27 Apr 2021 · 25m · Sam Ransbotham, Shervin Khodabandeh, JoAnn Stonier · 17 quoted facts
JoAnn Stonier, Mastercard's chief data officer, argues that AI is only as fair as the data it learns from. Every data set carries gaps and bias, and that is acceptable if teams know what is in it, what is missing, and design questions that fit its limits. Used with that care, AI can reveal hidden bias and help organizations become more inclusive.
The tension is between the promise of AI and the slow, unglamorous work needed to deliver it. Stonier stresses that nothing she describes is easy: it means tracing data lineage, watching for model drift, scrutinizing vendors, and building teams that bridge business, technology and policy. Her design training treats constraints as materials to work with, not obstacles.
Stonier uses 1910 voter rolls, which leave out women and people of color, to show a data set can be valid for one purpose and misleading for another. The risk comes from forgetting what is missing, not from the gap itself.
Early analytics tested whether a set condition appeared in the data. Machine learning instead asks what the data itself reveals, so any unexamined flaw in the data can quietly shape what the system learns.
Teams must watch for model drift and check whether an output is accurate or just reflects an input inequity. If an incorrect result isn't caught, the machine keeps learning from it and can amplify bad outcomes.
Stonier says that when a vendor presents a combined solution, buyers should be even more curious about the recipe inside it. Mastercard traces data lineage, quality and consistency back up the data food chain.
Stonier describes a three-sided triangle of skills, with her team sitting in the middle. Her lawyer-engineers translate between business, technical and regulatory worlds, and she says leaders must admit which side they don't know.
According to Stonier, yes, if teams understand what is in their data and what is missing, because AI can surface hidden bias. Sam Ransbotham adds that biased data is a fact that can't be retrofitted away, so you have to know its provenance.
Stonier says those records are a valid data set for studying 1910, but they inherently miss women and people of color. An AI trained on them for a different purpose would learn from a skewed picture of society.
Stonier uses the term for outputs that wander away from accurate results, perhaps because of an inequity in the inputs. If it isn't caught, the machine keeps learning from the error and can amplify bad outcomes.
Stonier says to be even more curious about what goes into the vendor's recipe. That means asking about the origin, quality and consistency of the data behind the product.
Stonier says her interior design training taught her to treat problems as constraints, like plumbing and wiring in a room. She applies the same approach to data restrictions, such as designing a fraud model using only synthetic data.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
JoAnn Stonier helped Mastercard become compliant with the European Union's General Data Protection Regulation (GDPR) while she was its chief privacy officer.
“I helped, um, the company become GDPR compliant, which really was, I think, uh, the moment in time when many, many companies were coming to the place where they needed to operationalize a lot of data risk.”
Mastercard uses a process called privacy by design to build privacy and security into its products.
“We had a whole process called privacy by design. We still use that process.”
Before Mastercard, JoAnn Stonier was chief privacy officer at American Express, where she also held other roles.
“I was previously the chief privacy officer for American Express.”
JoAnn Stonier teaches business strategy in the design management master's program at Pratt Institute.
“I teach in their design management master's program, and I teach business strategy”
Stonier says 1910 United States voter rolls inherently omit women and people of color, so they suit some inquiries but not others.
“you need to know that inherently that data set is going to miss women. It's going to be missing people of color.”
Stonier suspects airlines sized overhead bins using average heights that did not account for enough women, as an example of averages that are not fit for purpose.
“it was on very average heights, okay, but the average height did not include enough women.”
Stonier says the first generation of data analytics used rules that acted when a set condition appeared in the data.
“we were creating algorithms or analytic questions that really asked the question of this is the question. If this condition exists in the data set, then do something, right?”
Stonier says machine learning starts from the data set and asks what it reveals, in contrast with rules-based analytics.
“Instead, we take the data set and say, what does the data tell us? What does the data tell us, and what does the machine then learn from them?”
Mastercard's data teams check early model outputs for model drift and accuracy, so input inequities are not amplified.
“we're gonna be looking for the model drift. We're gonna be looking to see is this an accurate output?”
Stonier says teams should trace data lineage, quality and consistency before using data sets in analysis.
“you've got to back up the data food chain to understand your data lineage, understand the quality, the consistency of the data sets”
Stonier says data can be restricted by contract or by its type, such as financial data, which constrains how Mastercard may use it.
“Can we not use the data because it's contractually restricted, or it's restricted because it's a certain type of data, it's financial data, right?”
Stonier's team at Mastercard includes people with both law and engineering degrees, who translate between business, technical and regulatory domains.
“I have lawyers and engineers on my team, right? They have both degrees.”
Stonier says that when COVID began, many merchants moved online and sought data and guidance to make decisions.
“So many merchants were going online. If they didn't have a huge presence, they needed to create one.”
Stonier gives designing a look-alike fraud model that uses only synthetic data as an example of a constraint-driven data challenge.
“the outcome is to design a look-alike model for fraud that only utilizes, you know, synthetic data, right?”
Mastercard uses a specific framework for ethical AI and is applying it to different contexts.
“So we have a very specific framework that we're using around ethical AI, but then we're really getting into the nitty-gritty around different kinds of context.”
Sam Ransbotham says Mastercard chooses projects by strategic focus rather than choosing a project in order to do AI.
“they are picking projects based on their strategic focus rather than picking a project to do AI.”
Shervin Khodabandeh says the chief data officer role is somewhat precarious, and many CDOs last only a couple of years.
“the CDO role is actually a, you know, somewhat precarious role.”
Stonier outlines a broad mandate covering data strategy, governance, quality, AI enablement, platforms and compliance built into product design.
“embed compliance and responsible data practices right into product design”
Stonier explains how the shift from rules to learning from data makes understanding the data's condition essential to avoiding unintended bias.
“Artificial intelligence and machine learning actually flips that a little bit on the head.”
A historical example shows that a biased data set is acceptable when its gaps are known and the question is designed around them.
“you can design an inquiry that is fit for purpose”
Preparing for unknown bias requires tracing where data comes from and questioning vendor-supplied solutions.
“you need to be even more curious about what is going into this recipe”
The overhead bin example shows how averages can fail groups, and Shervin frames the lesson as a mindset change across the ecosystem.
“Averages sometimes are the worst result because they're not fit for purpose.”
Stonier describes sitting between business, technology and policy, and why data leaders must admit what they don't know.
“have to be able to admit you don't know one aspect of that triangle”
Stonier traces her path from American Express and GDPR work at Mastercard to design school after September 11.
“I got a design degree, and I'm really glad I did”
COVID forced companies online and sped up the need to interpret data, which raises the value of translating skills.
“we've been in a time compression where data and digital has mattered”
Sam and Shervin highlight that Mastercard picks projects by strategy, not to do AI for its own sake, and that data is not neutral.
“Strategy with AI, not strategy for AI.”
We hear a lot about the bias AI can exacerbate, but AI can help organizations reduce bias, too. Find out how when we talk with JoAnn Stonier, Chief Data Officer at Mastercard. Welcome to Me, Myself and AI, a podcast on artificial intelligence in business. Each episode, we introduce you to someone innovating with AI. I'm Sam Ransbotham, professor of information systems at Boston College. I'm also the guest editor for the AI and Business Strategy Big Idea program at MIT Sloan Management Review.
And I'm Shervin Khodabandeh, senior partner with BCG, and I co-lead BCG's AI practice in North America. And together, MITSMR and BCG have been researching AI for five years, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and to deploy and scale AI capabilities across the organization and really transform the way organizations operate.
Today we're talking with JoAnn Stonier. JoAnn's the Chief Data Officer at Mastercard. JoAnn, thanks for taking the time to talk with us today. Welcome.
Thank you, happy to be here.
JoAnn, let's start with your current role at Mastercard. Can you give us a quick overview of what you do?
Sure, I'm currently, as you said, the chief data officer for the firm. And, uh, me and my team are responsible for ensuring that Mastercard's information assets are available for information while navigating current and future data risk. So my team has a very broad mandate. We work on, uh, helping the firm develop our data strategy and then work on all the different aspects of data management, including data governance, data quality, as well as, uh, enabling things like artificial intelligence, machine learning.
And then, uh, we also help design and operate some of our enterprise data platforms. It's a very broad-based role. We work also on data compliance and how do you embed compliance and responsible data practices right into product design? We start at the very beginning of data sourcing all the way through to product creation and enablement. It's a lot of fun.
Because our show is about artificial intelligence, let me pick up on that aspect. Can you give us examples of something you're excited about that Mastercard is doing with artificial intelligence?
Oh my gosh, I've had so many conversations just this week about artificial intelligence, most of them centered on minimization of bias as well as how do we build an inclusive future. The conversations that really excite me is how the whole firm is really getting behind this idea and notion. I've had conversations, uh, with our, uh, product development team and how do we develop a broad-based playbook so that everybody in the organization really understands. How do you begin to really think about design at its very inception so that you're really thinking about inclusive concepts?
We've had conversations with our people and capabilities team, or what's, uh, more commonly known as human resources, talking about the skill sets of the future. How are we going to n- not only have a more inclusive workforce at Mastercard? And what do we need to do to provide education opportunities both inside the firm, but also outside the firm? So that we can create the right kind of profile of individuals so that they, um, have the skill sets that we need.
But also, how do we upskill? How do we really begin to create the opportunities to have the right kinds of conversations? We're also working really hard on our ethical AI process. So there's so many different aspects of what we're doing around artificial intelligence, not just in our products and solutions, in fraud, in cyber, um, in general analytics and in biometric, uh, solutions around digital identity. That it's just it's really an interesting time, uh, to do this work and to really do it in a way that I think needs to last for the generations ahead.
JoAnn, that's really interesting. I'm particularly excited that one of the things you talked about is use of AI to prevent bias. I mean, there's been a lot of conversations around the unintended bias of AI and how to how to manage it. But I I heard you also refer to it as a tool that can actually help uncover biases. Can you comment more about that?
yeah, but it's hard, right? This this is a lot of hard work. I do a lot of conversations both with, um, academic institutions, other org civil society organizations, right? We're early days of AI and machine learning. I think we're probably at generation maybe 1.5, you know, heading into generation two. But I think the events of this past year have taught us that we really need to pay attention to how we are designing products and solutions for society.
And that, you know, our data sets are really important in, um, what are we feeding into the machines and how do we design our algorithmic processes that we also are feeding into the intelligence? And what is it going to learn from us? And so I try to explain to people that, you know, our first generation of data analytics, we were creating algorithms or analytic questions that really asked the question of this is the question. If this condition exists in the data set, then do something, right?
So we were looking for the condition to exist in the data set and then we were acting.
Rules-based?
Artificial right, rules-based, exactly. Artificial intelligence and machine learning actually flips that a little bit on the head. Instead, we take the data set and say, what does the data tell us? What does the data tell us, and what does the machine then learn from them? Well, the challenge of that, though, is if you don't understand what's in the data, right, and the condition of that data as you bring it into the inquiry, and you don't understand how those two things fit, then you wind up with a biased perspective.
Now it's unintended, it's inherent, right, to your point. But nevertheless, it's something that we have to back up and readjust our lenses as we begin to look at our processes around artificial intelligence. So we start with the data sets, and we understand that data sets are going to have all sorts of bias in them, and that's okay. First of all, bias is always gets, like, a, you know, kind of a a very gripping reaction, right? I use the example all the time.
If you go back to, like, the 1910 voter rolls in the United States, that's a valid data set. You may use that for whatever purpose you may have for evaluating something that happened in 1910 or 1911. But you need to know that inherently that data set is going to miss women. It's going to be missing people of color. It's going to be missing parts of society. As long as you know that, then you can design an inquiry that is fit for purpose.
The problem is if you don't remember that, or you're not mindful of that, then you have an inquiry that's gonna learn off of a data set that is missing characteristics that's going to be important to whatever that other inquiry is. Those are some of the ways that I think we can actually begin to design a better future. But it means really being very mindful of what's inherent in the data set, what's there, what's missing, what also can be imputed.
We can talk about that, what kind of variables can be created by the machine, right? It can be imputed as well. And so all of those things are things that we're looking at at MasterCard. So we have a very specific framework that we're using around ethical AI, but then we're really getting into the nitty-gritty around different kinds of context.
So I want to build on that because I think it's a very critical topic to so many of our listeners. So you talked about as long as you know, you could look for it, but then the things you don't know and then you find out in the hindsight. How do you prepare for those?
Well, I think you have to be aware again, you know, for what purpose to what end? And now let's go look at the data sets. And this is where you've got to back up the data food chain to understand your data lineage, understand the quality, the consistency of the data sets that are going into the analysis. And then this is not easy. So nothing I'm saying, I want everybody to hear me clearly. Nothing, nothing I'm saying is easy, right?
Everything requires a lot of scrutiny and time, but that's easiest if you're doing it yourself. But if a vendor's doing this and is presenting this to you as a combined solution, you need to be even more curious about what is going into this recipe. Every one of these elements becomes super important. Why? Because if we do it when we're building it, those of us who are the data scientists and the data experts, the data designers, right?
We understand it because when we're gonna be looking at the information models and the outputs as they come out in the first generations, we're gonna be looking for the model drift. We're gonna be looking to see is this an accurate output? Is this truly the result, or is it because of some inequity in the input? Which is okay if it's accurate, that's okay, because the machine's gonna then start learning and we're gonna put more data and more information.
But if it's incorrect and we haven't caught it, it could get amplified to the incorrect result. And it could have, and this is what we also really care about. That's when we get to the output analysis, it could amplify really bad outcomes. And the outcomes can be significant or insignificant, depending upon if it's being designed for an individual or if it's being designed for groups, or if it's just being something, you know, I talk about how I suspect that when the airlines were first modeling the size of the overhead bins, it was on very average heights, okay, but the average height did not include enough women.
How many ladies climb on seats to get their luggage at the very back of the bin or need to rely on one of you nice gentlemen to grab our overhead suitcases down to assist us, right? Averages sometimes are the worst result because they're not fit for purpose.
I think you're talking about injecting an entire new layer in the data analysis or data usage process, Maybe right? Because was it not the case, Sam, that twenty years ago, people thought of data as neutral, like data is neutral, you use what you want, but data could be potentially harmful. And I think what you're talking about is you're trying to create a mindset shift that says it's not just data comes in, answers go out. But you have to be mindful of why you're using it, how you're using it, and this mindset shift just permeates not just your organization, but also, you're saying, your vendors and other parts of the ecosystem.
It's a huge undertaking.
it is a huge undertaking.
so th- those are some challenging aspects. What's rewarding? What kind of exciting things happen? Uh, these these seem a little bit sort of like, oh gosh, uh, we have to worry about this, have to worry about that. Where's the get us excited about the fun part?
Oh, I think all of this is fun. I do so, I'm, you know, but I'm a data geek, right? So what's fun is actually seeing my firm come to life around this. These are actually very exciting conversations inside of the firm. I don't mean to make them sound like just risk conversations at all. It's intellectually challenging work, but we all agree that we think it's going to bring us to better places and already is bringing us to better places in product design.
You talked about the workforce aspect of it and that technology is necessary, but is is far from sufficient. We heard from one of our other speakers, the president of one hundred, hundred flowers about learning cautions, the desire to wanna learn. How do you bring all of that together in terms of the future of the workforce that you guys are looking at and the kind of talent and skill set and attributes that are are going to be successful in different roles?
Are you are you taking actions in that direction?
We really look for people who also can design, you know, controls and processes and make sure that we can navigate those. You know, w- we live in a world of connected ecosystems, and so we're only as good as our partners. But it means those handshakes, right? Those connections are super important to understand. And so how do you make those inquiries of other organizations so that you can create those connected ecosystems that are only going to grow in size and scale for the future?
And how do we help design those? How do we be the leaders in designing some of those is is really, really important for us. So it really is this meshing. Um, I talk about being like a three-sided triangle of data skills,
Is there a different kind?
but we sit in the middle, right? We sit in the middle of the business, we sit with technology skills, as well as there's an awful lot of policy that we need to understand. Often, it's the other design constraint of where are laws evolving, where is the restrictions, the other type of restrictions that we need to know. Is it gonna be on soil? Can we not use the data because it's contractually restricted, or it's restricted because it's a certain type of data, it's financial data, right?
It only can be used for one specific type of purpose, or it can only be used at an aggregated level, for example, it can be used at a segment level. Those type of restrictions as well, at that compliance level, also needs to be understood. So sitting in the middle of the middle is sometimes really hard, so that's why you have to be able to admit you don't know one aspect of that triangle.
so you mentioned the word design well a couple of times at least. Uh, can you tell us a little
here it comes.
bit about can you tell us a little bit about how you got to your role? You know, cause I I think that design thinking is coming from your your background. Uh, can you share with with us how you got to where you are?
I think it's probably easier if I just go quickly backward. So prior to being the chief data officer, I was the chief privacy officer for the firm, which was a lot of fun. Um, I enjoyed that role immensely. I helped, um, the company become GDPR compliant, which really was, I think, uh, the moment in time when many, many companies were coming to the place where they needed to operationalize a lot of data risk. And we had been on that journey.
We had a whole process called privacy by design. We still use that process. Lots of companies use that phrase cause it's it's a regulatory phrase. But we had already been looking at our products and solutions to try to understand how could we embed privacy and security into their very design and into their fabric, and we still do that to this day. But as we were doing the compliance work and planning it out for the GDPR, the European privacy law, we were realizing we were gonna need additional platforms and systems to be built.
And as I was doing that, I kept saying to anyone who had listen, who's gonna own all the data? Cause I really need I need somebody to speak with. And they just kept patting me on the head saying, you just keep designing everything and it'll be fine. Well, okay, so here we are, you know, that I'm the chief first chief data officer. But it was kind of an an expansion and then a severing of a lot of the work that I had been doing.
But I came to be the chief privacy officer. I was previously the chief privacy officer for American Express. I came to that role after a pretty good-size career at American Express as well, working in a variety of roles. I understood how that firm works. So financial services is something I've been doing for a while. But the design piece comes in because of nine eleven. I had the misfortune of being downtown that day, and the American Express building is r- right across the street from the world what was the World Trade Center.
And, um, so saw a lot, um, and lost colleagues that day. And we were all relocated for for several months while, um, the building was repaired. And so on New Year's of the next, I don't know, it was probably, like, January fourth, actually, of 2002. I was thinking, well, there's a lot of people who were not alive that New Year's. And I thought, Well, what can I do? What haven't I done in my life that I really wanna go accomplish?
And so I decided to go back to design school. Um, this is after law school, so I got my law transcripts were part of the process, which was fun getting them to send my transcripts to a design school. But I got a design degree, and I'm really glad I did because it's made me a better business person and it's given me a whole different way of thinking about problem solving. And then I've had the good fortune.
So I I do, yes, I do interior design work, and yes, I can help you with your kitchen and your bathroom and all the other things you wanna do during COVID. I've helped several colleagues pick out tile and other things.
all right, we'll talk after the podcast You can talk about that. after the podcast All right.
about that. Um, but I also then had the good fortune to meet the Dean at Pratt Institute, and I teach in their design management master's program, and I teach business strategy, and I've taught other courses in that program. And that's also helped me really evolve my design thinking as well. And so all of these things, yes, I have a law degree, yes, I'm a data person, but having the design thinking as well really makes you not think of things as problems, but just constraints around which you have to design things and that they will shift over time.
So it's no different than, well, the electric is over here and the plumbing is over here, and you only have so much space. So how can you utilize this, right? It's the same thing for, well, I can only use the data in this way, but I want to achieve an outcome. How do I do that? And so it is the same kind of strategic toolkit. You just kind of flip it for, okay, well, the powder room has to be only this big, so that's one challenge.
Or the outcome is to design a look-alike model for fraud that only utilizes, you know, synthetic data, right? Um, or something along those lines. But it does kinda give you a little bit of a can-do spirit, cause you figure something has to be possible because there's raw material in the world. So that's kind of how I approach things.
It's very well said.
That's a very weird triangle. It's a triangle with some very My, yeah, my weird.
triangle is probably very weird. I I get that a lot.
Although that's a bit of a theme, though, you know, we when other people we've talked to, they they're bringing in these experiences from other places, and very often they're not super technical things. We're talking about artificial intelligence, you might think that it would quickly go down a path of extreme technical, but it, I think, is just as likely to go down a path of design.
Yeah, no, uh, what we find on my team, and and it generally is that if you find people that have kind of that combination of right brain, left brain. So I have lawyers and engineers on my team, right? They have both degrees. It's a really interesting match of being able to translate the business to the technical, the technical to the business, right? Translating the the regulatory to the business or the regulatory to the legal. It it kind of works as a giant translator role.
And if you think about it, that's kind of the moment that we're in right now is the ability to be fluid in translating concepts from one domain into another. And I I think it works. Um, but I I do think that some of those competencies are gonna be equally important or more important as as we continue to evolve, as we begin to develop. There'll be the core skill sets that we've always had, uh, you know, leadership skills and problem solving skills and analytic skills.
But I do think that the ability to translate that and then also be able to derive from the output, right? Whether it's from a dashboard or reports or whatever, I think that's also gonna be really important business skills for any, uh, businessman or woman.
You said right now, you said, like, that's the place we are right now. What do you think is cause that? What's making that so important right now?
I mean, I I I don't know that I was emphasizing that, but I think when I look over the past year and I think about we're just on the year mark of COVID for us. You know, I can remember when we went virtual, I remember so many of the customer and client calls that I was helping to field at the time. Companies were in a a moment, right? So many merchants were going online. If they didn't have a huge presence, they needed to create one.
Business models were really shaken. But everybody was looking for data to try to help make some decisions, right? And they needed guidance. If they didn't have the data, they were looking for data. If they had the data, they were looking for guidance on how to interpret it. If they knew how to interpret it, they were looking for look-alike information to make sure they were making the right decisions. So if you just look at that maturity curve, we have been thrust forward in time.
So I think it would have been happening, maybe in a less compressed way. I think we've been in a time compression where data and digital has mattered, and so we can't undo that. And so I think these skill sets matter more and more just because of the compression we've been under as a society. And so we're going to see that be part of what this next normal or next generation, cause I don't really like the word normal, next generation.
This is now part and parcel of how we interact with each other. We will never go back to just being in person. This kind of, you know, connecting digitally will always now be part of our mandate, right, and part of the toolbox. So I just think that there will be more of a need to be able to interpret and to utilize than we had a year ago.
That seems like your superpower is taking some situation, whether it's nine eleven and turning it into design school, or taking a pandemic and turning it into a more fluid approach to things. It sounds like your superpower in many cases. Thank you. I'll Yeah,
I'll take that one.
JoAnn, thank you so much for talking to us. You really picked up on a lot of points. We've covered a lot of ground. Thanks for taking the time to talk with us today.
Yeah, thank you so much.
You're welcome. Thanks for having me.
Sure, and that was great. JoAnn covered a lot today. What struck you as particularly interesting?
I mean, a lot of things struck me in a positive way. I liked how she talked about design thinking, she talked about emotional. And I actually think that's been part of her secret sauce and part of Mastercard's success with these kinds of initiatives. Because, like, if you look in the industry, the CDO role is actually a, you know, somewhat precarious role. Because, you know, there's very successful CDOs, and there's many CDOs that just don't have that vision or that ability or that autonomy or sort of that multi-sided triangle as she talked about.
And so they don't succeed, and they they last for a couple of years and and so I think there's something to be said for that. What did you think about Sam on her comments about implementation, the challenges?
Yeah, exactly. I thought it was interesting that they've not changed what they're doing in terms of AI projects because of some of the the fact that it's hard. I think they they recognize They've doubled down that it's hard
on it, yeah.
and and they've doubled down on it. And I think that's, uh, that in and of itself is a little bit of a a validation that what they're doing, they believe in that. Even though it's hard, they're they're still continuing. And and I think they are picking projects based on their strategic focus rather than picking a project to do AI. And, you know, that's, uh, something that just keeps coming up. Is and no one's out here saying, Hey, you know, let's do some AI today.
It's so it's Monday morning, let's do some AI. These people are are trying to deliver on things that are in their organization's strategy. And it turns out that in many cases, AI is is the tool. But if it's not, that's okay, too. But it often is.
Strategy with AI, not strategy for AI.
Where did I read that? Yeah,
Yeah, I know, right, didn't you write it?
that was all David.
The other thing I thought was quite insightful, and she went there very early on, is about trust and ethical use of AI and sort of reining in the AI solution to make sure it's ethical, but also using it to uncover hidden bias so that you could be more inclusive, you could be more aware of the workforce of of your customers of the ecosystem. What I liked was particularly insightful there was the need not to view all data as equally neutral.
And that depending on where the data goes to, very harmful things can come out of it, which necessitates another layer. I mean, she hinted on it, right? There's a layer of awareness and governance and a mix of tech artifacts as well as process protocols to make sure that the outputs are ethical and unbiased. And she talked a lot about the need for mindset change for that to work, that people have to ask, What am I using it for?
What is it gonna do? What could go wrong? And that constantly probing for things that can go wrong so that you could preempt it.
But also, sh- she saw it as a tool to prevent bias. I mean, we we've seen so many stories out there about, Oh, this AI system's biased, that AI system is caused bias. And I'm not denying those are true. I'm not a biased denier. I wanna it's just there's a there's an opportunity that she saw in that data to to rectify that. They are a fact, and there's not gonna be a no no amount of retrofitting is gonna take that away.
But she also said, We don't ignore the data either. Uh, we just have to to know what that provenance is.
I think that's a great point, Sam.
Thanks for joining us today. Next time, Shervin and I will talk with Chris Couch. Chris is the senior vice president and chief technology officer at Cooper Standard. Please join us.
Thanks for listening to Me, Myself and AI. If you're enjoying the show, take a minute to write us a review. If you send us a screenshot, we'll send you a collection of MITSMR's best articles on artificial intelligence free for a limited time. Send your review screenshot to SMR feedback at MIT dot edu.