EP 268: AI’s Data-Driven Decision Paradox

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life. We are in the age of data. We have so much data, almost too much. But what are we doing with it, right? Especially now that we have generative AI and large language models that, that can help us do infinitely more than we could in years past without it. So how do we still have all of this data but so few people are using it in their decision making? It's actually something I think about a lot, but today maybe I'll get some answers and hopefully you will too. So what's going on, y'?

All?

My name is Jordan Wilson. I'm the host of Everyday AI. We are your guide to learning and leveraging generative AI with our daily podcast newsletter and our live stream. So thank you for joining us. If you're listening on the podcast, make sure, as always, check out your show notes. There's going to be a lot of links to check out with more helpful information as well as a recap of today's show in our daily newsletter. And if you are joining us on the live stream, like Juan coming in from Chicago, like myself, thank you. Get your questions in. So, before we get into that topic, and I'm extremely excited to talk about this kind of not a new intersection, but a very exciting intersection of so much data and what we can do with it with generative AI. But before we do, we're going to start as we do every single day with the AI news. All right, got a lot going on today. So OpenAI has kind of told us how chat GPT works. So OpenAI has released a proposed framework called Model Spec to shape how AI tools respond in the future. The framework includes principles and rules that aim to assist developers and end users like us what they say to benefit humanity and to reflect well on OpenAI. The company hopes to gather public input and incorporate that feedback to ensure responsible development of their AI models. So AI tools behaving badly has become a common issue, as we all know, highlighting the need for responsible development. OpenAI's model spec framework aims to guide the behavior of AI tools with principles such as assisting developers and end users and complying with applicable laws. So we'll obviously have a link to that in today's newsletter if you want to check it out. It's actually a pretty light read, so don't, don't think it's, you know, crazy, crazy advance or anything. All right, speaking of our newsletter, we snuck this in in our newsletter. Didn't get to it on the podcast yesterday because it hadn't come out, but worth talking about. Google DeepMind has unveiled the latest version of its artificial intelligence, Alpha Fold, which can predict how proteins behave and interact with other molecules. This is huge. So the breakthrough has the potential to advance research in fields such as medicine, agriculture, bio, everything. So the new version, alpha fold AlphaFold3, can predict how proteins interact with each other with other molecules and has an accuracy of reportedly between 62% to 76%. Scientists can use AlphaFold 3 to design molecules and antibodies for medical treatments, potentially saving time and accelerating research. Also, this new AI may contribute to understanding complex biological systems such as photosynthesis and plants as an example. But if nothing else, here we are now where generative AI is definitely already discovering new drugs. But with Alpha Fold 3, yeah, it's. I think it's going to expedite that process a lot. Last but not least, kind of a small one, but I think very relevant for our audience. Zapier has released Zapier Central in AI workspace. So Zapier, known for obviously business automation services, has launched Zapier Central, an AI workspace that allows customers to create AI bots using natural language. So Zapier Central allows for the creation of AI bots without coding, making it easier for customers to integrate automated services. So pretty cool. They just released kind of a demo video a couple hours ago. It does have this kind of chat GPT like interface where you can chat with Zapier and it does just kind of create little agents that can do your work. Isn't that what we all want? Yeah, that's what I want. All right, so there's, as always, there's going to be more on those stories and more AI news and just what's happening in the world of AI in our newsletter. So make sure to go to your everydayai dot com. All right, but today we are talking about data, right? So there's literally so much data, too much data, I can't make sense of it all. So a lot of us use AI to help make sense of it. And, you know, there's pros and cons to that. So I'm not going to talk about this by myself. Very excited to welcome on to today's show. Let's go ahead. There we go. We have him there. All right, so we have Aakash Indurka, who is the head of product at Virtualistics. Akash, thank you so much for joining us.

Yeah, thanks, Jordan.

Great to be Here I think I got like a C plus on the pronunciation of your name there. But anyways Akash, tell us a little bit about what you do in your role at Virtualytics.

Yeah, yeah, so I'm the head of product so really helping to navigate kind of the future direction of the product, making sure that we're constantly creating new value for our customers. Also you know, a lot of the product leadership kind of figuring out how we're going to, you know, embed generative AI and other other new AI technology into our platform. And you know, as a company we really focus on using AI to change the way that people approach analytics and get past data driven decisions to impact.

Give us the super high overview. What's the average client or customer? What are they using VirtualYtics for?

Yeah, so I'll give a quick example. We've done a lot of work with the maintenance operations world. So talking about maintenance of physical assets, you know, aircraft facilities, things like that. And so really common problem for maintenance, maintenance and repair operations is they might have predictive maintenance that tells them when something's going to break. Right. But that's not really a decision, right. The decision is okay, what do you do about it? Right. And to be able to answer that question you need to be able to know, you know, when it's going to break. But also do you have the parts, the people, any special equipment to do that maintenance job. And so that's like an example of a problem that we solve. Right. So we do a lot of work with maintenance operators but also their analysts and our platform in general kind of extends to developers and even executives. Right. So really industry specific decision intelligence if you will.

So Akash, let's start just at the end. Let's, let's fast forward, let's get to the, to the point here. So you know, there is this time now I think we're in this sweet spot here in the, you know, age of the Internet where we have so much data and it's so easy. Right. And we have generative AI which is actually pretty good at, you know, making sense of all that data. So you know, can you talk a little bit about what this paradox even is? You know, it's so much data. People want data, but are they using it?

Right? Yeah, no, I think it's something that ultimately is really relatable. I'll give an example at the end. But you know, over the last decade or really 15, 20 years or so now, there's been a huge proliferation in data collection and also data analytics. Right? That's both. Things have become easily accessible to every business out there. The problem is that even though 80% of people out there want to make data driven decisions, 70% of people are also saying that analytics and the data itself is the reason they can't make a decision. Right. So that's kind of what I call the data driven decision paradox. And that decision paralysis is something that we honestly experience in our own life. Right. You know, we were chatting before and I kind of mentioned like, you know, if you had to go buy a toothbrush, you know, 15, 20 years ago, you would just go to the store and buy a toothbrush. Right. Nowadays, if you need to do something that simple, a lot of us will end up going on Amazon, looking at reviews, checking if some of the reviews are fake, really getting into it, but also kind of getting in our own head. So you can kind of imagine that if buying a toothbrush has become that complicated and you can get into this data driven decision paradox, just imagine making an actually complicated business decision. Right. With a lot on the line, it just really ups the ante. And I think with all of that, a lot of people have become kind of overwhelmed or just really unhappy working with data at this point.

You know, I'm laughing because I feel personally called out by that. Yeah. I've literally spent hours, you know, literally scraping data from review sites, running semantic analysis on a purchase that's like $50. Right. So why do you think it is? Right, it's, it's, it's. No, you know, data has been the new gold for, you know, a while, right. Or whatever you want to say, the, the new oil. So why do people have all this data and why can they not actually do anything with it? Is it too complicated? Is there not the right systems? I mean, what are some of those actual reasons? You know, I even think myself, right, Like, I'm a big data person. I probably use the data once or twice a week, but I have to be very intentional about it and go in and maybe it's a manual process, even still using generative AI. So why this big paradox? Why the big drop off between that, you know, 80% of people wanting to use it in only 70 or 70% of them not doing it.

Yeah, yeah. I mean, couple answers there. So, one, why do we have all this data? I mean, just, we started collecting it and we never stopped and I don't know that we should. Right. I think it, you know, we actually can put it to good use. That being said, 97% of data that's collected out there just goes completely untouched, right? So, like, even though we have it, we're definitely not getting our full bang for the buck. As for why I think people run into this whole paralysis situation, it's really interesting, right? So I think there's a few key reasons. One is that nowadays when we make. We want to use data to make decisions, the problem is it's not just like one data set or one column, and it's like, yeah, just pick the thing furthest to the top, right? And that's it, right? It's the best. It's not that simple anymore. Data is really complex, and honestly, you have to fold in different layers of the data. And one of the biggest problem is that the layers of data might start to disagree with each other, right? You know, one, you gotta be able to figure out how to join that data together. That's. That's part of the battle, right? But ultimately that's, that's pretty doable. You just have to get through it. The problem is, even after you do that, what if data set A tells you that your decision is right, but data set B says there couldn't be anything worse than the decision you're about to make? What do you do then? Right? Classic decision paralysis. I think the other thing, you know, beyond the data quality and kind of the conflicts between different layers of data, there's also a whole other layer to it, kind of like the human layer of it, right? Which is first you got to get to the point where you feel comfortable making the decision, right? You have the transparency around it. You can explain it to other people, you could show them the data, but then you have to socialize it, right? It's very rare that you have a single. You know, we're not all the president, right? We can't just say, this is how it's going to be and then it just happens, right? We have to kind of socialize our decisions, get other people bought in on it. And while that's necessary, it also sometimes invites what I call decision congestion, where it's like, that's where the bureaucracy comes into play. It slows things down, right? Too many cooks in the kitchen. So there's a few different layers to how it happens. And I think it's created like a big gap where there's all these expectations to be able to use data to do amazing things. In some cases, cases we're doing that, but I think in the majority of cases, we're nowhere near what we should be able to do with it. And, and I think, like Some of the things I highlighted or play a big reason in why.

Explain, explain that decision congestion. Let's, let's go into that a little bit more because I can kind of imagine how that plays out, because I personally feel that way a lot. I'm sure a lot of people do. But explain specifically what that decision congestion is and how, you know, whether they're business owners, marketers, data scientists, how they can work through that.

Oh, absolutely. Yeah. I think it happens for a few different reasons, but basically at its core it's like, I think it's often called bureaucracy in other situations. But you know, you may see that there's a clear idea for what needs to happen and you can explain it. You have transparency, you've got the data on your side, but when you present it to other people, because that data, you know, even the whole concept of making a data driven decision is really new for people. They're more based on relying on their hunches, what they're used to doing. Hey, it's the status quo. Why break it if it's not, you know, why? Why fix it if it's not broken? Getting past that mentality and often it's because the people that are kind of pushing back on it, they don't have visibility into the data. Right. And I think one of the things that we've really been exploring a lot as a company is, you know, you've got kind of these different Personas, you've got analysts, you've got developers. Right. That think about things really differently. And then you've got executives, which is like polar opposite of a developer in a lot of ways. Right. And so you actually need an experience that you need a common language across all of those. Right. Otherwise you're just going to talk straight past other people as a data scientist for, you know, 10 years. Right. I think what can often happen is you just talk straight past an executive who. Or they talk straight past you and it's because you're not talking in each other's language. So having that common language and having an experience that fits the mold for each of those Personas, but then is easily translated to the world of the other person you're talking to. So they actually get it in their language and in their world. I think that's a big. That hasn't really existed in the market. And so that's one of the ways we're trying to approach solving this problem.

You know, one thing when I specifically think about data and decision making is I think people underestimate how powerful a Large language model can be. Right. I think that there were all these narratives maybe very early on, right. Like when companies were still, you know, blanket banning generative AI and large language models. And, and when people would, you know, share screenshots and say, oh, you know, ChatGPT doesn't know what, you know, four plus four is yet today, if you know how to use generative AI and large language models correctly, they can save you even for myself, 90% of time, right. Like I can have a spreadsheet with a hundred thousand data points. I used to spend hours on that. Now I can get that decision in seconds. So how can the average person maybe who's not like yourself, you know, with a decade or more of data science experience, how can they still find that sweet spot, right. Of using a large language model, using it responsibly and not going into that, you know, decision congestion or, you know, being able to actually use data and in large language models.

Yeah, I mean, I think there's, you know, on an individual basis, right. For just someone that wants to be able to do more data analytics, more data science and leverage generative AI. I think one thing that's key across all generative AI is the conversational interface has been really powerful because there's kind of check ins, very intermittent check ins that are very regular check ins that we're not getting too far off base, right. And the user is still the pilot and the generative AI is kind of there as a co pilot to help produce options for the next step. And then the human can kind of say, okay, I want to go more this way. And I think generative AI is also kind of going through another wave with RAG technology that allows it to be a little bit more clear where it's getting its information from. So I think all of those need to kind of also translate into the data analytics, you know, AI for analytics world, right. Where it can't just be full automation from, you know, one end of Here's a question all the way to okay, we solved your problem, right? There needs to be check ins. So I would say as that technology kind of comes to market, it's going to help a lot of people and allow them to do it in a responsible way. Which I think especially when you're making important business decisions, it's like so critical, right? So yeah, yeah.

And those critical business decisions, I think, you know, you mentioned rag there, which I think is extremely important because then I think that could in theory improve, you know, the outputs. Right. So if for those that don't, Know, retrieval, augmented generation is rag. So Aakash, can you talk a little bit more about how RAG may or may not? I mean, I think it will help. Right, but how can rag, when we start bringing in our own data, how can that help maybe close that original gap that we talked about of all these people wanting to use data but not using it. How can RAG help in that?

Yeah, I mean, I think it's really the concept of rag. Right. Being able to pin a path forward to something you've already seen. Right. Tying it down to some sort of ground truth, if you will. Right. That whole concept is super critical in data science because you have to have like breadcrumbs along your path. Right. We always refer to that. The path from data to impact has to be, you know, with breadcrumbs for each mini decision that went along the way. And so RAG is just kind of that concept of you can always kind of refer back to some source data and kind of cite it. Right. I think when you're doing analytics or making a decision, you need, you know, one. Yeah, RAG for generative AI will help, but you need that whole concept of being able to trace back your decision through the analytics, through the visualization, back to the source data. Right. Or even any of the transformations or AI you apply along the way. So I think that concept of being able to cite it, it has that kind of scientific method about it that really gives people more confidence as they're going through that analytic journey. Yeah.

And I think even because I know that there's some people who haven't even touched their toe with rag, but the way I like to describe it is essentially think of it as a layer of your own company's data that is between you and a model to ensure, not ins, but you know, improve the likelihood that your data is being used in a responsible way and you can trust the outputs. So one thing actually have two very similar questions here from, from Brian and Cecilia. So I'll go with Cecilia, since there's actually two questions here, so asking what are the best tools for data driven communications and asset management. Kind of specific, but we'll see if we can tackle that one. And then the second one, or maybe we just do, you know, the best tools in general. That's also kind of what Brian was asking here for data analysis. But then also what are, are there AI based text or communications driven by data that provide alerts?

Yeah. Okay, so this is, this is tricky for me because I'm going to have to plug my own company. Yeah, I think the platform we're working on, virtualytics AI platform and the decision intelligence apps that live inside of it are really built on to address like everything, you know, that that was mentioned in the questions. Being able to improve your communications, really having that common language across your different Personas and then also being actually able to get into the weeds of the analytics. Right. And, and apply, you know, maybe not generative AI methods every time, but even just predictive models, you know, tradition, traditional quote unquote AI. Right. To, to better your, your analytics. So you know, we have AI kind of at a few different layers, but I think really the focus is still on helping the humans through, through their analytics.

Yeah, no, and, and, and you know, I'm curious. So yeah, we'll, we'll definitely have a link to VirtualYtics in the newsletter for those of you that are interested in learning a bit more. Maybe Akash for you know, your, your standard tools. Right. Large language models. Because I'm sure a lot of people, you know, especially our, our listeners use those and they're trying to, you know, figure out. Okay, well number one, we should always say don't, don't upload confidential sensitive information. Right?

That's a big one.

Get that blanket disclaimer out of the way. But you know, do you have any best, you know, best advice for people who do want to bring, you know, kind of public or non confidential data into large language models? Is that a good idea? And how can people best use those large language models?

Yeah, so I mean one, again, like you said, you definitely don't want to bring in confidential data. Right? That's, that's almost a whole separate conversation where I find that it can be very helpful in. And this is something that we're also doing is I do think you can ask, you know, GPT and these other large language models about methods. Right. You can kind of inform it. Here's what I'd like to be able to do. Right. But I'm not a data scientist. How do I get to that point? Right. And it's actually pretty helpful at being able to suggest ideas for how to get there. You might still have to do some of the legwork. Right. Or as you know, the technology is maturing in the market. The LLMs themselves will be baked into all of these analytics and AI platforms.

Right.

Which is, that's really the future that I'm envisioning.

Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on genai. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies in the world do. Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen. What? Speaking. Speaking of the future, right? How, you know, how can, you know, I'm sure this is, you know, a big part of what you're working on at Virtualtics, but how can, you know, business owners, decision makers, how can they start to close that gap? Right? And the reason I am, you know, talking over and over and trying to hammer this point home about this like, sweet spot of data is because, you know, big companies are making it harder to collect data. You know, we've seen iOS updates from Apple making it harder. You know, Google's been talking about getting rid of cookies now for four years, right? And they say, oh, this year's the year. But you know, we have this sweet spot of having so much data and we have this, you know, extremely powerful technology in generative AI in large language models. How can we actually take advantage of this time and not have that huge gap?

Yeah, I would say, honestly just getting started, right? Because it's all about turning those unknown unknowns into at least known unknowns so you can start improving it. I think one of the biggest problems in general is people just don't know what's in their data. And you're not going to know what's in it until you just start getting hands on with it. Right? So you need the right tools, but you also need people that are going to feel comfortable to look into it. And I think there's always this pressure of, well, there's data, so you better do something good with it. Having a little bit more openness about, okay, maybe all of this data isn't that good. And even though you might have a terabyte of it, it Might be better off to learn what's wrong with it and go collect another terabyte, Right. Because you can do that faster now than you could have before. But if you don't start that journey and start discovering how broken it may be, you're never going to get past that first step, right. So. And there's all sorts of tools out there that can help with that.

Give me, give us an example of, you know, a couple. Because I'm sure what you said there probably just resonated with a lot of people, right. Because we've all spent, you know, hours staring at an Excel sheet or a Google sheet.

Oh, yeah.

Find out like, oh, this is garbage.

So I, I've seen this before. I'll kind of go back to that, that maintenance example, right. For maintenance operations, you know, suppose that you've got maintenance data about, you know, different assets. Let's talk about H vac systems, right? In large buildings, maybe you've got a bunch of different H vac systems across a series of buildings. And some of that you have the data on whether they're breaking. You've got sensor information, and then separate from that, you've got your technician data, right? Who's going to different sites and doing maintenance and completing which jobs. Imagine having both of those data sets and wanting to put them to good use, and then you discover you've got no ability to track which asset they were going to each time. So you know that there was maintenance going on, but you don't know on which assets or you can't tie it to the sensor data. That's a great example of where it's just, you're missing a link. And if you don't start going down that road to figure out how to link your data, you're not going to discover it and you're not going to fix it. Right. I've literally, like walked around campuses with people and just said, look, the thing you need to do is just take out a notebook, write down the asset ID here, and then go back to your desk and figure out which asset ID that that lined up to. Right. So sometimes making the fixes, it's like you have to get a little bit literal and like, put your finger on it a bit. But it is solvable. You just have to get started to discover it.

And, you know, it's interesting, Aakash, and I love that you gave that example. Very illustrative of literally, you know, you could be a data scientist and maybe, you know, we always talk about, okay, what do the future of jobs look like, right? Like if a large language model can pretty accurately crunch, you know, millions of data points in a couple of minutes that it used to take days, what does the future of jobs look like for those people? And, you know, maybe that example of what you just said is a good one, is having that human piece and going out there literally and looking with your eyes and writing things down in the notebook and starting to connect data points and tell stories. So is that the future? Right, Because a lot of people think like, okay, you know, with coding and data analysis and, you know, I don't know, bookkeeping and all of these more manual jobs specifically around data. People say, okay, well, these are going to be large language models. How do we think that, you know, the, the role of humans is going to change in helping us maybe close this, this data paradox?

Yeah, I mean, one, I think, you know, analytics in general is about to get much more industry specific, or it's going to need to be, because the value proposition for analytics in general only really makes sense when you're talking about within the confines or the context of a specific industry. I think how humans can help, that is focus on the real world constraints that the data is not going to know about. Right. Discover the issues in your data, but then also really try to understand the information that's not actually going to show up. I sometimes call it the physics of the data. There's the data, which is just what you collect, and then there's the physics behind it of what constraints were there that aren't documented anywhere. But there's a reason why the data looks the way it does. Right? So, you know, become experts in your industry. That's the part that humans need to, to provide.

How can, you know, you make it sound so easy, right? But you, you're obviously an expert in this field. How can people better do that? How can they better understand? Right. Because I do think, you know, people who are going to excel in this new age of generative AI, they have to think about their roles differently. I think you just gave a fantastic example. How can people do that? How can they, you know, kind of take, take a step out maybe from, you know, metaphorically or literally. Or literally. How can they step away from the data and start to better understand the physics of it? How can, how can people do that?

Yeah, it's a, it's a good question. I don't know that there's like a single bulletproof answer. I would say, especially as the data and the generative AI aspect sort itself out more, just start communicating More with subject matter experts. Whatever industry you're in, someone at your company is going to know in depth how your manufacturing process works, how does distribution work go and have those conversations increase the communication about those things. Because the guy that, that's, you know, the guy or gal running your, your data science, if they don't know that, they're never going to come to you with good solutions to those, those questions or those business problems. So I mean maybe it's a cop out answer but you know, just getting out there and communicating and you know, just taking an interest in it, you.

Know, sounds crazy but something that might be just the advice some people need to hear is like get out in communicate. Because I think in, you know, in today's world sometimes you know, that, that, that gets lost. It's a lost art form. I think we have one more question here before we wrap up today's interview. So Douglas asking, do you have recommendations for using AI to visualize findings? Asking is it just Python based reporting or are there other approaches?

Yeah, so this question is very much at the core of virtualytics technology. So I gotta again plug it. You know, that's, that's where we really got started is using AI to guide what to even look at in your data to surface where those interesting intersections. You know, you can use Python and other, you know, programming languages are to do some of this but it's kind of a different way of thinking about AI. Right. Typically people think of AI as more like predictive. Right. And they really go for it from that. This is kind of flipping it and saying AI will help you discover what to look at rather than looking at stuff so you can make a predictive model. So I think programming languages are good if you actually know how to write code. And that's the barrier that most people get stuck in. Right.

So we've talked about a lot here. So we've gone over how the role of data analysts might change, about how not all data is, is good data. We've talked about rag and how large language models can actually help you just look at data differently. But you know, as we, as we wrap here, Aakash, what is maybe your best takeaway or your best piece of advice for maybe people either who are working in data or maybe leaders who are having to make decisions on it? How can they start to close kind of that gap that we talked about earlier between wanting to do something with the data and, and actually doing something with it?

Right. I think one just be really honest with yourself about data quality and issues that are coming up when you have to start integrating with data. Two would be, you know, if you want to get past this data driven decision paradox, you're going to need tools that assist you with that. So invest in your tools. And I don't think kind of the traditional BI approach is really getting it done. So you kind of need a different set of tools. Right. BI is great for reporting seeing what maybe isn't working or is working, but if you actually need to get into the art of solving a problem, you need different tools. And then third is really start to practice decision socialization. But it has to fit in that sweet spot between being able to communicate about it and congestion. Right. So those are the three things. Right. If you, if you want to get past this whole paradox, you have to invest in all three of them.

Geez, my, my fingers literally hurt because I'm typing so many notes and so many takeaways because I think you just helped. I think a lot of people out there better tackle this, this huge problem of having so much data and having great AI but not being able to do anything with it. So thank you so much, Akash, for joining the Everyday AI Show. We very much appreciate your time and your insights.

Yep, thanks, Jordan.

All right, and as a reminder, everyone, like I said, so much great information. This is one. Whether you're here listening on the live stream or on the podcast, make sure to check out your show notes or just go to your everydayai.com we recap every single day's conversation by me, a human. I go and write this and I'm excited to give you more insight. So make sure you check that out and make sure you join us tomorrow and every day for more Everyday AI. Thanks, y'.

All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic. Visit your everyday AI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.