A Wharton AI Research Leader's Formula for Responsible AI

**Stefano Puntoni** (0:00)
To make good decisions with data, you have to do a lot of thinking yourself without data first and knowing exactly what it is that you're trying to achieve.

**Cindi Howson** (0:19)
Hi, I'm Cindi Howson, host of The Data Chief. If you've heard me talk about data and AI before, you know that I believe everyone, not just data teams, should be able to get the insights they trust. That's why I'm proud to say that ThoughtSpot sponsors this podcast. ThoughtSpot's agentic analytics platform lets you simply ask a question in natural language and get clear, governed answers right when you need them. No fuss, no waiting. It's why companies like Cisco, Lyft, Hyatt, and Roche count on ThoughtSpot. See what the future of analytics feels like at thoughtspot.com. thoughtspot.com.
Welcome to The Data Chief. I'm your host, Cindi Howson. Today, we're joined by Stefano Puntoni, co-director of Wharton Human-AI Research and professor at The Wharton School. Stefano is a leading researcher on how AI is changing decision-making, work and society. In this episode, we'll explore why data and AI insights so often fail to translate into action and how psychological threat and change management shape the success of AI inside organizations. Stefano, welcome to The Data Chief.

**Stefano Puntoni** (1:49)
Thank you for having me, Cindi. Great to be here.

**Cindi Howson** (1:51)
Yeah. And where in the world is here today, Stefano?

**Stefano Puntoni** (1:55)
I'm in Philadelphia, in my office at The Wharton School.

**Cindi Howson** (1:57)
Philadelphia, just a little bit north of me, but I feel like I should ask you, what is your favorite town or city in Italy?

**Stefano Puntoni** (2:06)
In Italy? Oh, I'd say Florence. I grew up in Tuscany, so I would be biased.

**Cindi Howson** (2:11)
Yes. Yeah. No, Tuscany is beautiful. Beautiful. It is perhaps one of my favorite countries to go visit in the world. And there are many times.

**Stefano Puntoni** (2:21)
I recommend it to everybody for vacation. You know, Tuscany is one of the best places maybe for an AI professor. And that's, I think, you know, the US is a lot more exciting that way. But in terms of like, you know, enjoying life and food and wine is great.

**Cindi Howson** (2:36)
Oh, for sure. The wine. The wine. I'm picturing a village in Montalcino atop a hill. Yes. You'll have to come visit me in Lewis. And we will open this huge bottle of Brunello. That would be fun.
So, Stefano, you are a prolific writer, researcher. You're educating executives and youth on how AI is changing the world. I want to ask you about a point that really stuck with me, though, in one of your books, Decision-Driven Analytics, and how sometimes more data does not lead to better decisions. Why is that?

**Stefano Puntoni** (3:22)
I think that we have this almost like instinctive reaction that when we see intelligent algorithms and amazing data systems, we take them almost as an excuse for thinking less. Because the machines are so hard, then we can rely on them for insights. We don't need to think as hard. I think that's a common fallacy, but I think it's the bad one. Because in a way, the smarter the machines, the smarter we've got to be. So I advise everybody that I think to make good decisions with data, and you have to do a lot of thinking yourself without data first, and knowing exactly what it is that you're trying to achieve, framing the problem, clarifying alternative courses of option. How do you even know what success looks like here, and what are you doing in the first place? So I think a lot of these questions are not very well thought through in a lot of data analytics programs, and that ends up often being an issue, because there is a chasm being created between the decision makers, the subject matter experts, the people who are accountable for decisions, like brand managers, product managers, people like that. And on the other hand, the technical function, the analysts, the AI engineers, and people who maybe lack domain expertise, but they really have deep expertise on the tools. I think not having that good communication, not knowing why we're doing this exactly what we're looking for, then leads these analytics to be not useful.

**Cindi Howson** (4:45)
Yeah. So that's interesting. It's almost like you're saying the pre-work is more important or equally important as the insights themselves.

**Stefano Puntoni** (4:58)
Yeah. In a way, it's like, you know, you get what you put into. So if you thought very carefully about what you're doing and why, it's much more likely it's going to be helpful than if you were not. Now, to be clear, there's a lot of insights that can be gained using just data mining and exploratory analysis, just to see what you find in there. People can learn from data. But a lot of the times, that's not what you're trying to do. You're trying to know, for example, whether this ad is better than that ad, or whether this target audience will appreciate this feature or whatever it might be. So in that case, you have to have thought through exactly, what are the options, how do we know which option is best, and what kind of data do we need in order to answer that question. That is basically more of a conceptual challenge. It's not a data challenge. So it becomes a data challenge eventually, but I think first you need to have done a lot of thought.