research, for example, it also requires additional work for fact-checking. This blog post explores the paradoxical impact of generative AI in the workplace.

AI-based chatbots such as Open AI’s GPT-4 are highly versatile. They can be used for searching, analysing, and summarising information, as well as generating and editing ideas, texts, and codes. The technological maturity of these AI applications and their wide range of uses make them attractive to both companies and employees. Yet, opinions diverge when it comes to assessing their impact.

Generative AI as a productivity catalyst

A few months after the hype around ChatGPT began, initial working papers were published, attesting to the program’s significant potential to boost employee productivity (Brynjolfsson et al., 2023; Noy & Zhang, 2023). These and similar studies have led to a media narrative depicting generative AI as a productivity catalyst, reducing workload and accelerating processes for employees. Headlines claim that AI boosts worker productivity by 14 percent, and another article compares the anticipated productivity gains from generative AI to those of the Industrial Revolution. However, these articles often overlook that the underlying studies are based on experiments that do not necessarily reflect real working conditions and are focused on specific professions (such as customer service).

Generative AI as a catalyst for additional work and bureaucracy

Contrary to the dominant media narrative, critical voices emerged soon after the release of ChatGPT. In an article for The Atlantic, Ian Bogost argues that ChatGPT burdens us with more work. On top of all our existing tasks, we now need to spend time distinguishing between human and AI-generated content and dealing with the associated bureaucracy. Numerous examples illustrate this point. Universities must decide how to handle AI-generated student papers, academic journals must determine the extent to which AI can be used in submissions, and societies must address AI-generated misinformation.

Paradoxes in the context of technologies

Understanding these contradictory perspectives can be aided by a paradox perspective. In the context of technological change, a paradox refers to inherent contradictions, dilemmas, or unexpected consequences arising from the use and development of technology (Mick & Fournier, 1998). A paradox means that technologies can simultaneously produce X and -X, such as fostering independence while leading to dependency, conveying the feeling of intelligence alongside the feeling of ignorance, bringing people together while also causing isolation. A prime example of this is the smartphone. It facilitates staying in touch with friends and family yet hinders deep connections during in-person interactions.

Paradoxes in working with ChatGPT

A recent study involving 48 interviews with advertising industry professionals explored how ChatGPT impacts their work. The findings reveal that employees experience ChatGPT’s impact as contradictory, identifying three paradoxes in their interaction with the program (Osadchaya et al., 2024). First, ChatGPT both facilitates and obstructs research. While employees find AI-based chatbots such as ChatGPT more effective than search engines, they also bemoan the extra effort required to verify the answers. Second, AI-based chatbots promote both creativity and standardisation. Though respondents use ChatGPT beneficially in the creative process, they also observe an increasing homogenisation of advertising messages. Third, from the respondents’ perspective, using ChatGPT leads to both efficiency gains and losses. You can quickly generate a blog post such as this one, but then you have to spend a lot of time revising it, both validating facts and adjusting the tone to avoid the text from sounding too AI-generated.

Outlook

My own preliminary analyses of online forum discussions among creatives in the advertising industry align with the study’s findings. This does not imply that generative AI cannot enhance employee productivity. However, the productivity effects may have been overestimated in initial studies regarding actual work conditions. Understanding why this is the case and when negative impacts outweigh positive ones remains an open question, which requires further research that considers the application context more thoroughly. For this reason, our project “Generative AI in the Workplace” investigates both the applications and impacts of generative AI in selected professional fields such as human resources and marketing.

References

Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative AI at Work. https://www.nber.org/papers/w31161

Mick, D. G., & Fournier, S. (1998). Paradoxes of Technology: Consumer Cognizance, Emotions, and Coping Strategies. Journal of Consumer Research, 25(2), 123-143. https://doi.org/10.1086/209531

Noy, S., & Zhang, W. (2023). Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. Available at SSRN 4375283. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4375283

Osadchaya, E., Marder, B., Yule, J. A., Yau, A., Lavertu, L., Stylos, N., Oliver, S., Angell, R., Regt, A. d., Gao, L., Qi, K., Zhang, W. Z., Zhang, Y., Li, J., & AlRabiah, S. (2024). To ChatGPT, or not to ChatGPT: Navigating the paradoxes of generative AI in the advertising industry. Business Horizons. https://doi.org/10.1016/j.bushor.2024.05.002