Additional tools for RAI

Responsible AI Practices

Esther Montoya van Egerschot

Chief AI Governance @ DigiDiplomacy

Risk versus impact assessments

 

RISK

  • Foresee potential issues
  • Negative impact

 

IMPACT

  • Foresee potential and risk
  • Positive and negative impact
Responsible AI Practices

Risk assessments help:

  • anticipate harmful impacts
  • to mitigate them proactively

 

Examples:

  • NIST Artificial Intelligence Risk Management Framework (AI RMF) plus AI RMF Playbook
  • New York City's Local Law 144
  • France's CNIL
  • OECD's AI Risk Evaluation Framework

A universal approach for AI risk assessments is still lacking.

Responsible AI Practices

AI impact assessments help organizations:

  • Identify the broader effects
  • Ensure alignment with human-centered values
  • Foster trust and transparency

Examples:

  • The Dutch AI Impact Assessment (AIIA)
  • Microsoft's Responsible AI Impact Assessment Standard
  • IBM® watsonx
  • Human Rights Impact Assessment (HRIA)
  • The Responsible AI Institute RAISE Benchmarks
Responsible AI Practices

Common pitfalls

  • Ethical blind spots: implicit bias
  • Bluewashing or ethics washing: pretending
  • AI shopping: business over ethics
  • Shadow AI: no explicit approval or oversight

Integrating AI impact assessments alongside risk assessments:

  • ensures a comprehensive evaluation of AI systems
  • avoids harm and actively contribute to societal good

An image of an eye in a dark background as a blind spot

1 DALL·E
Responsible AI Practices

Let's practice!

Responsible AI Practices

Preparing Video For Download...