Establishing Trustworthy Artificial Intelligence in Automated Feedback

Abstract

Providing educational feedback has been widely acknowledged as an effective method to assist student learning. To offer feedback for a large student cohort, educational researchers started developing artificial intelligence (AI) systems. Though the existing AI-driven feedback achieved certain success, many concerns still exist during the deployment of AI (e.g., privacy, AI explainability and accountability) and further shape the untrustworthy barriers between the educational stakeholders' needs and AI efforts. To reduce the barriers, it is necessary to identify the concerns based on the principles of trustworthy AI before the development of automated feedback. Therefore, in this paper, we aimed to tease out the concerns raised in the existing literature. In particular, we mainly discuss trustworthy AI trough five dimensions -- i.e., acceptance, explainability, accountability, fairness, and privacy -- to shed light on the future design of automated feedback.

Kekurangan Aplikasi Grammarly dalam Mendeteksi Kesalahan pada Karya Tulis Mahasiswa

“…Salah satu hasil hasil dari kecerdasan buatan manusia dan juga sedang marak adalah mesin otomatis pemberian umpan balik terhadap hasil tulisan para siswa (corrective feedback on writing). Lin et al, (2022) menyatakan bahwa mesin seperti ini dapat memberikan umpan balik terhadap karya para siswa. Namun kata mereka, masih banyak kekhawatiran yang ditimbulkan mesin seperti kerahasiaan, penjelasan dan akuntabilitas dari mesin hasil kecerdasan buatan seperti ini.…”

Application of CHATGPT in civil engineering

“…CHATGPT is trained through reinforcement learning based on human feedback. Many researchers are currently studying the application of CHATGPT in different disciplines such as finance, health, education, and biotechnology (Antaki et al, 2023;Branco et al, 2023;Dowling & Lucey, 2023;Dwivedi et al, 2023;Agathokleous, 2023;Gomez-Flores et al, 2023;Holzinger et al, 2023;Khogali & Mekid, 2023;Lecler et al, 2023;Li et al, 2023;McGrath et al, 2023;Peres et al, 2023;Saggu & Ante, 2023;Sitapure & Kwon, 2023;Valencia et al, 2023). In their short communication, Gill and Kaur examined the applications of CHATGPT as shown in Figure 1 (Gill & Kaur, 2023).…”