Please enable JavaScript.

Coggle requires JavaScript to display documents.

Data, Raw facts that describe the characteristics of an event or object -…

- - - - Data Source Bias
        
        Historical data carries traces of social discrimination (e.g., historical unfair records in recruitment and credit fields)
        
        Limited data collection scope (ignoring data of minority groups and edge scenarios)
        
        Subjective bias in data annotation (cognitive bias of annotators embedded in data)
      - Algorithmic Design Bias
        
        Embedding of developers' subjective tendencies (implicit values and cognitive limitations reflected in program logic)
        
        Defects in algorithm model structure (e.g., excessive weighting of specific features by the model)
        
        Single optimization objective (only pursuing efficiency/accuracy while ignoring fairness indicators)
      - Data Processing and Application Bias
        
        Bias in data preprocessing (amplifying original data imbalance during cleaning and integration)
        
        Bias in model training process (marginalization of data from minority groups and sample imbalance)
        
        Mismatch of algorithm application scenarios (blindly migrating models trained in single scenarios to diverse scenarios)
    - - Impaired social equity (exacerbating discrimination against groups based on gender, race, class, etc.)
      - Infringed individual rights and interests (affecting access to key opportunities such as employment, credit, and education)
      - Disrupted market order (enterprises abusing biased algorithms to gain unfair competitive advantages)
      - Social trust crisis (reducing public recognition of algorithmic decisions and digital services)
    - - Legal Level
        
        Improve laws and regulations on algorithm governance (clarify requirements for algorithmic fairness and division of legal responsibilities)
        
        Establish an algorithm review and filing system (mandatory review of algorithms in high-risk fields)
        
        Strengthen the accountability mechanism for algorithmic infringement (clarify the compensation and liability obligations of developers and operators)
      - Technical Level
        
        Build diverse and balanced training datasets (supplement data of minority groups and eliminate data imbalance)
        
        Develop interpretable algorithm models (improve the transparency of the algorithmic decision-making process)
        
        Introduce fairness constraint technologies (incorporate weights of fairness indicators in model optimization)
      - Industry and Organizational Level
        
        Formulate industry algorithmic ethics norms (clarify fairness standards and industry self-regulation guidelines)
        
        Strengthen the main responsibility of enterprises (establish internal algorithmic ethics review teams and disclose algorithmic decision-making logic)
        
        Promote the development of third-party algorithm evaluation institutions (provide independent and professional fairness testing services)
      - Social Level
        
        Improve public algorithm literacy (popularize knowledge on algorithmic bias and enhance awareness of rights protection)
        
        Encourage multi-subject participation in supervision (collaborative supervision by media, academic institutions, public welfare organizations, etc.)
        
        Strengthen interdisciplinary research cooperation (promote joint research by computer science, law, sociology and other disciplines)