在医学语言模型中评估和减轻认知偏见
Samuel Schmidgall1, Carl Harris2, Ime Essien2
1Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA. sschmi46@jhu.edu.
NPJ digital medicine
|October 21, 2024
概括
大型语言模型 (LLM) 在偏见的医学问题上显示精度降低. GPT-4表现出弹性,而其他模型则扎,突出了在临床AI应用中改善偏差缓解的必要性.
科学领域:
- 人工智能在医学中的应用
- 自然语言处理自然语言处理.
- 临床决策支持 临床决策支持
背景情况:
- 大型语言模型 (LLM) 在医疗应用中表现有前途,其在标准化考试中的表现证明了这一点.
- 医学检查并不完全代表现实世界患者与医生的互动,这涉及诸如认知偏见等复杂因素.
- 在临床偏见的情况下评估LLM绩效对于安全有效的医疗整合至关重要.
研究的目的:
- 评估临床认知偏差对医学问题答案中的大语言模型 (LLM) 准确性的影响.
- 为了比较不同LLMs的偏差弹性.
- 在医疗环境中测试LLM偏差缓解策略的有效性.
主要方法:
- 开发BiasMedQA数据集,包括1273个USMLE问题,修改以包括常见的临床认知偏见.
- 使用BiasMedQA数据集对六种不同的LLM进行评估.
- 实施和评估三个不同的偏见缓解策略.
主要成果:
- 在大多数LLM中,当与临床偏见问题相比,与无偏见的问题相比,观察到显著的绩效下降.
- 与其他评估模型相比,GPT-4对认知偏差具有优越的弹性,包括Llama 2 70B-chat和PMC Llama 13B.
- 偏差缓解策略显示LLM准确度有所改善,但并没有完全消除业绩差距.
结论:
- 目前的LLM对临床认知偏见具有脆弱性,这影响了他们在医疗保健环境中的可靠性.
- 进一步的研究和开发是必要的,以提高LLM强度对认知偏见可靠的医疗应用.
- 确保LLM对偏见的弹性对于其在临床实践中安全有效地部署至关重要.
更多相关视频
相关概念视频
Language and Cognition
329
Language serves as a bridge between ideas and communication, influencing how individuals perceive and interact with the world. Psychologists have long debated whether language shapes thought or vice versa. This discussion gained grip with Edward Sapir and Benjamin Lee Whorf in the 1940s, who proposed that language determines thought, a concept known as linguistic determinism. They suggested that the vocabulary and structure of a language influence how its speakers think and perceive reality.
329
Beck's Cognitive Therapy
39
Cognitive therapy is a psychological approach designed to address distortions in thinking, which can lead to negative emotions and unrealistic beliefs. These cognitive distortions often influence how individuals interpret and respond to situations, exacerbating emotional distress. Below are some prevalent cognitive distortions, their characteristics, and examples of how they manifest in thought processes.
Arbitrary Inference
Arbitrary inference involves making conclusions without sufficient...
Arbitrary Inference
Arbitrary inference involves making conclusions without sufficient...
39
Bias in Epidemiological Studies
161
Biases can arise at various stages of research, from study design and data collection to analysis and interpretation. Recognizing and addressing these biases is essential to ensure the validity and reliability of epidemiological findings.Broadly speaking, biases in epidemiology fall into three main categories: selection bias, information bias, and confounding. A more detailed description of possible biases is:
161
Cognitive Theories: Lazarus Mediational Theory of Emotion
669
Richard Lazarus' cognitive mediational theory highlights the pivotal role of cognitive appraisal in shaping emotional responses. According to this theory, the evaluation of a stimulus — based on personal values, goals, beliefs, and expectations — mediates the emotional response. This appraisal process is immediate and often occurs unconsciously, influencing the intensity and nature of the resulting emotion.
Cognitive Appraisal and Emotional Response
Lazarus proposed that...
Cognitive Appraisal and Emotional Response
Lazarus proposed that...
669
Cognitivism
1.4K
Cognitive psychology emerged as a significant field in the mid-20th century. It focused on understanding humans' internal mental processes. This approach emphasizes how people perceive, remember, think, and solve problems—elements critical to human cognition.
Previously dominated by behaviorism, which prioritized observable behaviors and largely ignored mental processes, psychology transformed in the 1950s. Cognitive psychologists argue that understanding how we think and process...
Previously dominated by behaviorism, which prioritized observable behaviors and largely ignored mental processes, psychology transformed in the 1950s. Cognitive psychologists argue that understanding how we think and process...
1.4K
Bias
3.7K
Bias refers to any tendency that prevents a question from being considered unprejudiced. In research, bias occurs when one outcome or answer is selected or encouraged over others in sampling or testing. Bias can occur during any research phase, including study design, data collection, analysis, and publication.
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
3.7K


