인공지능(AI) 양형과 법관의 양심

Appropriate and reasonable sentencing is realized by pronouncing sentences as envisaged by the Constitution and the law in specific cases, while ensuring consistency across identical cases. Currently, however, sentencing in Korean criminal courts relies heavily on excessive judicial discretion, underpinned by broad statutory ranges and the lack of specific criteria for discretionary mitigation (정상참작감경). This reliance has led to significant issues, including judicial arbitrariness and sentencing disparities that undermine public trust. To address these concerns, the Supreme Court of Korea has decided to pilot an Artificial Intelligence (AI) sentencing system starting in 2028, with full-scale implementation scheduled for 2030. Despite its potential, constitutional controversies have arisen regarding whether such a system might infringe upon the principle of trial according to a judge’s conscience, as stipulated in Article 103 of the Constitution, particularly if AI effectively binds judicial decisions. This paper addresses these concerns by reconceptualizing the ‘conscience of a judge’ under Article 103—not as a private, subjective conviction, but as ‘conscience as social universality’ and ‘conscience as institutional morality’. From this perspective, the AI sentencing system achieves constitutional consonance by serving as a tool for self-reflection. It allows judges to guard against personal dogma and prejudice by cross-referencing their pre-understandings with empirical data. Ultimately, rather than replacing the judge’s judgment, the AI sentencing system functions as an external benchmark that compels judges to provide rigorous legal reasoning—explaining why their decision is superior from the standpoint of institutional morality whenever it deviates from the AI’s recommendation. This process prevents the intervention of an arbitrary conscience and, consequently, leads to the substantive realization of the rule of law.