Scholarlink

Definition of the problem <p>Assistance systems based on machine learning are increasingly being used in medical imaging diagnostics. But can physicians still take responsibility for a&#xa0;diagnosis if they do not fully understand how it came about? Those who blindly trust in technology can hardly fulfill the epistemic condition of responsibility.</p> Arguments <p>Qualitative interviews with radiologists and pathologists on the topics of “responsibility” and “trust” show that physicians only trust artificial intelligence (AI) results if they have the opportunity to control them. Control can create the basis for <i>justified</i> (as opposed to <i>blind</i>) trust. It is worth considering whether such justified trust can also fulfill the epistemic condition of responsibility: Physicians would then no longer have to (be able to) check a&#xa0;specific individual result of the AI, but it would be sufficient that they have <i>previously</i> checked and continuously re-evaluate the expertise of AI for this task. It is clear that the challenges outlined above are by no means specific to AI: Assuming responsibility despite a&#xa0;lack of knowledge is also encountered, among others, in the area of delegation. The fact that internal processes are difficult to comprehend also applies to humans and other technical devices.</p> Conclusion <p>In order to trust AI in a&#xa0;responsible way, physicians need a&#xa0;basic understanding of information technology and a&#xa0;heightened awareness of the respective strengths of humans and machines. They should generally have the opportunity to understand AI results, but there should be a&#xa0;differentiated evaluation of the processes for which traceability is necessary and helpful. AI developers could be held more accountable through product liability.</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 in der bildgebenden Diagnostik verantwortet vertrauen. Erfahrungen aus Radiologie und Pathologie ethisch diskutiert

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月2日

查看全文
  • Wiebke Brandt,
  • Alexis Fritz,
  • Angelika Kießig,
  • Philipp Lerch

摘要

Definition of the problem

Assistance systems based on machine learning are increasingly being used in medical imaging diagnostics. But can physicians still take responsibility for a diagnosis if they do not fully understand how it came about? Those who blindly trust in technology can hardly fulfill the epistemic condition of responsibility.

Arguments

Qualitative interviews with radiologists and pathologists on the topics of “responsibility” and “trust” show that physicians only trust artificial intelligence (AI) results if they have the opportunity to control them. Control can create the basis for justified (as opposed to blind) trust. It is worth considering whether such justified trust can also fulfill the epistemic condition of responsibility: Physicians would then no longer have to (be able to) check a specific individual result of the AI, but it would be sufficient that they have previously checked and continuously re-evaluate the expertise of AI for this task. It is clear that the challenges outlined above are by no means specific to AI: Assuming responsibility despite a lack of knowledge is also encountered, among others, in the area of delegation. The fact that internal processes are difficult to comprehend also applies to humans and other technical devices.

Conclusion

In order to trust AI in a responsible way, physicians need a basic understanding of information technology and a heightened awareness of the respective strengths of humans and machines. They should generally have the opportunity to understand AI results, but there should be a differentiated evaluation of the processes for which traceability is necessary and helpful. AI developers could be held more accountable through product liability.