into_to_AI_Ethics.pdf

Transcript

2. IINTRODUCTION  AI increasingly has n impact on everything from social media to health care. Contents A HCD fairness A bias Model cards
3. HUMAN –CENTERED DESIGN FOR AI  HCD is an approach to designing systems that serve people’s needs.  HCD involves people in every step of the design process  Applying HCD for AI designs involves 6 steps totally Step.1 Step.2 • 1.Understanding people’s needs to define the problem • Check if AI adds valve to nay potential solution Step.3 Step.4 • Consider the potential harms that the AI system could cause • Prototype , starting with non-AI solutions Step.5 Step..6 • Provide ways for people to challenge the system • Building in safety measures
4. IDENTIFYING BIAS IN AI  Bias refers to the negative, unwanted consequences of machine learning ML applications, especially if the consequences disproportionately affect certain groups.
5. TYPES OF BIAS  HISTORICAL BIAS occurs when the data generated is flawed  REPRESENTATION BIAS Occurs if the datasets build for training a model poorly represent the people that model will serve.  MEASUREMENT BIAS Occurs when the accuracy of the data varies across groups.  AGGREGATION BIAS occurs when groups are inappropriately combined  EVALUATION BIAS Occurs when the benchmark data does not represent the population that the model will serve when evaluating a model.  DEPLOYMENT BIAS Occurs when the problem that model is intended to solve is different from the way it is actually used.
6. AI FAIRNESS  Fairness in machine learning totally depends on the industry and the purpose it serves. For example: lets consider the model that approves credit card applications Is it fair if the approval rate is equal across genders, or Is it better if gender is removed from the dataset and hidden from themodel ?
7. FOUR FAIRNESS CRITERIA STATISTICAL PARITY EQUAL OPPORTUNITY EQUAL ACCURACY GROUP UNAWARE
8. MODEL CARDS  model card is a short document that provides key information about a machine learning model  Model cards increase transparency by communicating information about trained models to broad audiences.  Examples of model cards  Salesforce’s model cards  Open AI’s model card for GPT-3  Google cloud’s example model cards
9.  A model card should strike a balance between being easy- to – understand and communicating important tecnical information.  SECTIONS OF A MODEL CARD • Model details • Intended use • Factors • Metrics • Evaluation data • Training data • Quantitative analyses • Ethical considerations • Caveats and recommendations
10. CONCLUSION  AI plays a major role in many aspects like credit card decisions, to make video surveillance in airports, and to inform military operations. also these technologies have the potential to harm or help the people that they serve. So an initiative can be taken in identifying the harms by applying the ethical lens and can build designs to reduce these harms.