Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence

The article introduces the impact of generative artificial intelligence (AI) on production and labor markets, highlighting its potential to replace or complement human workers. The authors note that recent advances in generative AI, such as ChatGPT and DALL-E, can automate creative tasks like writing and image generation. However, the impact of generative AI on productivity, wages, employment, and inequality is unknown. To address these questions, the authors conducted an online experiment with 444 experienced college-educated professionals, asking them to complete two occupation-specific writing tasks. Half of the participants were told to use ChatGPT for the second task while the other half used a LaTeX editor. The authors collected data on output quality, time taken, job satisfaction, self-efficacy, and beliefs about automation. They found that ChatGPT had mixed effects on productivity, being beneficial for some subcomponents of complex tasks but hindering others. They also found that ChatGPT usage did not significantly impact job satisfaction or self-efficacy, suggesting that the technology did not have a negative psychological impact on workers. Nonetheless, the authors acknowledge that their study has limitations, and further research is needed to examine the impact of generative AI on a broader range of occupations and tasks.