Recently, experimental findings from a research team at the University of Toronto and the Massachusetts Institute of Technology have revealed that AI systems trained with descriptive labels may make decisions that are stricter than those made by humans, posing a threat to future life. The study highlights profound flaws in AI algorithm training, which could lead to decision-making issues in areas such as housing, loans, and surgeries. Using a dog image dataset as an example, the results showed that descriptive labels led to harsher judgments against non-compliant dogs, potentially causing social injustice. Algorithms could exacerbate societal biases in fields such as PhD applications and recruitment, intensifying inequality. Researchers urge the early correction of this issue, emphasizing the critical importance of accurately labeling data to avoid algorithmic bias in an AI-dominated era, as failure to do so could have a severe impact on human life.