The Human Code

Unlocking Business Potential with AI: A Conversation with Ari Kaplan

Don Finley Season 1 Episode 41

Send us a text

Exploring Human Behavior with AI: A Conversation with Ari Kaplan

In this episode of The Human Code, host Don Finley sits down with Ari Kaplan, Databricks' global head of evangelism and a leading influencer in data and AI. They discuss Ari's impressive career in sports analytics and AI, the integration of AI in business, and the future of human-machine interaction. Ari explains the importance of capturing human behavior through AI and shares insights on how businesses can stay competitive by integrating AI effectively. They delve into AI's role in enhancing business decision-making and the significance of using AI to process both structured and unstructured data. The episode highlights the advancements in AI technologies and their potential to revolutionize how we interact with technology.

00:00 Introduction to The Human Code 

00:49 Meet Ari Kaplan: AI and Data Influencer 

01:09 Exploring AI's Impact on Business and Society 

01:50 The Intersection of Humanity and Technology 

05:59 Challenges and Opportunities in AI 

08:12 The Role of Data in AI Advancements 

13:04 The Future of AI and Business 

16:31 Data Intelligence Platforms and Governance 

28:39 Vision for the Future: AI in the Next Decade 

32:35 Conclusion and Sponsor Message


Sponsored by FINdustries
Hosted by Don Finley

Don Finley:

Welcome to The Human Code, the podcast where technology meets humanity, and the future is shaped by the leaders and innovators of today. I'm your host, Don Finley, inviting you on a journey through the fascinating world of tech, leadership, and personal growth. Here, we delve into the stories of visionary minds, Who are not only driving technological advancement, but also embodying the personal journeys and insights that inspire us all. Each episode, we explore the intersections where human ingenuity meets the cutting edge of technology, unpacking the experiences, challenges, and triumphs that define our era. So, whether you are a tech enthusiast, an inspiring entrepreneur, or simply curious about the human narratives behind the digital revolution, you're in the right place. Welcome to The Human Code.

In this episode, we're thrilled to welcome Ari Kaplan. Databricks global head of evangelism and leading influencer in data and AI. Our he's impressive career spans decades of work across sports analytics, major league baseball, McLaren formula one, and his role in shaping modern AI technologies. Today, Ari. And I will share insights on how AI is evolving to capture human behavior and emotions, providing more nuanced and actionable insights. The impact of AI on business decision-making and how companies can stay competitive by effectively integrating AI into their operations. And to look into the future of AI, discussing rapid advancements and potential for revolutionary changes in the way we interact with technology. Join us as we dive into these fascinating topics with Ari Kaplan. This episode is packed with valuable insights that will change the way you think and how you approach AI and its role in the future of business in society. You won't want to miss it.

Don Finley:

welcome. I'm here with Ari Kaplan, and we're just going to discuss the intersection of humanity and technology today. Ari and I had a brief conversation beforehand, and I think that we're in for a real treat today. He's a wealth of information and knowledge, and Ari, I just want to thank you for being on, and Want to know what is it that got you interested in the intersection of humanity and technology?

Ari Kaplan:

Don, thank you so much for having me on. I loved, listening to some of the other podcasts, but yeah, ever since I was a kid, I've been fascinated like other kids with, how does the universe work, how do things work and, started realizing throughout history, part of human nature, whether it's cavemen scrawling on walls to represent the real life. So in something artificial like a cave painting to hieroglyphics to now the modern computer age, everyone's just fascinated. How can you encapsulate the real world into something, in this case, digital? And so databases, structured data, images, unstructured data has always been fascinating. That's pretty much been solved. But to me, the next fascinating part is like this whole human behavioral science aspect. maybe we can talk about some of my experiences, but sports analytics is a big one or, human behavior in retail or consumer purchasing or marketing is another where, yes, there are ways you can make a math formula on how many, somebody buys a Band Aid, but from the human aspect, if your kid scraped their knee and you need a Band Aid, you're going to. pay more money. And then similarly in athletics, people always say, yes, you can capture how quickly they threw a ball or how many times they dribbled before they shot. But what's hard to say is like how they adapt to real life. How are they team players interacting with others? So for me, I personally love exploring the part where it's the innovative edge. of the world, but it's not so far ahead that, you can't actually do something in real life with. So for me, that edge is like human behavior. That's, I think, where Gen AI is starting to figure out, actual insights that traditional, number based analytics couldn't.

Don Finley:

I love this because you've just described the history of computing like we look at it from the standpoint of like we had switches at first right then We had punch cards and then we have keyboard Mouse like that interaction that way that you go But now we're sitting at a point where and i'm skipping over mobile phones and all that but like just to make that point like you're now looking at that computer human interface, but also how The emotional aspect of what gen ai can create with the person alongside.

Ari Kaplan:

exactly. And to jump to the somewhat near future, I have right behind me, the Oculus Quest. I'm a huge virtual reality fan, and that's like extremely immersive, but that like any, I call it a gaming platform, maybe there's some business uses, but for now a gaming platform. you are limited to what other people have produced and it's very, time consuming to make a video game, rightly but I've started seeing demos where you could just say, I want to be in a field. I want non aggressive lions to come up and play with me. I want to be able to fly over this, mountain range and hopefully in the future we'll be creating real time using AI. These immersive experiences. So that, I think we're already even starting to see.

Don Finley:

I love this aspect of What we can create, what we could interact with, how we can, meet our needs in different ways and get that playful side of us into this with an experience that is much richer than like the standard environments because you run out of creative spaces when it comes to just doing something one off. What do you see holding us back from having those opportunities or what is it that is, really opening us up to having more of these, exciting and experiences.

Ari Kaplan:

Yeah. great question. And, there's two worlds, I think of this gen AI. space. One is the consumer aspect, creating these virtual environments. And then the other is the business aspect, companies being able to get better insights from all of their data. Their structured data like numbers and, text and date fields and unstructured like video images and, LLMs, text based, so what's holding us There's a few things. One is our imagination, like what are the use cases that you can bring into production with this? And then the other is just, one reason NVIDIA stock has been going tremendously is the compute power. And then the, the other is the ease of use to be able to. scale out and do massive compute and don't want to jump the gun, but I'm at Databricks. I'm the head of, global evangelism. And, about a year and a month ago, I'd say we acquired Mosaic AI and that combined with a platform. It's called the Lakehouse platform or data intelligence platform. Structured unstructured in one environment. Fine. It's called multimodal analytics. How can you use structured data like. How many sales did that Band Aid do? And unstructured data, I don't know, social media, Yelp descriptions, images of where in the store it's being placed. And if you have all different types of data, you can make many more informed decisions. So one thing holding back was there wasn't really a platform. a unified platform that did all of that. the other thing holding back is, sometimes companies don't have the data or the right people to ask the right questions or to know how to, do these data science AI questions. democratization is how do you democratize it? Make non technical people able to do it is, I think really one of the keys.

Don Finley:

I think you're hitting on a point because when we talk to partners and our clients about how we're doing AI implementations now, we talk about it as a digital transformation. And likening it to the idea of what happened in e commerce, not many people in your organization knew what e commerce was right. But you had an understanding that the world was going in that direction and then it was going to be an opportunity. And at first we saw that there was partners that were creating platforms that allowed you to do this and that it was one of those, the questions that you're asking are the activities that you were trying to accomplish help. But what we're seeing today. As far as generative AI is that there's not a lot of people that understand really how to get it to go. most people are doing zero shot prompting, and that has an accuracy of let's say between 60 and 70%. But in order for us really to do good things in business, we got to be above 95%, And that requires either adding agentic frameworks or something else to the solution. But really what we're learning is that the activity of having individuals utilize AI and having a structure in the business that allows them to experiment and play is really helping to ensure that like the entire company can level up and start to use these processes and these tools so that the data science team can focus on the really hard problems. Whereas like the tools that you can get out today are enabling the rest of the business to actually do their role and to create, actually to enhance that imagination of what's possible.

Ari Kaplan:

Yeah, a hundred percent. there's two parts. There's the data and then there's doing the, analytics or the gen AI on the data. So from the data, That's probably one of the biggest challenges is data is typically in different silos, different clouds, online, offline. If you have different partners, it's hard to share data, typically. And one cool thing with Databricks, we have clean rooms and, data marketplaces where you could share data either freely with the public or, abstract out, for example, the emails or private information. And you can only share the data that you want to share with other companies. And it's fully masked. in the past, you would share all the data and then hope the target would mask it. And that's a. potential risk. Then the other thing is, I think a lot of the structured data has been solved, in many companies, but unstructured, they used to refer to it as a data swamp, something like 90 percent of unstructured data is not being accessed. Companies spend all this money to collect it and put it in some repository. And people don't even know it exists, this unstructured data. So that's one huge challenge. how do you get the people in your company or your customers to be able to easily, Get access to the questions, but then how do you get technical and non technical people able to ask questions? So one awesome thing that's pretty much working now are large language models to ask questions of a company's own data. So for example what I do just co-authored a book and we have access to our sales force. We have this media campaign where we can track, for the ebook version you have to register. And so we have, thousands of names, in our registry and I. it's hard to write a SQL question. What is the revenue from customers who downloaded the ebook starting at the time they downloaded it? I don't know our database, maybe I could figure it out in a couple hours, but the. Kind of cute saying now is the number one programming language in the world is English or native language. So I literally can just ask, Hey, what were the revenue from companies after they downloaded the ebook? And it knows what I mean. It knows the context of the data where the word sale could be the word revenue. It could be the word, revenue. A R, and it just moves.

Don Finley:

Oh, see, these are the solutions that are just so awesome, cause if you're looking at a large company, you're gonna have data over here. You're gonna have data over there, but actually it's not even data. You're going to bring that into a central place. It's the intrinsic understanding of either what that unstructured data is, or that expertise lives in multiple spaces. And also one department may register it revenue. One may register it as ARR, but the system itself is able to understand that both those things. are the same. And that's such a powerful tool that you have at your hands today.

Ari Kaplan:

Yeah. And it's cool. And, in your point of like accuracy, that, that is another challenge, there's a, longstanding, quote that no model is perfect, but some are more useful than others, same thing, with these, generative or traditional AI insights is at some point. I worked with Major League Baseball. You could say, what is the expected performance of a player one or two years from now, and it'll give a result, but then there's a metric of how accurate that prediction is. And, It's pretty much not perfect, but is it 95 percent like you mentioned for one use case good enough to be production if it's landing a spacecraft on the moon, I want it to be as close as possible, but if it's a marketing campaign, and the click through rate. goes from 3 to 3. 1%. It's not the end of the world if that's, less accurate. It's still up to a human to determine the accuracy. taking that thought and applying it to, that question of, knowing, what is the revenue from this or knowing that the word revenue is the same as A R, or sales. It's called reinforcement learning. So it comes up with an answer and then you have the user click thumbs up or thumbs down, or to say, what was good about the outcome or not? write a little review. And then over time, at Databricks, we crowdsource our employees to help build our models. If you're an end customer, pick any company in the world as Their customers are using the LLMs. They give it their feedback and over time it learns and says, yeah, this word means this. Like in baseball, the word deceptive is actually good. It means you're a deceptive player. In the world of finance, maybe not so good. So you want the answer in the context of what your business is and, also just the questions being asked is information itself. So the more frequent somebody asks the question. or uses a certain phrase, it learns from that, gets better each time.

Don Finley:

there's a model that I've been working on as far as, how would you like AI to be in your organization? And I think that we're in a environment that has many, and especially smaller businesses, you have a number of Companies that you're working with your data may be populating in a number of different environments. And while all of them are adding AI into their product, there's nothing that has your brand voice. But what you've been describing is that data breaks can be at the center of your organization. It can be at the heart of really that the triangle that I'm calling is the, the, your partners, your customers, and your internal. organization as well as being that AI that really has an understanding of what's happening. We're able to get the information to the human who can make the decision. So it's augmenting that intelligence, that's there. Yeah. And I really, I love that. this is how we're situating and talking about Databricks because it is something that really sits in the middle and allows you to, understand the fullness of your business, instead of understanding it in those silos of the individual like tools or organizations that you have. where do you see like success? And then also what are some of the challenges of, these implementations that You've discussed.

Ari Kaplan:

Yeah. So, just with Databricks, just to educate the listeners, we're now the fastest growing, tech company over a billion dollars. which is awesome. And one of the reasons is it's like the infrastructure to enable everything. That you've been saying, just the data part. there's something called the Lakehouse technology. Databricks, CEO coined that term, but now a lot of companies do it. And actually, 74 percent of all enterprises have a Lakehouse, be it Databricks or one of our, competitors, other choices out there. And that is. Based on open source technology, like Apache Spark, Delta Lake MLflow that enables you to do structured data like a data warehouse and unstructured data like a data lake. in one platform. So that's now an industry standard. That's to get the data in. It's open source based, which has a lot of benefits, but it brings the costs super down. You don't have to have multiple copies of data moving, or multiple copies of data costing twice as much. So that's the Lakehouse. That's what the foundation is, and then, just co authored a book, the Data Intelligence Platform for Dummies by Wiley, but, which is available free, ebook download. But that is what you're saying. It brings the intelligence, understanding your data that's housed in the lake house. So you or your customers can ask questions, every aspect leverages AI. So it's getting information from your questions. Also don't overlook that there's a lot of software developers out there. People that write Python, R, SQL, and being able, there's lots of co pilots in the industry, but being able to generate or fix a code based on the understanding of your actual data and your code out there is super important. So if you use a GitHub co pilot or you use, chat GPT, that's has valuable information, but it's not your corporate information. And if I were a financial company, I would not send my code out to a third party, to say, fix my code up, just since people may see the data that is our bread and butter, our secret sauce. So having it, In a platform like Databricks, it's your own code. All secure is key. So yeah, there's asking questions of the data. There's the co pilot, code. But then there's also just the engineering. I used to work for Oracle as president of the Worldwide Oracle User Group one day and, You, have to have humans to size your database to understand what format should the file be in. It's like a big deal to get the best performance, but now AI is helping automate or your term, I loved you using augment intelligence to make the administrators of the infrastructure, like automate. so you never have too much and you're overspending or too little and you're bottlenecked on performance. So how do you scale up and down using AI? this is another feature that we have, and you're going to see all across the industry.

Don Finley:

that's fantastic. I love that concept because basically you're describing how do I have the right amount of processing, the right amount of compute, how do I have my infrastructure set up in a way that allows me to support everything yet at the same time? I know I'm overspending when it's two o'clock in the morning and I have a full environment stood up. is that what you're AI to understand where you're going to peak?

Ari Kaplan:

Exactly. Yeah. When are you going to peak? But then. with Gen AI, you probably hear in the media, somebody just asks a question like, what will my sales be the next four quarters? And then it's like a runaway query. And the company gets charged 10, 000 you know what I mean? Being able to have like safeguards and governance around it so you can limit it or come back and say, this will cost you 10, 000. Are you sure you want to know? but yeah, and short of that, yeah, you just, throttling things up and down, there's different concepts, materialized views and real time streaming and, travel and things like that, that have different costs associated with it. And in the past, you had to be very conscious of what you're doing, but now it's just like more simple to just go in, ask questions, not be overcharged, be charged only for what you're computing on.

Don Finley:

Oh, nice. you're reminding, and I'm going through a little PTSD of sometimes creating agent swarms of AI and then having them just get chatty with each other, and it ends up, costing you hundreds and thousands of dollars if you run it unchecked. And that is a really strong feature to be having and implement. cause sometimes you don't need, you don't need the thousand dollar answer. You need the one dollar answer. that's

Ari Kaplan:

Yeah, good enough answer. Sometimes. Yeah,

Don Finley:

we were on another call and I, it was like, done is better than perfect. So at least give you enough information. I got to say, what is, the data intelligence platform for dummies? what's the core basis of that? I'd love to get into it. You

Ari Kaplan:

I went over most of it in that it's getting the most intelligence from the data that you have from all of the data that you have. So companies, if you're a marketer, you have historical information, you have maybe web trends, you maybe have a background information. On your users. And, I'll admit it. I love Tik Tok. I love X Twitter, Meta slash Facebook. There are these ads that are targeted for me. And I buy them and I love them. I use the products. It's awesome. and in the past, there would just be like random, do you want a window washer for your house? No, I don't. I'm like, don't give me that. But now if it's something like, oh, there's a cheap USB. like, I always wanted a microscope only 50. I'll buy it. So the AI knows what I've watched and seen, and I'm actually happy with that. but, yeah, so data intelligence platform, helps the companies get more focused. Insights. It helps them better connect with customers as actual humans. And it's just finding that, but the important thing again, to stress is it's of your company's data. So the world today, when you do gen AI, let's just say open AI, chat GPT. that's a generalized model based on training on Taylor Swift songs based on. Russian Soviet

Don Finley:

a wide net. that you've cast as far as there's no, personification of your brand, of like your, of your company's experience at OpenAI, right?

Ari Kaplan:

Exactly. And that has some purpose, but it's more, appropriate to base your insights on your own company's data. And when I say that the company's data with governance, so it's based on data that Is allowed to be analyzed and allowed to be shared out. You don't want leakage of information that went through AI, but having it in the, language that your company, your industry uses is going to be way better context, way better received. If it's based on your information and, there's so many examples, different words, a can of corn and baseball, it's catching an easy fly ball, but, means different thing to a farmer. So over time. You want things based on your data. And then there are times you want things based on your proprietary information. If it's a medical information that can't be shared publicly and you're a doctor and you want to understand what are the prescriptions I should use, what drugs interact with each other, better than ever before, based on someone's personal information that we can't share, that's where a data intelligence platform also helps out. So you want it. based on the context of your own information and then all the other benefits of, code, writing software, helping maintain an environment, and then being Flexible open since like in the last couple of weeks, there's been a gazillion new LLMs coming out. There's been a gazillion new updates from every single vendor. So how can you make sure that best ideas, the best technology gets incorporated? So that's the data intelligence platform. and I believe there's going to be a lot of vendors out there having their own version, with Databricks, we've. been in the driver's seat for that and have a lot of customers and, sets us apart. Is that data intelligence? We're the first to do it at the scale and then our governance, something called Unity Catalog, which we actually open sourced a couple of weeks ago. but Unity Catalog is like the governance framework. How do you make sure that you, Don, have access to the data? to all of this, and I have access to this subset, and it controls all of it. It's, not just access, but auditing. Something called lineage. you could see from the raw data, all the way through the machine learning prediction, or the Gen AI outcome, like what, was the data based on. So me, if I make a Gen AI model on some marketing campaign, I want to be able to see the data, the sales data was up to date from yesterday, or was the sales data from 10 years ago. And unless you have that lineage, the underlying data, you may not quite know how your model has been built. So lineage is. It's been a godsend for a lot of companies just to figure out where all their

Don Finley:

And I'd imagine that lineage works on the unstructured data as well. if you have training manuals or you have a conference keynote that your CEO gave 10 years ago, it probably has less significance the report that you're doing today, but you'd still want to know. that's where the information was coming from. And that's, also a key challenge that we have with open AI or like just LLMs in general is you don't know where that data came from unless it's actually citing the source

Ari Kaplan:

Exactly.

Don Finley:

And yeah, like we still need to have That human in the loop or that human verifying that it's verifiable by humans, in some capacity. And then especially I think where, if we talk about using AI for an underwriting solution,

Ari Kaplan:

Yes.

Don Finley:

being able to know where that data is coming from, how those decisions are being made is really important to know that you are providing a fair product that is lacking in bias.

Ari Kaplan:

Exactly. And I loved your example of, the versions of the instructions are like underwriting, each different state or country has different rules. So you want to see, was this something from the 1950s or most updated? So when you do some traditional machine learning AI or generative AI, you want to be able to see where all the sources and then how is the data merged or sliced up along the way. so exactly. So you get confidence and then, 100 percent agree that you, still need humans to, make that final assessment. It's just augmenting. It's helping those humans, feeding them the information. since that research part is just typically, whether you're a lawyer, an underwriter, the, government, whatever it is, that research part is Like the most time consuming, bored and repetitive part.

Don Finley:

I can definitely share that sentiment. All right. I gotta say it's been absolutely fantastic having you on. What I would love to leave the audience with is really your vision for, and I know you're living at like the edge of innovation, but where do you see the world in five, 10 years?

Ari Kaplan:

That's a

Don Finley:

what excites you? Yeah.

Ari Kaplan:

Yeah, and not too long ago, just a few years ago, the question is where's the world in 20 years. And now like asking the question, where's it in five or 10 years is mind boggling, where's the world in even two years from now is, going to be pretty wild to see the advances, that we're seeing both on the AI side and then the capacity side, the amount of data being collected is incredible, the ability. just to process unstructured data shouldn't be understated. To take every YouTube video out there, and be able to process it. what percent of speaking on a Zoom call is by what person? How do you summarize things? these are all mind boggling and they, mind boggling to me. They didn't really exist. and like an easy way that it does now a year and a half ago. So it's incredibly mind boggling. I'm seeing, like from the business standpoint, you're going to see companies, like you do better, make better insights. Things are going to be better governed. the actions are going to be more, in step. So the, like the saying is, your job, isn't going to be replaced by AI, but it's going to be replaced by somebody using AI more effectively than you are. so it's going to really, improve a lot of business insights, democratizing the barriers to entry to be able to do this that used to take. a lot of programming knowledge. It's going to open it up to everyone. So I think people who have skills like knowledge of your industry and being able to communicate things and interpersonal skills are going to be way more important than ever before. So that's like the business side. And then on the consumer side, it's just from Elon Musk coming out potentially with these robots that are all purpose for not much money. that sounds wild, but if that works, that will change a lot. on the, like who knows side, like the job automation, a lot of jobs will be automated, and some of them that will be a good thing, like telephone switch operators, no longer needed, that's fine. The world, it's fine for that. since people were able to like up level, but it might happen so quickly that, it could be great. It could be horrible for, the job market, different. Parts of the world. that's like a macro level and, we'll see there's going to be inventions. There's going to be people with just one or two people or 10 people being able to start enormously successful companies, that will be incredible and I can't wait. My brain can only think a little linearly and maybe I can do one level abstracted but in five years, 10 years. we'll be able to be created in real time. We'll defy our imagination now.

Don Finley:

I absolutely love what you're saying. And I agree with you. thinking 20 years right now just seems too far out. we've been hit with a wave of exponential growth. in this field that just seems like very surprising. and even two years ago, I was telling people, I was like, look, we're probably going to automate blue collar jobs before we automate white collar jobs away, and then you come to see where LLMs have advanced to, and you're like, you know what? No, white collar, everybody's job is changing. And you're right. there's the possibility that we have for mass change seems to be right on the edge. So Once again, man, I really thank you for taking the time today to talk to us.

Ari Kaplan:

it's been so exciting. Thank you. Appreciate it.

Don Finley:

Thank you for tuning into The Human Code, sponsored by FINdustries, where we harness AI to elevate your business. By improving operational efficiency and accelerating growth, we turn opportunities into reality. Let FINdustries be your guide to AI mastery, making success inevitable. Explore how at FINdustries. co.

People on this episode