The goal is to help organizations navigate the turbulent AI landscape by mitigating risks and helping them achieve goals
A leader in data and AI, SAS introduces new products and services Trustworthy AI to strengthen AI governance and promote the reliability and transparency of models. Model cards and new AI governance consulting services will help organizations navigate the turbulent AI landscape by mitigating risks and helping them achieve AI goals more confidently. SAS has also published a Trustworthy AI Lifecycle Workflow, mapped by the National Institute of Standards and Technology (NIST) AI Risk Management Framework.
“Our clients are excited about the potential of AI, but remain cautious about when and how to use it,” says Reggie Townsend, vice president of SAS’ data ethics practice. “They have good questions about responsible AI and ethics. Our goal is to provide tools and guidance informed by decades of experience to integrate AI, drive profitability while reducing unintentional harm.”
Model cards: the “nutrition labels” of trustworthy AI
It can be difficult to convert something as complex and sophisticated as an AI model into something understandable for everyone involved in the AI lifecycle. And with the approval of new rules and regulations around the world, the ability to understand and share with regulatory authorities the performance of a model becomes crucial. Model cards, a future feature of SAS® Viya®, will be useful to stakeholders throughout the AI lifecycle. From developers to directors, everyone will see the value of a curated tool that supports both proprietary and open source models.
Scheduled to launch in mid-2024, model cards are best described as “nutrition labels” for AI models. The SAS approach is to automatically generate model cards for registered models with content coming directly from SAS products, relieving the burden on individual users who need to create them. Additionally, as SAS Viya already has an existing architecture for open source management, model cards will also be available for open source models, starting with Python models.
Model cards will highlight indicators such as accuracy, fairness, and model drift, which is the deterioration of model performance when conditions change. They also include governance details such as when the model was last modified, who contributed, and who is responsible for it, which allows organizations to internally handle abnormal model performance. The model usage section addresses intended use, out-of-scope use cases, and limitations, which will be key factors as model transparency and auditing will likely be regulated corporate operations. The model cards were presented earlier this year at SAS Insight, an event for analysts.
“SAS has taken a cautious approach to guiding its customers in adopting artificial intelligence, focusing on the practical realities and challenges of implementing AI in real industry environments,” says Eric Gao, research director at IDC. “Model cards will be valuable for monitoring AI projects and promoting transparency.”
New AI Governance Group Led by an Ethical AI Expert
With the proliferation of artificial intelligence, SAS customers have become increasingly concerned about how to use their data productively and securely. To help them on their data and AI journeys, SAS introduces AI Governance Advisory, a value-added service for today's customers.
Starting from a brief meeting, SAS AI Governance Advisory helps the client reflect on the meaning of AI governance for their organization. SAS has tested this service, and customers have noticed several benefits:
- Greater productivity with reliable and distributed decisions.
- Greater trust with more responsibility in the use of data.
- Acquire and retain the best talent, which requires responsible innovation practices.
- Increased competitive advantage and market agility thanks to “future conformity”
- Greater brand value to face possible impacts on society and the environment.
Polish PZU Insurance is one of the largest financial institutions in Central and Eastern Europe. A long-time SAS customer, PZU implements AI in areas such as claims, sales, fraud detection and customer service.
“Our conversations about AI governance with SAS helped us identify potential blind spots that would cause problems for customers and the business,” says Marek Wilczewski, managing director of information, data and analytics management (Chief Data Officer/Chief Analytics Officer) of the PZU. “We better understand the importance of having more perspective when embarking on AI projects.”
Industry veteran and ethical AI expert Steven Tiell has been hired by SAS as global director of AI governance. Tiell, who led Accenture's global data ethics and responsible innovation practice, is also the former vice president of AI strategy at DataStax.
Development on Emerging Government Standards
Last year, the US National Institute of Standards and Technology (NIST) released an AI Risk Management Framework. It has become a valuable tool for organizations to create and manage trustworthy and responsible AI in the absence of official regulations.
SAS has established a Trustworthy AI Lifecycle Workflow to make it easier for organizations to adopt NIST recommendations by specifying individual roles and expectations, gathering necessary documentation, outlining factors to consider, and leveraging automation to facilitate adoption . Organizations obtain a production model with documentation demonstrating due diligence to ensure the model is fair and its processes do not cause harm.
The workflow allows organizations to record their considerations about the impacts of AI systems on people's lives. It includes steps that ensure that the data being trained represents the affected population and that model predictions and performance are similar across protected categories. These steps help prevent the model from causing disparate impacts or harm to specific groups. Additionally, users can control the accuracy of a model over time, creating human intervention tasks for when special attention is needed.
