Watch this Ask the Expert session to learn how a trustworthy AI approach injects fairness and oversight at every step of the AI and analytics journey.
Watch the Webinar
You will:
- Learn how trustworthy AI helps create fair business decisions.
- Discover how to put trustworthy AI principles to practice.
- Get inspired about the end-to-end trustworthy AI capabilities of SAS® Viya®.
The questions from the Q&A segment held at the end of the webinar are listed below and the slides from the webinar are attached.
Q&A
What are ways in which we can foster transparency in our AI & Analytics projects?
By building those natural language insights, utilizing those that are present in SAS Viya right now, and having that explain-ability at the global level of the model but also at the local level with the means of Shapley values and LIME. The other thing is also having that human in loop concept by implementing the AI workflow where you are not only having your model outputs, but you also have some of the business decisions that could be implemented in the workflow concept. Then you have that always human loop to have that auditability trail, and, thus, alluding to the transparency.
How can we use SAS Studio for fairness & bias assessment?
In SAS Studio, we have this Fair AI action toolset. You can use the CAS actions that are present to do the bias detection that I showed in the Fairness & Bias tab in Model Studio. It could be done programmatically to retrieve all the bias metrics.
When using any of the AI tools in SAS Viya, is there a boilerplate disclaimer that should be footnoted so that viewers of your results understand the potential for bias?
We don't currently have a disclaimer across all the tools.
Do these fairness views incorporate priors, or only model output?
Right now, I think it is focused only on the model outputs. That is what you saw in the prediction bias and the performance bias plots where the sensitive variable that was asked for the user to select prior to building the models and then it showcased the bias metrics from that perspective. So, I would say the latter.
Can you export the interpretability in terms of rules?
We have SAS Intelligent Decisioning that allows the user to export all the decision flow that was made to make this decision. It is exportable.
Bias mitigation is only available using CAS action. Are you covering the mitigate Bias action set?
In this particular session, it was geared towards showcasing end-to-end stuff. But that is something that we might consider for later webinars because we do have bias mitigation action set in the fair AI toolset, which can do the bias mitigation in your in-process model development stage.
Could you expand on the programmatic bias assessment you mentioned SAS Studio can do? Is it outlier detection? What would define bias?
For Bias detection there is a programmatic way that we can do within SAS Studio. There is assess bias action set that comes out of Fair AI toolset, which could be used in your CAS actions program in your SAS Studio. It will give the same result as the graph like performance bias metrics, prediction bias metrics, the equalized odds, demographic parity, and all those things. It can create those programmatically, as well. Here’s a good reference.
Recommended Resources
Responsible Innovation at SAS
Trustworthy AI Blogs
Moving from SAS®9 to SAS® Viya®
Move to Viya Board
Please see additional resources in the attached slide deck.
Want more tips? Be sure to subscribe to the Ask the Expert board to receive follow up Q&A, slides and recordings from other SAS Ask the Expert webinars.
