Articles Vol. 5 No. 2 (2026) · pp. 63-69 Jurnal Nasional

The Transformation of Learning Assessment in the Era of Generative Artificial Intelligence: A Literature Review on the Redesign of Authentic Assessment, Instrument Validity and Reliability, and the Shift Toward Formative Assessment in Higher Education

Transformasi Evaluasi Pembelajaran di Era Kecerdasan Buatan Generatif: Kajian Pustaka Atas Redesain Asesmen Otentik, Validitas-Reliabilitas Instrumen, dan Pergeseran ke Evaluasi Formatif di Pendidikan Tinggi

DOI: https://doi.org/10.64464/tarbiyah.v5i2.269
  • Saidatun Nisa Nasution(1) * Corr.
  1. (1) Sekolah Tinggi Ilmu Tarbiyah Serdang Bedagai, Indonesia

* Corresponding author: saydanasty08@gmail.com

PDF (Indonesian)

Abstrak

Proliferasi perangkat kecerdasan buatan generatif (AI generatif) telah mengguncang asumsi fundamental evaluasi pembelajaran di pendidikan tinggi, khususnya premis bahwa sebuah produk tulisan yang dikumpulkan mahasiswa secara andal mencerminkan kompetensi individu yang bersangkutan. Penelitian ini menerapkan desain kualitatif dengan pendekatan studi pustaka (library research) untuk mengkaji bagaimana bidang evaluasi pembelajaran merespons disrupsi tersebut, dengan tiga fokus, yaitu: (1) redesain instrumen asesmen menuju tugas otentik yang tahan terhadap AI; (2) validitas dan reliabilitas instrumen deteksi AI yang digunakan untuk menjaga integritas evaluasi; serta (3) pergeseran paradigma evaluasi sumatif menuju formatif yang dimungkinkan oleh umpan balik berbasis AI generatif. Data dihimpun dari jurnal terindeks, satu studi empiris multi-alat berskala internasional, dan literatur teori evaluasi klasik yang terbit dalam rentang 2011-2025, kemudian dianalisis secara tematik. Hasil kajian menunjukkan bahwa literatur redesain asesmen mengerucut pada tugas berorientasi proses dan berpikir tingkat tinggi yang secara struktural sulit diotomatisasi penuh, selaras dengan teori penyelarasan konstruktif (constructive alignment). Namun demikian, pengujian empiris terhadap alat deteksi AI mengungkap persoalan reliabilitas yang serius, dengan tingkat akurasi di bawah 80% pada mayoritas alat yang diuji serta risiko tuduhan keliru (false accusation) yang terdokumentasi, sehingga alat tersebut tidak layak dijadikan satu-satunya bukti dalam keputusan evaluasi yang berisiko tinggi. Sementara itu, umpan balik otomatis berbasis AI generatif menunjukkan potensi untuk menskalakan umpan balik formatif yang cepat, meski kualitas pedagogisnya masih bervariasi dan tetap memerlukan pengawasan manusia. Penelitian ini menyimpulkan bahwa sistem evaluasi yang dapat dipertanggungjawabkan di era AI generatif harus secara simultan meredesain objek yang dinilai, memperlakukan teknologi deteksi sebagai bukti pendukung dan bukan bukti final, serta mengarahkan ulang budaya evaluasi dari penjagaan gerbang sumatif menuju praktik formatif yang berkelanjutan dan dialogis.

The proliferation of generative artificial intelligence (generative AI) tools has shaken the fundamental assumptions of learning evaluation in higher education, particularly the premise that a piece of writing submitted by a student reliably reflects that individual's own competence. This study employs a qualitative design with a library research approach to examine how the field of learning evaluation is responding to this disruption, focusing on three areas: (1) the redesign of assessment instruments toward AI-resistant authentic tasks; (2) the validity and reliability of AI detection instruments used to safeguard evaluation integrity; and (3) the paradigm shift from summative toward formative evaluation enabled by generative AI-based feedback. Data were collected from indexed journals, one international-scale multi-tool empirical study, and classical evaluation theory literature published between 2011 and 2025, and were subsequently analyzed thematically. The findings indicate that the assessment redesign literature converges on process-oriented and higher-order thinking tasks that are structurally difficult to fully automate, in line with constructive alignment theory. However, empirical testing of AI detection tools reveals serious reliability issues, with accuracy rates below 80% for the majority of tools tested, as well as documented risks of false accusation, rendering such tools unsuitable as the sole evidence in high-stakes evaluation decisions. Meanwhile, automated feedback based on generative AI shows potential for scaling rapid formative feedback, although its pedagogical quality remains variable and continues to require human oversight. This study concludes that an accountable evaluation system in the era of generative AI must simultaneously redesign the object being assessed, treat detection technology as supporting rather than definitive evidence, and redirect the culture of evaluation from summative gatekeeping toward sustainable and dialogic formative practice.

Keywords / index terms
  • Learning Evaluation
  • Generative Artificial Intelligence
  • Authentic Assessment