大型语言模型中的性别和内容偏见:关于谷歌双子座2.0闪光实验的案例研究
1Department of Arts, Università di Bologna, Bologna, Italy.
Frontiers in artificial intelligence
|April 2, 2025
概括
这项研究比较了Gemini 2.0 Flash实验和ChatGPT-4o的内容调节. 双子座2.0显示性别偏见有所减少,但允许更多的暴力内容,这引发了人工智能开发的伦理问题.
科学领域:
- 人工智能的人工智能
- 计算机科学 计算机科学
- 人工智能中的伦理学
背景情况:
- 大型语言模型 (LLM) 需要强大的内容调节.
- 之前的工作分析了ChatGPT-4o的伦理调节实践.
- 评估像双子座2.0这样的新型LLM中的偏见至关重要.
研究的目的:
- 在双子座2.0闪光实验中评估内容调节偏见.
- 将Gemini 2.0的性别偏见和道德立场与ChatGPT-4o进行比较.
- 分析双子座2.0的调节政策的影响.
主要方法:
- 双子座2.0闪光实验和聊天GPT-4o的比较分析.
- 用性别特定提示进行内容调节的测试.
- 对性和暴力内容的接受率的评估.
主要成果:
- 与ChatGPT-4o相比,双子座2.0表现出较少的性别偏见,对女性特定提示的接受率更高.
- 双子座2.0显示对性内容的宽容性增加.
- 双子座2.0保持了对暴力内容的接受率很高,影响了两性.
- 针对男性的提示通常仍然比针对女性的提示有更高的接受率.
结论:
- 双子座2.0降低了性别偏见,但增加了对暴力内容的宽容性,造成了道德困境.
- 减少偏见和内容宽容之间的权衡需要仔细考虑.
- 为了实现透明,公平和包容的节制,需要进一步完善人工智能模型.
相关概念视频
Stereotype Content Model
The Stereotype Content Model (SCM) was first proposed by Susan Fiske and her colleagues (Fiske, Cuddy, Glick & Xu, 2002; see also Fiske, 2012 and Fiske, 2017). The SCM specifies that when someone encounters a new group, they will stereotype them based on two metrics: warmth—or that group’s perceived intent, and how likely they are to provide help or inflict harm—and competence—or their ability to carry out that objective. Depending on the warmth-competence categorization, a person will feel...
Improving Translational Accuracy
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
X and Y Chromosomes
Among mammals, the gender of an organism is determined by the sex chromosomes. Humans have two sex chromosomes, X and Y. Every human diploid cell has 22 pairs of autosomes and one pair of sex chromosomes. A human female has two X chromosomes, while a male has one X chromosome and one Y chromosome.
The germline cells such as egg and sperm cells carry only half the number of chromosomes, i.e., 22 autosomes and one sex chromosome. All eggs have an X chromosome, while sperm cells can carry an X or...
The germline cells such as egg and sperm cells carry only half the number of chromosomes, i.e., 22 autosomes and one sex chromosome. All eggs have an X chromosome, while sperm cells can carry an X or...
The Y Chromosome Determines Maleness
The Y chromosome is a sex chromosome found in several vertebrates and mammals, including humans. In addition to 22 pairs of autosomes, the human males have one X chromosome and one Y chromosome. In these organisms, the presence or absence of the Y chromosome determines the development of male traits.
Evolution
Around 300 million years ago, the two sex chromosomes diverged from two identical autosomal chromosomes. Over time, the Y chromosome has lost most of its genes, shrinking in size. Today,...
Evolution
Around 300 million years ago, the two sex chromosomes diverged from two identical autosomal chromosomes. Over time, the Y chromosome has lost most of its genes, shrinking in size. Today,...
Improving Translational Accuracy
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
Bias
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...


