AI risk assessment

Artificial Intelligence (AI) Strategy

Vidhi Chugh

AI strategist and ethicist

Ethical guardrails

 

• UNESCO
  • Need guardrails
  • Prevent risks such as bias and discrimination
  • Fundamental human rights

 

¹ https://www.unesco.org/en/artificial-intelligence/recommendation-ethics

AI risks

 

• National Institute of Standards and Technology (NIST)

  

 

• Probability of occurrence
• Degree of the consequences

¹ https://www.nist.gov/artificial-intelligence

Data risks

• Data privacy
  • Sensitive information

• Data bias
  • Perpetuate societal biases  

AI gone wrong

 

• Gender bias

 

• Racial discrimination

Data privacy breach

• Heavy fine for non-compliance with the GDPR
• Unlawful processing of data
• Scraped images for facial recognition

 

Business implications

Legal

Business implications

Legal, reputational

Business implications

Legal, reputational and financial harms.

Ethical frameworks

Ethical frameworks

Ethical frameworks

Key pillars of risk assessment

                Privacy

                Fairness

             Transparency

             Accountability

Regulations and compliance

• AI project lifecycle

• Regulations and compliance
  • Accountability requires governance
  • Define roles and responsibilities
  • Compliance with data privacy regulations
  • Seek user consent and anonymize data

Data collection and preparation

Assess

• Internally available or procure externally?
• Sensitive information such as Personal Identifiable Information (PII)?
• Well-represented population?

Mitigate

• Processes for data integrity
• Identifier encryption, synthetic data generation
• Create balanced data

Model development and evaluation

• Explain model outcomes
  • Explanation and bias identification
  • Fair treatment across categories

Model deployment and monitoring

 

• Degree of impact

• Establish mitigation strategies

 

 

• Cost is high in healthcare and finance

• Realign model behavior

 

Let's practice!

Artificial Intelligence (AI) Strategy