Passionately curious about Data, Databases and Systems Complexity. Data is ubiquitous, the database universe is dichotomous (structured and unstructured), expanding and complex. Find my Database Research at SQLToolkit.co.uk . Microsoft Data Platform MVP

"The important thing is not to stop questioning. Curiosity has its own reason for existing" Einstein

Sunday 2 January 2022

Creating an AI Ethics panel

A data ethics council helps maintain an organization’s values-based intentions, and increases transparency into how we use Data & AI. This enables a focus on three areas for growth:

  • Establish governance for data ethics & AI and consider the importance for data collection and sharing.
  • Describe how/when fairness happens and how/what biases have been accounted for
  • Provide mechanisms for recourse.

Harvard business review has an article Create an Ethics Committee to Keep Your AI Initiative in Check

Accenture have a summary page Building data and AI ethics Committee in your business

Accenture have a full report on building a data ethics committee. 

Ethics Frameworks to help implementation

Four frameworks are covered below, the UK government data ethics framework, Data Ethics Decision Aid (DEDA), the UK statistics authority ethics self-assessment tool and Microsoft Responsible AI Framework. 

The government data ethics framework  has a self-assessment for the 3 overarching principles.

The self-assessment has 5 specific actions. A score lower than 3 requires review.

The areas covered in each area:

1 define public benefit and user need
  • understand unintended and or negative consequences 
  • human rights considerations
  • justify the benefit
  • make user needs and public benefit transparent
  • check everyone understand user need and how to use the data

2 involve diverse expertise
  • ensure diversity within your team
  • involve external stakeholders
  • effective governance structures with experts
  • transparency

3 comply with the law
  • compliance with GDPR and DPA 2018
  • data protection by design
  • accountability
  • transparency
  • project complainant with the equality act 2010
  • ensure effective governance of your data

4 review the quality and limitations of the data
  • data source being used
  • meta data understood
  • processes to maintain integrity
  • is synthetic data appropriate for the project evidence based caveats
  • bias in data to train the model
  • determine proportionality
  • data anonymisation
  • robust practices - demonstrated reproducibility, quality of the model
  • make data open and shareable whenever possible
  • think about transparency of sensitive models
  • explainability

5 Evaluated and consider wider policy implications
  • repeatability
  • project influences
  • accountability structures
  • skills, training and maintenance for longevity of the project
  • share learnings

The 3 areas summary

Another model is the Data Ethics Decision Aid. DEDA is a tool-kit facilitating initial brainstorming sessions to map ethical issues in data projects, documenting the deliberation process and furthering accountability towards the various stakeholders and the public.

Ethics Self-Assessment Tool from the UK statistics authority is another framework .The self-assessment tool provides a timely means to identify ethical issues, shape future discussions and support an accurate and consistent estimation of the “ethical risks” of research proposals.

The Microsoft Responsible AI Framework looks at 6 principles
Fairness - should treat all people fairly
Reliability & Safety -  should perform reliably and safely
Privacy & Security - should be secure and respect privacy
Inclusiveness - should empower everyone and engage people
Transparency - should be understandable
Accountability - People should be accountable for AI systems

No comments:

Post a Comment

Note: only a member of this blog may post a comment.