There is a sentiment out there among some that data governance, defined by Informatica, as a set of principles, standards and practices that ensures an organization’s data is reliable and consistent, is both dull and boring.
According to Carter Cousineau, vice president of data and model governance with Thomson Reuters, that is completely untrue, especially today in a world dominated by artificial intelligence (AI) and machine learning (ML) advances that have become almost a daily occurrence.
Speaking last week at the second annual Analytics Unleashed event, organized by IT World Canada and sponsored by SAS, Informatica and shinydocs, she said good governance is not just about complying with regulatory mandates, but about improving how an organization operates.
She kicked off the presentation with what she described as a “few fun facts,” courtesy of McKinsey & Company, that revealed data driven organizations are six times more likely to retain their customers, the most innovative companies are 2.4 times more profitable and upwards of five million businesses are actively using WhatsApp to connect to customers.
All that data being produced means rules must be in place, but Cousineau said it is incorrect to view governance as simply a collection of policies and standards developed after a series of endless meetings.
An example of that from an operational sense would be a marketing department and governance: “marketing must create standard guidelines and/or principles on how to do successful, consistent branding.
“That’s the beginning of governance — you do want structure around policies, standards and templates — but then, as soon as you have that consistency, it is then when you can start to unleash the true power of governance in terms of how it fits in your organization and allows you to be agile to ever-growing changes.”
However, before that happens, she told the live and virtual audience, a key question must be asked, namely, “do you trust your data?”
There is a whole subset of additional questions that also need to be asked, said Cousineau, and these include: What is it? Where did your data come from? What are you assuming to be true about the data? What consents are in place? Do you know what was added, changed, filtered out or deleted? Why was the data collected?
This must all be correlated, answered, and followed, not only to make sure an organization follows data governance rules and policies, but that it is ethical, particularly when that data is used as part of an AI configuration model.
Good or ethical data has the ability to help solve problems, she said. An example of that, she said, occurred during the pandemic, when money launderers developed new ways to operate since the bulk of bank branches around the world were closed. She referenced an article from the MIT Technology Review that discussed how lawmakers and financial institutions “have built an AI model to help forecast and identify frauds in advance of them occurring.”
On the flip side, what happens when ethics in data and AI has not been considered? Again, this exercise also took place at MIT and is known as the trolley problem. As another article from MIT Technology Review notes, “in 2014 researchers at the MIT Media Lab designed an experiment called Moral Machine. The idea was to create a game-like platform that would crowdsource people’s decisions on how self-driving cars should prioritize lives in different variations of the trolley problem.”
Cousineau said it “brings back that ethical dilemma of ‘should the driverless car kill the baby or the elder?’ The answer is no one, of course. But when you are facing those decisions as a human, how would you behave? They created this gamified view to better understand humans’ decision-making process in that very question.”
What MIT lab found is that the answer depends on where you are from. Those in Asian countries selected the elder, or more frequently were to select the elder, and those in North American countries selected the baby, she said.
“The reason this is such a big question or important ethical dilemma is that when you feed that data into a self-driving vehicle, how would it behave? And you can see in the study of human behaviors how, depending on the countries or jurisdictions you are in, that behavior may be different.
“The decision of that model or data set for the self-driving vehicle would be different. There are cultural differences within our society, and we cannot neglect those. We must ensure our data and models and systems are fair, because that is the data that is fed into them.”
Another area she mentioned involved AI in healthcare.
“This is a big bucket of opportunity, but at the same time, you do not see the full adoption, and the question becomes, why, despite the advanced research that we have seen in cancer treatments? We know that there are technologies around AI that can detect cancer rates far quicker and easier than that of doctors. Why are they not adopting? Doctors cannot take that technology unless they fully trust it, they need to be able to trust that technology 100 per cent or they will not adopt.
“That’s something where even though we’ve seen advancements, you don’t actually see that curve and shift in terms of actual adoption.”
As far as the global policy landscape is concerned, Cousineau said that when it comes to legislation, what happened in the world of privacy will be mirrored when it comes to ethics.
“If you look at the history of privacy, we are at that same trajectory around ethics. Privacy had a lot of different regulations that were forming in different geographic locations and regions.
“We see the landscape coming around ethics. We know these ethical concepts are being established from a regulatory standpoint, but they may not be speaking commonly to one another. And this is like what we saw in privacy until it reached that final convergence.
“From a policy landscape, I expect the same trajectory. We will see this divide, or slight variance between regulations from countries, to the point it reaches that convergence. And that will be that go-forward from a global perspective. Right now, as it stands, regulations are not as closely tied as they could be.”
