Scholarlink

Background <p>The development of generative artificial intelligence (AI) offers considerable promise for supporting the process of visualization for the purpose of teaching in forensic medicine; however, preliminary observations suggest that AI systems decline to depict scenes involving injury, violence or death, domains that are central to forensic pedagogy.</p> Method <p>To assess the hypothesis of an emergent epistemic distortion driven by AI-related restrictions, three major language and image models (ChatGPT, Gemini and Stable Diffusion) were confronted with five targeted prompts on forensic topics. Outputs were rated on a&#xa0;5-point scale. All prompts were issued in both Germany and Hungary to examine potential jurisdictional variation.</p> Results <p>In Germany, the systems refused to generate forensic visualizations in half of all cases, compared with only 5.6% in Hungary. Across both settings, depictions of violence-related mechanisms were largely incorrect or unusable.</p> Discussion <p>The pronounced divergence between Germany and Hungary intimates the presence of a&#xa0;deeper, systemically active logic of distortion. The results indicate that moderating regimes primarily define the epistemic scope and not technical constraints. A&#xa0;feasible solution could be the creation of academic special access pathways to generative AI, including locally hosted open-source models operating under institutional ethical oversight, thereby enabling a&#xa0;methodically governed, filter-reduced environment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

KI-basierte Bilder zu forensischen Demonstrationszwecken

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月27日

查看全文
  • Frank Ramsthaler,
  • Marcel A. Verhoff

摘要

Background

The development of generative artificial intelligence (AI) offers considerable promise for supporting the process of visualization for the purpose of teaching in forensic medicine; however, preliminary observations suggest that AI systems decline to depict scenes involving injury, violence or death, domains that are central to forensic pedagogy.

Method

To assess the hypothesis of an emergent epistemic distortion driven by AI-related restrictions, three major language and image models (ChatGPT, Gemini and Stable Diffusion) were confronted with five targeted prompts on forensic topics. Outputs were rated on a 5-point scale. All prompts were issued in both Germany and Hungary to examine potential jurisdictional variation.

Results

In Germany, the systems refused to generate forensic visualizations in half of all cases, compared with only 5.6% in Hungary. Across both settings, depictions of violence-related mechanisms were largely incorrect or unusable.

Discussion

The pronounced divergence between Germany and Hungary intimates the presence of a deeper, systemically active logic of distortion. The results indicate that moderating regimes primarily define the epistemic scope and not technical constraints. A feasible solution could be the creation of academic special access pathways to generative AI, including locally hosted open-source models operating under institutional ethical oversight, thereby enabling a methodically governed, filter-reduced environment.