Is ChatGPT detrimental to innovation? A field experiment among university students

Abstract

AbstractChatGPT represents a momentous technological breakthrough whose implications – along with other AI innovations – are yet to fully materialize. This paper is among the first attempts to experimentally test the effect of AI applications (in the form of ChatGPT) on three dependent variables usually assumed to be AI-collaterals: innovation, readiness to exert effort, and risk behaviour. We took advantage of the delayed introduction of ChatGPT in Egypt and conducted a pre-registered field experiment with nearly 100 senior university students at a public university. Over one month during term time, participants were asked to submit three graded essay assignments. In the treatment group, students were asked to write the essays using ChatGPT whereas in the control group, such option was neither mentioned nor allowed (the experiment was fielded before ChatGPT was legally operable in Egypt). One week after all assignments were submitted, the two groups were invited to the lab to play an innovation game (deploying multiple strategies to increase the sales of a hypothetical lemonade stand), a risk game (bomb risk elicitation task), and do a real effort task. The ChatGPT group was significantly less innovative, significantly less risk averse, and exerted less effort (however not statistically significant). Our results point to possible negative effects of AI applications but need further testing and larger samples to be confirmed.