Befriending ChatGPT and other superchatbots: An AI-integrated take-home assessment preserving integrity

TL;DR

This study proposes an AI-integrated take-home assessment method that involves students using ChatGPT as a primary source, encouraging critical engagement and reducing academic dishonesty.

Contribution

The paper introduces a novel assessment technique that explicitly incorporates ChatGPT, promoting critical thinking and skepticism, thereby maintaining academic integrity.

Findings

No significant answer overlap among groups.

Wide variety of ChatGPT handling approaches.

Enhanced student skepticism towards AI responses.

Abstract

With the launch of ChatGPT, serious concerns have reasonably been raised of its ill-effect on the integrity of remote take-home exams. By way of mitigating the concern, in this study, a rather straightforward Artificial-Intelligence (AI)-integrated take-home assessment technique is proposed, and the outcome of its practice is discussed. Despite involving AI, in the form of ChatGPT, the assessment adheres to the convention of posing questions invoking critical thinking and problem solving skills. However, AI is characteristically integrated in this assessment by instructing the learners to employ ChatGPT as one of the primary sources. The learners are directed to report the use of ChatGPT by including both the prompts and its responses, before expressing their thoughts on AI-generated answers and their own concluding statement. These three characteristic components of the present techniques -- the handling of ChatGPT through the prompts, comments on the AI-responses and the concluding thoughts -- are evaluated to gauge the learning. The proposed assessment was assigned as a take-home group activity for a batch of seventy eight students, divided into thirteen groups. Despite addressing the same questions, there was no significant overlap in the answers. Moreover, a wide range of approaches were adopted by the groups in handling ChatGPT, which in-turn rendered different responses, ultimately drawing distinct answers. Besides preventing the undesired use of ChatGPT by explicitly integrating it, the proposed assessment seemingly helped the learners question the accuracy of its responses. This self-realised skepticism can be expected to curtail blatant malpractices involving ChatGPT in the long run.