Algorithmic assessments with context in mind

We have discussed various facets of assessments and risk in the last couple of weeks. Many of you have reached out asking for templates and examples. There can’t be “A” template which covers everything.

The trivialization of the assessment process with a focus on templates leads to artifacts that serve the spirit but not the intent of comprehensive assessments.

I have seen enough “half-baked” data cards and model cards which don’t serve any purpose but completes the check-box exercise. In this newsletter, I will focus on how to design algorithmic assessments with context in mind through an example. I will pick a hypothetical example and discuss how to structure an algorithmic assessment with context in mind. Let’s first define the process we will use.

• First, define the goals of the use of the application and the context of use. Engage all stakeholders and get agreement on the intent and specific needs and assumption.
• Second, ask why you need an algorithmic assessment. You must have a clear goal on why an assessment is being done in the first place. Is it to verify whether the application meets performance or usage goals? Is it to validate whether the application serves the intended purpose. Is it to check what may go wrong? Is it to understand the effects of various socio-technical aspects (fairness, explainability, security etc.)?
• Third, Ensure you have to have a clear understanding of the algorithmic system, its design, assumptions and how it works and what the dependencies for the processes are.
• Fourth, engage stake holders and build a clear scoping document. Determine who will conduct the assessment? Will it be internal? Will it be with assistance of external experts? What metrics will you use ? How do you define independence? What risks do you anticipate and how will you identify, assess and evaluate it? ?etc. etc. (These issues have been covered in previous newsletters)
• Fifth, conduct a robust assessment. Leverage the tools, expertise of stakeholders, etc. to conduct assessments. Ensure you keep the context in mind and use the right tools and metrics when evaluating various facets .
• Sixth, Analyze and evaluate your findings and engage the various stakeholders to discuss the findings, recommendations and ensure all aspects of the scope are covered.
• Seven, develop a plan to addresses the found issues. Get agreement from stakeholders on the plan and their commitment to implement changes.

This is not a prescription for all assessments.!

I have kept it generic so you can customize it to your needs. Note, I am not emphasizing tools, dashboards or automation in the process. There is a place for that as you hone your assessment process.

Focus on the key themes with context in mind when you embark on assessing algorithmic systems.

We will illustrate these steps through a hypothetical example on Credit risk in the next few newsletters.

??Keep on learning!

?? Want to learn more? QuantUniversity is offering a LIVE Algorithmic Auditing course on August 22nd 9.30 AM to 4.00pm in partnership with PRMIA. If you are interested, check out details here:?https://prmia.org/Shared_Content/Events/PRMIA_Event_Display.aspx?EventKey=8906

??Many of these topics will be elaborated in the?AI Risk Management?Book published by Wiley. Check updates here ->?https://lnkd.in/gAcUPf_m

??Subscribe to this newsletter/share it with your network ->?https://www.dhirubhai.net/newsletters/ai-risk-management-newsletter-6951868127286636544/

I am constantly learning too :) Please share your feedback and reach out if you have any interesting product news, updates or requests so we can add it to our pipeline.

Sri Krishnamurthy?

QuantUniversity

#machinelearning?#airiskmgt?#ai