AI and Data Ethics: Accountability and Transparency

businessman uses looking glass to look at computer code

Continuing our series on AI and Data Ethics, this conversation with Don Weinstein is about accountability and transparency.

In this series on artificial intelligence (AI) and data ethics, we're talking about ADP's ethics principles of explainability, data governance, ethical use of data and privacy, and accountability and transparency.

We talked with Don Weinstein, ADP's Chief Product and Technology Officer, about accountability and transparency.

What does accountability and transparency mean in the context of AI?

Weinstein: Accountability means we strive to do the right thing and make sure that the ways ADP data are used are both useful and fair. We have to ask questions about what could go wrong, how we can detect and manage for bias, and how we can respect and protect people's privacy.

We also want to give people the tools and information to make better decisions. But AI systems don't give us answers. Instead, they help us ask better questions.

This is where transparency comes in. Transparency means we tell users what factors go into the analysis they're getting. That way they can see where it applies and where it doesn't.

With AI, there's an opportunity to learn and understand things that were not possible before. This gives us richer insights and reveals questions and approaches we may not have thought of. It also allows decisions to be based on better information and hopefully, made with more confidence.

How does ADP incorporate accountability and transparency into its systems and products?

Weinstein: We're helping clients understand what they're getting by exposing the factors that go into our analysis and calculations. When you know which ten things we looked at, you can tell what matters to you.

For example, our turnover predictor can give you the probability that someone will leave based on different factors such as time in role, tenure with the company, comparative pay, and commute distance. Based on this data, we can tell you that someone has a 27% probability of leaving in the next year. Depending on your industry and organization, this could be high or low risk. Even if it is a high risk for you, there's still a 73% chance that they won't leave.

If you decide this is someone you want to keep, then you need to know what went into that 27% prediction so you can figure out how to approach the issue. But we won't be able to tell you whether that person's partner got a transfer and they are moving to Spain or whether a promotion would work to keep them. Employment decisions are complex, just like people.

We can't see the future (besides the parts we're making). We can give our clients tools that offer relevant and useful insights along with the information they need to use those tools effectively.

Why is accountability and transparency an important part of data and AI ethics?

Weinstein: We understand that our products are used in connection with decisions about people's careers and lives. ADP started by helping our clients make sure people's paychecks were right. We have always been aware that what we do affects our clients and their employees in important and fundamental ways.

We also have a lot of sensitive personal data about people and protecting privacy and data security is fundamental to everything we do at ADP. As we moved into machine learning and exploring how we could get deeper insights from the data we have, we wanted to do it as carefully and thoughtfully as possible.

I've also talked to a lot of providers who make claims about what they can do with data that are just not true, and in some cases not possible. It's data malpractice.

We are very careful about overstating what we do and don't know. We continuously test, monitor, and improve our models so changes to the data or a model don't inappropriately impact the results.

We also created ADP's Data and AI Ethics board. We've included outside people to bring different perspectives. We've also developed ethics review processes and resources that are a required part developing all our solutions that involve data and machine learning.

As much as we love data and technology here at ADP, we know our tools affect people. We take this responsibility seriously and ethics are an important part of getting it right.

Learn more about ADP's privacy commitment and read our position paper, "ADP: Ethics in Artificial Intelligence," linked from the AI, Data & Ethics blade on the Privacy at ADP page.

Related articles: