Responsible innovation is critical because technology does not exist in a vacuum. It affects us all in unexpected ways. We know analytics has an undeniable impact on society. For example, analytics can help hospitals manage their inventories for essential items like wheelchairs and bladder scanners, help sports teams curate a perfect fan experience and even help governments identify when a new federal office is needed. With this great power comes a responsibility to be cautious and intentional when applying powerful analytics.

The responsible approach to innovation means asking "could we?" and "should we?" when developing new technologies. Answering these questions requires an unwavering set of principles rooted in core beliefs and proven effective strategies.

The core values at SAS translate into human centricity, inclusivity, accountability, transparency, robustness and privacy and security. From product ideation to development and deployment, these principles are reflected in the people, the processes, and the products. Read on to learn more about each of the principles and discover how they appear in SAS® Viya®.

Read more stories in the series about data ethics principles

Focus on the people

Human centricity aims to promote human well-being, agency, and equity. When creating analytics models, it is essential to remember that models impact real people. Data-driven systems must be designed to protect and preserve human dignity, promote overall well-being and embrace fairness.

While the specifics of the design may vary by use cases and context, all designs can benefit from an intentional level of human oversight and caution around delegating decision-making authority to AI.

The analytical workflow should require a human to complete and approve certain tasks. This design promotes human oversight and agency in the analysis. Humans should be in the loop in many contexts.

Learn more about how SAS is driving responsible innovation 

Include before you analyze

Inclusivity focuses on the accessibility and inclusion of diverse perspectives in designing, developing, and deploying data-driven systems. In the early phases of the analytics lifecycle, organizations may inadvertently introduce biases through incomplete data. It is important to assess data quality metrics like completeness, uniqueness, distinct values and semantic types. During the modeling phase of the data lifecycle, correlated variables, mismatched values and outliers may bring unintended consequences.

It is also crucial to perform data exploration that lets the user see the distributions of variables in training data and analyze the relationships between input and target variables before training models. Data exploration ensures that the data's insights represent diverse populations.

Solve the most complex analytical problems with a single, integrated, collaborative solution

Build accountability through decision workflows

Accountability is recognizing the role a model developer, deployer and user plays in the model's outcomes. Accountability necessitates individuals and organizations proactively identifying and mitigating adverse impacts from data-driven systems. All parties involved in the system must recognize their role and work together to minimize harm.

Incorporating a decision workflow would allow users to create, approve, annotate, deploy and audit decisioning processes. Intentional oversight in the decision workflow also helps build accountability in various decision checkpoints. Accountability helps organizations take a proactive approach to understanding and mitigating risks.

Be able to communicate what, why and how

Transparency within responsible innovation translates to clear communication about the model's intentions, scope, and limitations. Transparency fosters trust in data-driven systems and helps people understand how decisions are made. Organizations can achieve transparency through capabilities such as data lineage, which allows users insight into data assets in the context of their sources, outputs and relationships.

Model interpretability empowers users to interpret the results of AI models with techniques like partial dependence (PD) plot, individual conditional expectation (ICE) plots, local interpretable model-agnostic explanations (LIME) and HyperSHAP. You may even consider generating explanations of data and models written in simple business language understood by any audience through natural language insights.

Natural language processing: What is it? Why does it matter? 

Reliability and consistency are essential

Responsible data-driven systems also need to operate reliably and safely. A data-driven system that changes outputs depending on the day and the phase of the moon is likely useless to most organizations. This is where the principle of robustness comes into consideration. Robust data systems produce consistent and accurate results for organizations.

One way to achieve a robust model is by introducing the model monitoring concept within your data lifecycle. Model monitoring empowers the users to automatically monitor deployed models for accuracy, fairness and relevance over time. Robust data systems need to be tested against a range of inputs and real-world scenarios to reduce unforeseen harm and ensure consistency of performance.

SAS® Model Manager: Learn how to build a streamlined, secure ModelOps process 

Keep things private and secure

Finally, organizations must consider privacy and security roles within their data life cycles. Responsible innovators must meet regulatory requirements and respect the use and application of data about individuals or populations. This principle of privacy and security is based on the respect for autonomy of data subjects to dictate their privacy.

Organizations with an appreciation for privacy and security should consider implementing information privacy capabilities to indicate if the data contains potentially private information that could be linked to an individual. This would allow customers to treat sensitive data cautiously and empower them to consider whether additional protective measures are needed for specific models. When the best way to protect an individual's privacy is to hide the data values, consider embracing data masking techniques.

SAS® Information Governance: Understand, create awareness of, control and manage data assets while protecting the data

Minimizing harm and building trust

Ultimately, responsible innovation is an essential element for the ethical development of technology. The power of analytics can positively impact society, but it also requires responsibility and intentionality to ensure its impact is not harmful. To lead responsible innovation projects, organizations should identify core values that prioritize human centricity, inclusivity, accountability, transparency, robustness and privacy and security.

By incorporating these principles into their products, processes and people, organizations can proactively minimize harm and build trust. The principles of responsible innovation at SAS can serve as a template for other organizations that strive to innovate responsibly.

Read more stories from SAS bloggers on equity and responsibility

Vrushali Sawant and Kristi Boyd contributed to this article.


About Author

Vrushali Sawant

Data Scientist, Data Ethics Practice

Vrushali Sawant is a data scientist with SAS's Data Ethics Practice (DEP), steering the practical implementation of fairness and trustworthy principles into the SAS platform. She regularly writes and speaks about practical strategies for implementing trustworthy AI systems. With a background in analytical consulting, data management and data visualization she has been helping customers make data driven decisions for a decade. She holds a Masters in Data Science and Masters in Business Administration Degree.

Comments are closed.

Back to Top