We hear a lot about responsible AI or AI ethics in the marketplace today. At SAS, we believe there should be a larger conversation about responsible innovation. In reality, the decisions made by AI are the outcome of algorithms, data and business processes. This means ethical considerations must be applied in each area to ensure responsible innovation.
It’s critically important to understand that analytics, machine learning and AI study the past to make decisions for the future. However, if data from the past is biased or underrepresented, then analytics could perpetuate future decisions that are also biased and have unintended consequences.
Analytics does not understand our goals in society. When we strive to improve health care, criminal justice, economic growth and the environment, we must inform analytics with our objectives to ensure we achieve goals fairly and equitably.
Furthermore, as analytics, machine learning and AI become pervasive in society, automated decisions are being made on behalf of large populations. At the same time, society and the world we live in are incredibly dynamic and undergoing continuous changes. Organizations leveraging AI must be in a state of continuous learning. Anyone who develops technology that automates decisions for others should bear the responsibility of transparent and equitable outcomes.
In my meetings with customers across industries, they’re expressing a growing sense of responsibility for making fair and explainable automated decisions. Our customers want to know they’re innovating responsibly, and they’re looking to vendors like SAS to help.
This post from SAS Blog Editor Caslee Sims covers two areas of innovation where we see ethics beginning to take a front seat: health care and criminal justice.
For instance, in health care, innovations can improve patient care, customize treatment plans and identify abnormalities in medical scans. According to Centers for Disease Control and Prevention data, before the pandemic, Black women were three times more likely than Hispanic women and 2.5 times more likely than white women to die from causes linked to pregnancy. How can we use analytics, machine learning and AI to understand the indicators and eliminate this disparity?
In criminal justice, we want to reduce crime and improve equity in sentencing. So how can we use analytics, machine learning and AI to achieve our societal goals with better legislation, policies and procedures?
We could ask similar questions about AI innovations in manufacturing, government, education, travel and tourism – and more.
Responsible innovation in practice
In response to this growing need, we established a data ethics practice led by Director Reggie Townsend. The mission of the group is to build ethical considerations into our technology roadmap and analytics best practices for our customers.
Today, our efforts in responsible innovation include four priority areas:
- Policy: Governments and watchdog organizations are currently defining guidelines for machine learning and AI. The data ethics practice is available to help customers understand, anticipate and incorporate these policies when innovating with analytics.
- Process: Most of our customers are not driven by guidelines and policies alone. They want to develop innovations that do good in the world. With this in mind, we are driving best practices that help our customers use analytics for good.
- Product: We’ve developed a repeatable process that we translate into products and solutions that scale to solve complex and data-intensive problems with ethical considerations in mind.
- People: Most importantly, we want to ensure that wherever SAS technology shows up, it doesn’t harm people but instead helps people thrive.
Currently, SAS® Viya® products already support several capabilities for responsible innovation, including protected and sensitive data flagging, surrogate model interpretation and life cycle management. We also have more innovation ahead on our road map.
Responsible innovation is everyone’s responsibility
Organizations around the world are taking a harder look at equity and responsibility in all of their business activities. Not only is this the right thing to do, but consumers also value brands that prioritize these areas.
I challenge you to ask questions beyond the primary and immediate goals of your business. How do your products and services affect your customers and larger communities? What bigger impact do you have on the outside world? Are you taking different groups and populations into consideration when you create new products and solutions? These are all critical questions.
It is everyone’s responsibility to innovate responsibly. At SAS, we are fortunate to help you shape a more equitable future through analytics, machine learning and AI.
This right here, "Analytics does not understand our goals in society." This simple point should be the guidepost with every step we take toward AL and ML innovation. I am in the midst of reading "Weapons of Math Destruction," and what some of these "black box" algorithms have unleashed is actually tragic. I firmly believe that with SAS being the greatest data analytics company in the world, we must lead on moving progressing AI and ML innovation thoughtfully and ethically.
Even in the best-intentioned system design, the designers are not going to be able accurately to anticipate or predict possible potential harm to the most vulnerable populations. A mechanism for bottom up feedback from individual users would have to be built into the system so that it could be adjusted, or possibly learn to self-correct. This is assuming that the vulnerable are able to recognize, identify, and articulate harm that is done to them, which may not be the case. A fish that has spent its entire life in polluted water has no way of knowing that there is any alternative, and would not recognize pollution as a problem.
I think it is dangerous when we trust algorithms to make too many sequential decisions automatically (without human review and intervention). I suggest that instead of worrying about making algorithm's ethical, we should be identifying decision points in the analysis process where human review and intervention needs to be built in.