What We Get Wrong About AI Expertise

In this podcast episode, Dr. Beverly Wright, Vice President of Data Science & AI at Wavicle Data Solutions, is joined by Fouad Bousetouane, Professor and Chief AI Officer at the University of Chicago and 2ndsight, examine the complex and often misunderstood notion of making “everyone an AI expert.” From defining what true AI expertise really means to highlighting the dangers of solution-first thinking, this episode challenges common assumptions about democratizing AI.

Speaker details:

• Dr. Beverly Wright, Vice President of Data Science & AI at Wavicle Data Solutions
• Fouad Bousetouane, Co-Founder and Chief AI Officer at InterspectAI

Watch the full podcast here or keep scrolling to read a transcript of the discussion between Beverly and Fouad:

Beverly: Hello, I’m Doctor Beverly Wright and welcome to Tag Data Talk. With us today, we have Fouad Bousetouane, Professor and Chief AI Officer at the University of Chicago and Second Sight, respectively. And we’re talking about the challenging nature. Oh, I’m so excited to talk about this, the challenging nature of making everyone an AI expert. Thank you for being here.

Fouad: Thank you for having me, Beverly.

Beverly: Excellent. Well, let’s start off with a little background. Tell us, why are you so cool?

Fouad: That’s a tough question, but.

Beverly: Loaded question.

Fouad: Yeah, absolutely. I believe I am cool because I can wear multiple hats. The first hat is the hat of the AI scientist. My curiosity is inbounded. I love working on problems, tuning models and going beyond what is feasible. I like exploring obvious problems to apply AI and AI modelling.

The second hat. I love wearing the hat of business minded AI leaders in which I love working on business case problems, driving business value, leading a team, also leading cross functionally and embedding AI within every and every business process. The third hat, which is also cool. The educator. That’s why I’m engaged with the University of Chicago. I love sharing my acumen, my knowledge, guiding and coaching the next generation of AI builders.

Beverly: Nice. I love it. So, kind of, what people used to call the analytics unicorn plus an educator on top of that. So, you like to coach others up and lead the next. I love that. Very nice. So, we’re talking about the challenging nature of making everyone an AI expert. And I know we’re not really going to make everybody an AI expert, but where does this title come from? What are we talking about with this AI expert?

Fouad: Yeah, like nowadays, especially with the hype of generative AI, everybody’s getting interested in learning about AI and sometimes things go a little bit out of control. Like sharing some knowledge, which is a good thing. However, we have to be cautious of the way we approach AI and who we identify as an expert AI or AI practitioner. Let me take a step back here and define what’s an AI expert is.

Beverly: Exactly.

Fouad: AI expert, I believe is a problem solver, a scientist and engineer who is capable of mapping the right problem, real world, the problem to the right AI modelling strategy. Thinking backward, living in the problem space, I understand the dynamics of the problem reflected in hidden patterns within the data and map that to the right AI modelling strategy.

It could be when you say AI, we say ML, we say computer vision, pretty much cross spectrum of AI applications. If somebody is capable of doing that, then they have deep knowledge of how to solve a problem leveraging AI and what they have seen recently. It’s a little bit scary. Lots of AI enthusiasts and some professionals are doing solutioning. They don’t go from problem to solution; they start going from solution to problem.

Beverly: I see.

Fouad: Which can create pretty much suboptimal solutions. Sometimes some people are overpromising the business, and the outcome is not as expected. And when someone gives this impression to their business partners, they will lose trust in AI capabilities and applications.

We have to be really cautious. LLMs are very cool tools, capable of many things. Yeah, but LLMs are a general purpose. And they don’t apply. You can’t apply them out-of-the-box vertically. That’s why we don’t say for each problem, LLM can solve it. Let’s understand the problem first. Maybe a small, tiny model can solve it without going into the complexity of LLM.

Beverly: The beauty of Occam’s Razor that nobody pays attention to because they want to do something cool and complex.

Fouad: Absolutely. That’s true.

Beverly: If I’m hearing you right, first of all, you mentioned AI is a broad spectrum. This could be natural language processing. This could be computer vision. This could be predictive and machine learning. This could be LLM, LMS, of course. And so, it includes a wide range of possibilities, robotics, maybe even. And then the second thing you mentioned is to be an AI expert.

That’s a very interesting definition because you’re saying if you are able to hear the business problem, frame it and know what is the potential solution, even if it’s not necessarily an AI solution, but that makes you more of an expert if you can understand the problem and then know what the solution that’s needed.

Fouad: Absolutely. And go beyond that, like mapping a problem to a modelling strategy, because a modelling strategy doesn’t drive directly to the solution. You can go through multiple cycles of baselining, fine-tuning, and the first of all of these steps is enumerating solutions. Having that mindset, it could be organically embedded within your mental model, or it could acquire knowledge that puts you in that position of approaching problems from the lens of AI.

Beverly: If I’m hearing you right, my husband’s a woodworker and he goes to this place called Synergy Mill. They have all this stuff, and it’s a community. It’s way more than anybody could have individually. And so, they all just kind of share this equipment. So, if you’re looking at a piece of wood and you’re trying to turn it into something, you look at it and go, what are the tools I need to make this what it needs to be? That’s what you’re saying. And the difference being that some people are walking in with a hammer, and they’re like, where do I hit it?

Fouad: What do I hit it? Where do I hit it? Right? We have an LLM. LLM is really cool.

Beverly: Yeah, let’s use it.

Fouad: If even before identifying the tools, you understand the nature of the piece of wood, that’s how you can map it to the right tool.

Beverly: Right.

Fouad: What’s the thickness? What are? I’m not an expert in this domain or field, but you need to understand what you are dealing with. And if you make an analogy with AI or the problem, that qualifies to be a good problem for AI to solve it. You understand the data, the dynamics of the data, how the data looks like, what do I need to do, what are some hidden patterns I can either discover or start like plain with the data. It’s like understanding that the problem is half of the solution.

What they have seen also, and it is really common, is now the business partner shows up with potential use case and they bring a solution with that saying you know, you can use LLM for this problem. Like maybe this is not the mindset and that’s the risky part of everybody getting excited about AI, reading things about AI. But we have to be cautious on how to apply, when to apply, which problem qualifies for which modelling structure.

Beverly: Well, I think a big part of what you’re saying is just start with the actual problem.

Fouad: Absolutely, don’t.

Beverly: Don’t start with the solution of like, here’s what I want to do. I want to use this hammer and why do you think people are doing that? What is the driving force, no pun intended behind. We have to use this method or I’m dying to use Gen. AI. Help me use Gen AI and then they come up with a problem to try to fix it I mean.

Fouad: Yeah, when I say start with the problem, live in the problem space. You have to breathe the problem, you have to understand the problem and what are the requirements, what are the restrictions, and what business partners are expecting from you. Yeah, these are very important, like key parameters you extract from the environment right from the problem to map it to the right solution or tool.

Yeah, going back to why everybody is solutioning, is because of the trend like if you want to sound cool as a business partner, as an AI or ML person, you will tell your business partner like we are using Gen AI. Everybody’s talking about LLMS, everybody’s talking about Gen AI, it sounds cool. I want to make sure people believe and see and feel that I am on top of all these trends. It’s pretty much becoming a unique and important skill to have and to showcase in a business setting.

Beverly: So, being able to hear the business problem and then they know how to push a button to use the method that they’ve already got in mind is not really making you an AI expert, is it you’re saying?

Fouad: No.

Beverly: And going back to the woodworking example, I guess that explains why woodworkers are always like, oh, that’s this kind of tree and it has a disease and here’s how old it is, and all these things matter. You have to immerse yourself in this.

Fouad: You kind of press the button, and you can come up with an AI workflow or AI application, but it’s not going to necessarily work with the level of accuracy you are expecting, right? What you will end up doing is finding some hacks, some tweaks to make the solution work for the problem instead of understanding the problem, building the solution. Like if you go from solution to problem, you end up with a suboptimal solution that may not fit within the business.

Beverly: Right. And it may not even be the problem that you wanted to solve. Informs has a certified analytics professional designation, and different domains. In this process they call business problem framing. And then after that they have analytics framing. And I felt like so many people weren’t paying attention to it. I actually wrote an article on business problem framing. So, I totally hear you. I get what you’re saying, but surely, we’re not saying AI is only for PhDs. You know we’re not kicking people out, are we?

Fouad: No, absolutely not. I believe that AI is for everyone. AI is a very exciting technology and strategic. It is going to be embedded not only in our lives but also in all business processes. The way we are going to operate post AI and before AI is not going to be the same, I believe.Beverly: Do you think it’s a big transition?

Fouad: It’s like this paradigm shift.

Beverly: Pivotable, OK?

Fouad: And also, what I think is we don’t have a choice. Nobody has a choice to adopt or not adopt AI. It’s a necessity. Everybody should know and understand AI at some level of complexity, right? But what we need to design or embed within businesses is a wide spectrum of core functions that are dedicated to different aspects of AI. We draw the line between AI expertise, AI practitioners, between AI business.

Beverly: Leader. Consumer.

Fouad: Also, the prompt engineers work with AI testers. We need lots of testers down the road because now the biggest fear of companies putting AI and Gen AI in front of customers is the vulnerability of these LLMs. They hallucinate, they are biased, you can jailbreak quickly, you can make them toxic, and you can manipulate them. That is some layer of complexity. That’s why we have to define some level of expertise to make sure what we build is trustworthy.

Beverly: Yeah. Are there natives and immigrants in AI, or are we all sort of immigrants?

Fouad: There are some natives, lots of immigrants.

Beverly: Yes. OK, OK, gotcha. Yeah. And I think it was Shannon Harlow who first talked to me about, she said do you think AI is going to be convergent with data science or divergent? And I said both.

Fouad: Both.

Beverly: I think, I think it’s going to be both in that some people that kind of grew up, they learned SQL, they understand data, they know analytics and business intelligence systems, and they learn modelling, and now they’re in AI. You know, it’s kind of like all these experiences that you have kind of lead to that. And so now if it is hallucinating or doing something wonky, you can look at it and go like, I don’t know why it’s doing that. Whereas the immigrant doesn’t have that and they just come straight into the AI door. What’s the problem with that?

Fouad: If you look at the landscape of AI nowadays, everybody’s talking about the vertical agentic AI or vertical AI agents. And what is a vertical AI agent? It’s pretty much like an autonomous entity that is capable of reasoning, taking actions and adapting to the environment leveraging real world context of the environment. To be able to get to the point where this vertical intelligence is embedded, we have to fine tune these models. To be industry specific. The cycle of fine tuning the model relies on which model to fine tune.

Why? The other very important piece is the data integrity strategy. What kind of data are we going to leverage to fine tune a model? And the entire cycle of fine tuning, prepping the data, fine tuning, human in the loop, alignment with the model and data, and also continual training requires in depth expertise to make sure we hold this expert accountable. Because the data fits in with the model.

If it’s reflected into some hallucination or the model is not making the right decisions, at least we know who did what and to at least adjust what layer of accountability we have in place. It could be a human accountable for the outcome of AI because the data is not good, is not well curated, and is not embedded in fairness. Or it could be AI itself because of the limitations of the machine.

Beverly: Well, this feels like an overwhelming task. Like, what are we going to do? I mean, one thing you talked about was that we’re not really going to have a choice. Like AI, it is here. It’s not going to go away. And I’ve heard people in healthcare say, look, if you are going to need future surgery or some, I’m just making this up. But this is the analogy that they provided in healthcare is if you’re going to need surgery in the future, doctors are going to learn how to inject nanobots into your bloodstream.

And those nanobots are going to go work on your organ, and Med schools are not necessarily going to know how to do it. You can’t elect a traditional versus a non, you know, a non-nanobot surgery versus a nanobot surgery. And so, you may not have any choice. So, we’re in this. So, what do we do? Are we trying to say how do we solve this? Do we want to bring awareness to AI literacy? Is this an AI literacy problem?

Is this a problem framing like back to the fundamentals of understanding what’s going on, you know what kind of problem you’re trying to actually solve and working in a more scientific method kind of format for understanding the problem, find the right tool, come up with a solution instead of the other way around? Or are we saying you need to be a consumer, and you guys need to be the developers? You know, what’s the solution? How do we solve this?

Fouad: That’s a very interesting question. What I would say is that it’s a mix of everything. We have to level up. For sure, like I said, AI is a necessity. We don’t have a choice.

Beverly: So even for people that are like, I’m really good at supply chain, but even though that’s my expertise, I got to learn AI too.

Fouad: So, everybody has to take it to the next level. If I write emails, like all day long, I use AI to make them better. It goes all the way to simple tasks. Yeah, we are there, there is no way around it. Everybody should understand how to interact with AI, and how to maximize the output of AI. And we go to the cycle of prompting, how to prompt, how to design a question, and how to engineer it. I think this is a skill everybody should have. Besides levelling sets, we have to build a framework.

The framework is going to define, like I said earlier, what immersion functions within a corporation within a business organization, where and what duties and tasks are for each and every individual to accomplish. Otherwise, if you don’t draw the line between expertise and user and practitioner and business developer versus a product manager versus testing AI, it is going to be really, really challenging. Another aspect they believe is also very important is for all corporations or businesses to start the journey with AI. Even with small steps, it is very important. Nobody should be left behind. Is this the right time, if not adopted, to start the journey with baby steps?

Beverly: Yeah, love it. Well, that kind of ties into my last question, which is what final piece of advice would you give? Because we’re talking about, you know, this AI expertise and having to embrace it. So, a lot of people might feel a little overwhelmed about this environment because it’s everywhere. I actually had a student one time you were talking about teaching.

I had a student say, why don’t I just change my major to something that doesn’t involve AI? I mean, it’s in the arts, it’s in the sciences. It’s everywhere, right? It’s in business, of course, but what final piece of advice would you give people that are trying to better understand how to embrace, you know, and how to harness AI in meaningful ways?

Fouad: Yeah. Before answering your question, there is another point. They wanted to highlight this, but they have noticed this and it’s common across large corporations. There is some sort of a gap between the leadership and the operations team. There is a lack of understanding of problem complexities and especially with AI. You have this ML organization living in a different space from senior leadership, from the vision of senior leadership.

And I believe senior leadership should like to do some job and understand the different dynamics of AI and embed that within their vision. And I strongly believe in the future we will have more technical leadership, empowering technical folks, and AI experts to also take on end-to-end life cycle development of products.

Beverly: Wow, Rise of the nerd? I mean, this is such a paradigm shift, too, because it used to be that whoever had the good instincts, you know, who had the strong gut, who just kind of had that intuition and just knew those were the ones that typically got promoted. But you’re saying like you start, you’re starting to see, or you think that in the future we’ll start seeing CEOS that started off as you know, technical?

Fouad: I believe it’s a necessity. As a quick example, how to one expect the manager to set the right expectations for a PhD machine learning scientist? If the manager doesn’t understand the complexity of the problem, the complexity of training the model, the dynamics of fine-tuning, and hyperparameter optimization. If the manager is not there, there is a gap, and that gap is going to impact the organization is going to impact talents. Is keeping talents engaged and happy and retention. You know, it’s huge.

Beverly: Yeah, for sure.

Fouad: Like I believe empowering identified talents, people who are really technically sound, put them into management, like training and empowering them to lead visions and to lead product development.

Beverly: Nice, very nice. I love this. Well, thank you again so much to Faoud Bousetouane. Did I say that right, Bousetouane? I’m so proud of myself. I’m so proud of myself. I’m doing great for talking to us about the challenging nature of making everyone an AI expert. Thank you.

Fouad: Thank you.

As AI becomes ubiquitous across industries, the path forward isn’t about turning everyone into an AI engineer, but rather fostering AI literacy, thoughtful problem framing, and responsible integration. Understanding the problem is half the solution. AI’s future success hinges on empowering the right people with the right tools while building frameworks that support clarity, collaboration, and accountability.

Explore the full catalog of TAG Data Talk conversations here: TAG Data Talk with Dr. Beverly Wright – TAG Online.