Utilizing Tech - Season 7: AI Data Infrastructure Presented by Solidigm - 3x11: Putting Data Science Into Everyone's Hands with Amanda Kelly of Streamlit

Episode Date: November 16, 2021

Data science and machine learning developments can't have an impact if they don't get into everyone's hands. In this episode, Amanda Kelly of Streamlit joins Chris Grundemann and Stephen Foskett to talk about the challenges and opportunities in bringing data science to everyone's hands. How can we enable marketing, sales, marketing, and other elements of the business to access data and make informed decisions themselves? Data science teams have to meet business people where they are to better answer their questions rather than trying to create a perfect model in a vacuum. Streamlit helps to productize python scripts with a complete and flexible front-end and easy deployment, making it easy to share and iterate. These micro apps foster collaboration and interaction between data science and the business. Three Questions Stephen: Is AI just a new aspect of data science or is it truly a unique field? Chris: Can you think of an application for ML that has not yet been rolled out but will make a major impact in the future? Leon Adato: What responsibility do you think IT folks have to insure the things that we build are ethical? Gests and Hosts Amanda Kelly, Co-Founder of Streamlit.  Follow her thoughts on the Streamlit Blog.  Chris Grundemann, Gigaom Analyst and Managing Director at Grundemann Technology Solutions. Connect with Chris on ChrisGrundemann.com on Twitter at @ChrisGrundemann. Stephen Foskett, Publisher of Gestalt IT and Organizer of Tech Field Day. Find Stephen’s writing at GestaltIT.com and on Twitter at @SFoskett.       Date: 11/16/2021 Tags: @streamlit, @SFoskett, @ChrisGrundemann

I'm Stephen Foskett. I'm Chris Grundemann. And this is the Utilizing AI podcast. Welcome to another episode of Utilizing AI, the podcast about enterprise applications for machine learning, deep learning, and other artificial intelligence topics. We've recently and repeatedly discussed the challenges of bringing AI not to the ivory tower or even to the data center, but to the business. I remember we've had conversations about how AI is used to make decisions in the boardroom. We've talked about how AI can improve people's lives in third world countries. It's all about getting it out there. Isn't that right, Chris? Yeah, it really is. I think that one of the things we've talked about

is connecting domain experts with data scientists, but this is also about using those data scientists to really be able to answer quick questions for the business in real time or close to it, instead of sending folks off on a three or six month mission to maybe come back with the right answers or maybe not. Yeah. And it seems like that's one of the biggest challenges here because so much of the work that's happening in artificial intelligence and machine learning is very academic. And so it's a case of trying to figure out, okay, when can we be done enough to have a useful product that can then get out into people's hands? And then there's also a mechanical aspect of this, which is sort of how do we get the model into people's hands? And so that's why we decided to bring on today

Amanda Kelly from Streamlit. Amanda, it's nice to have you here. Hi, thank you guys so much for having me. Yeah, so as you said, I'm Amanda Kelly. I'm one of the co-founders and our COO over here at Streamlit. Long-time listener, first-time caller. I've worked in kind of AI machine learning for the past 10 years or so across kind of Google X, Google machine learning, Zoox on self-driving cars, and now at Streamlit. So really happy to be here and to talk to you guys about this. So why don't you kick off by talking to us a little bit about sort of the background and how the background led you to where you are today. In other words, you've worked at a lot of, pardon me, but a lot of egghead kind of environments.

And yet, you know, you're here saying, no, no, no, we really got to get this in people's hands. Yeah. You know, so, you know, I'm a co founder of a tech company and I've worked in tech for a long time. But, you know, I often say I'm the least technical person usually at the table, even though I might be more technical than, you know, 99 percent of the world. Right. I'm not I'm not a practitioner in the sense that I'm not in there. Right. Training the neural nets. Right. Writing the code. I'm fundamentally in many ways

the business user. Right. And so, you know, across my career, you know, a lot of what I've been doing is trying to translate the customer needs, whether that's coming from marketing, right, or sales or an operational group that's, you know, helping and using this kind of data and going to our engineers, our data teams and saying, you know, we need to have this thing, or we need this thing answered. My teams need a tool to do X, Y, and Z. And just too many times it's kind of like, well, not priority for Q4, right? Or maybe we can put somebody on that in Q1. And I'm like, you don't understand. Like we cannot move forward on this and we cannot make this decision or we're

going to be doing it in a much poorer way if we don't have this. But we have these kind of, you know, just really long, you know, cycles and we have these teams sitting off completely to the side. And it was, you know, really my feeling that we could be running a lot better, right. If we just had much closer collaboration between our business groups and our data teams. And, you know, that was one of the reasons that I came on to co-found Streamlit. And it's a lot of the work, right. That I'm doing today is, you know, trying to figure out how we can bring, you know, data science to more aspects of the business. Yeah, it's interesting. And it resonates with what we've heard from, from other folks we've talked to on the practitioner side. I wonder, I mean, and really what it is, is, you know,

I think a lot of folks have built up this big, to use Steven's word, ivory tower of data scientists, that's maybe a center of excellence for the entire company. And you've got to kind of book time on it. Like we're back to mainframes and I've got my terminal and I'm pulling in, but it's just a human power that I'm pulling in this time to get to the data. Is that the biggest frustration with, with, with data science in the enterprise today is, is, is getting access to folks who, who can analyze that data and really make it useful for the business? Or are there other challenges you've seen as you kind of come along this path? I mean, you know, I think if you, if you surveyed a bunch of people, we'd probably all, all of us would probably say in all of our functions that we need more resources, right. And we need more people, right. You know, things like that. But I think there's also just too often we're talking past each other, right. In terms of like, you know, what the data teams are trying to deliver, right. They're doing an excellent job, right. Creating these pipelines, creating these models and doing things. But, you know, while you're doing that,

while they're building that so many other things are going along on the side of the business, right? And businesses are often changing faster, right? Then, you know, our data and our kind of engineering pipelines. If you just think about like, you know, think about retail. And if I had asked a question of our supply chain three months ago versus now, right? Completely different, right? In terms of, you know, what the teams are worried about and what they need to do. And so a lot of it is just, you know, how do we enable those business teams, those supply chain managers, those salespeople, those marketing people, right? To really kind of keep updating, right? What they need, you know, what are the main needs of the businesses? And I think that one way we do that, right, is by not kind of forcing us to just have meetings with the

data team once a quarter where we try to write down everything and anticipate everything we might want to know, but creating more rapid processes, creating more tools that go directly to the hands of your business users, right? So they can just make those decisions themselves, or at least talk directly back to the data team, right, as things change. So that ties in with the challenge I've seen, which is a lot of data teams trying to anticipate needs or even the business forcing them to anticipate needs. You've got all these dashboards and reports and charts that are kind of pre-built and pre-canned around these different metrics that people want to track. And then you get three, six months into it and you realize that's not actually the data that you need, or it's telling you the wrong story because you're looking at it the wrong way.

So I think what I'm hearing you say is that we really need more kind of real-time access to that data, maybe via data scientists or maybe not, where I can actually ask a question right now that I need answered and not have to necessarily build a huge application around it, just get the answers in a way that I can present to myself, to my customers, my partners, et cetera. Yeah, I think that's absolutely right. Having these big dashboards kind of presupposes you know the question that people want to answer, and sometimes you do, and that's great. But too many times, I think that's absolutely right. You know, having these big dashboards kind of presupposes, you know, right, the question that people want to answer. And sometimes you do. And that's great.

But, you know, too many times, you know, I've been in meetings as someone who's managed data teams or even on the other side, and you have this giant PowerPoint presentation. And the first question that comes out of the exec's mouth is, of course, something that's not in the PowerPoint presentation, right? And then you're like, OK, well, we'll go back, right? And we'll do that. And it would just be so much better, right, if that conversation or even if it wasn't in a meeting, right, if you were just sending them something where they could turn the knob themselves and be like, well, let me look at this data just for last week. Let me look at this data, you know, just for, you know, our North American region, you know, let me, you know, look and see if I, you know, adjusted the timeframes, right, bringing their own assumptions, right, toying with the data, you know, building it themselves and then having, you know, a conversation with the data team if they need more.

Yeah, and I think that that's really important because too often data teams can seem sort of like, I don't know, like the cabal of wizards, like beyond the court, you know what I mean? And it's like, you know, oh, you know, King Arthur, what should we, should we go to war or should we not? And it's like, well, let me consult with Merlin, the data scientist.

He's like, well, let me take this back to my people. And then he kind of goes and does his thing. And then he comes back and says, yes. And that's it, right? Instead of saying like- The answer is yes. Yeah. Instead of saying like, okay, well, let's play with this a little bit.

Let's look at this data. And I think that that's a much more useful approach to be able to look at it and experiment. Yeah, I absolutely agree. And if you read some of these headlines or talk to data scientists about why their teams are a little bit frustrated, I think it's because we are expecting too much of them. The models are never going to be perfect. The data is never going to be perfect. You are never going to have framed that question in such a perfect way that it's going to come back and be like 72.7. That's it. That's the

exact price we should charge them, you know, go forth and make us all money. These things just evolve rapidly, right? I remember when I very first, you know, became a manager, I was so frustrated, right? When I would like, I thought I had perfectly articulated exactly what I would need somebody to do. And then they would come back and it would be nothing like what I had thought I had told them to do. Right. And ultimately that was a feeling of me as a manager. I wasn't able to kind of really articulate, but it was also a failure process that you just kind of give something and somebody goes off and they do a whole bunch of work and then they come back. Right. And so it's not the failure of the data science team. And I also don't think it's the

failure of the business. I think it's a little bit of a failure of the process, right? We need to just be, be talking more. We need to be enabling people more right to see the data for themselves, to inspect the models themselves, to have more informed opinions, as opposed to we're both in our side of the business, right? And we expect that the few emails and meetings that we have, we're going to perfectly articulate, right? Exactly what we want and need. So as we're talking about this, kind of getting that access to the data in real time and doing it kind of more off the cuff, giving these answers and not trying to get 72.6, but instead getting the, you know, yes, these two things are very different and this one's much

bigger, or, you know, this is a trend line that maybe you didn't see through the noise. Maybe you can tell us a little bit about how Streamlit specifically is approaching that problem. And, you know, because obviously, this is a company you're building to attack this area. Yeah, absolutely. So the big insight that we had at Streamlit was that we needed to empower the data teams, the data scientists, to directly build tools that they could give, right, to the other side of the world. And the problem that we see in a lot of companies is there's these very, very long pipelines. I think it takes, you know, often like nine to 12 months to implement kind of like a new AI model and get something out there. And they're very kind of waterfall things, you know, to build out, you know, all of these

processes. And that's great maybe for, you know, a model that's going to be your fundamental new recommendation system, right, for your e-commerce. You want to get that right. You want to roll that out, make it out. But there's a lot of things that are happening along the way that you just need to expose to other people to help them help with the work or just expose the kind of earlier things that you're doing as a prototype and say, hey, are we on the right track here?

Is this what your team needs? Is this going to happen? And so what Streamlit does is we make it really easy for data scientists to just quickly spin up these tools, right? Spin up, you know, take the data that they're working on, the models that they're working on, slap a front end on it really quickly, and then share it out to somebody, you know, on the business side, somebody else in the company, so they can start using it, right? And sometimes that's using it to just say, hey, does this thing even look right to you, right? You know, is this data makes sense, right? Kind of with what you're seeing live or is there something wrong with it? Sometimes it might be on the complete other end, which is like, great, we made this recommendation

system for this purpose, but hey, sales, maybe you could take it and, you know, upload a customer, you know, data CSV and get some recommendations, right, for what your customers can do. So there's lots of kind of little things that I think that teams are working on all the time that just, if we built, you know, kind of more and smaller tools, and if we had an approach that is basically more about sharing early, sharing often, right, can really unblock, you know, and the people down the road, as well as kind of to trickle out kind of more of that data insight through the rest of the organization. So specifically, how can we do this? How can we bring machine learning models into people's hands?

How does that work functionally? Yeah, well, I think there's two parts of it. First is, let's agree that that's what we should do, right? And I think that that is kind of a corporate mindset, right? That people have to have, which is it's okay to show things in progress, right? It's okay to involve other people in this process, right? To take a look at things. And honestly, sometimes that's the hardest part because it is a little bit scary to show people, you know, things in process, you know, and to trust that they're going to,

you know, ask questions and say things about it. But, you know, I always say just, you know, just put a giant work in progress up top, right? You know, put a paragraph of explanation about what you're doing, but the earlier that people see things, you know, and can provide feedback on it, right? The more you're able to course correct, or the more you're able to say, you know what, actually this is enough, right. This is all I really needed was this answer, right. You know, now we can move on to the next thing. So there is that kind of, you know, cultural shift, you know, are you ready kind of as a company to do this kind of launch and iterate and have these conversations. But if that, you know, that is something that you want as a company,

well then specifically, you know, the Streamlit specifically the Streamlit approach is we have a very popular open source library that's all in Python. And it allows you to take the code that you're already writing in Python. They're already writing for your models, for your datas. And with just a few kind of extra lines of code, you can put a front end on it. So it really kind of works naturally within the flow that data scientists, machine learning engineers are already doing, right? You're not asking them to just kick off a new two month process, right? Where they're having to build a front end and reactor or flask or whatever. You're literally like, I've had people say to me, wow, you know, another team did this in three and a half months.

I did in six hours, right? And that's kind of, you have to really bring down that time. If you want this rapid flow, it's very nice for me to sit here and say, Hey, you should share things with people and you should share things in progress. But if it's not easy to do that, if you can't do that really quickly, then it all falls apart. And that's where Streamlite I think really shines is that, that kind of ease and speed in terms of just getting something spun up and shared with somebody else. That's really powerful. Obviously taking those timelines and, you know, order of magnitude difference, or maybe even more than one order of magnitude difference. Does that put a

strain on data science teams? And what I'm thinking of is, okay, so once the business learns that they can ask these questions and answers, does it actually have the opposite effect of just sucking up all the time from the data scientists or, or how do you see folks culturally, you know, competing with that? Yeah. You know, I often tell my team, you know, the, the problem with data is that data always leads to you wanting more data, right. You know? And so, you know, once you see something, you're kind of like, well, wait a second, you know, the problem with data is that data always leads to you wanting more data, right? You know, and so, you know, once you see something, you're kind of like, well, wait a second, you know, what if we looked at it this way? What if we looked at it this way? And so, you know, I really think it is, there are cultural processes you have to add kind of along a lot of things because you can go, it is just, it is fun to look at data. It is fun to play with data. And

we would all like to have more, but I think you really have to have discipline, especially your executive layer, right? Who's kind of a, you know, allowing these things to say, is it actually going to change the needle, right? Is it, is it going to move things, right? If what we're deciding is, am I going to send, you know, email template A versus email template B, right? Like, let's be honest with ourself about like what we think the lift is going to be between these two, right? And then decide if we need, you know, to get a whole bunch of more data versus kind of putting something out there and trying it. And so I think you're absolutely right that it can lead to these, like once we unlock this, the sales team could be like, well, can we do this? Can we do this? Can we do this? And so, you know, there's new things I think you kind of have to build along

to say like, okay, we're going to stop this now, right? Or we're not going to revisit it for a while. But fundamentally, I think it comes back to just people understanding the decisions they're making and the impact that they're going to have. And then kind of, you know, focusing the efforts around that. Yeah. Cause I can see that in a way you could make it almost too easy for people to build something and roll it out. I mean, I'm thinking of, of basically like a, I don't know, a black hat, black box where somebody, you know, comes up with a thing that provides a lot of wrong answers and they make it look pretty and get into everybody's hands and say, hey, here's the magical Oracle that uses magical machine learning and artificial intelligence to answer all your business

questions and, you know, doesn't. Or, you know, more, you know, to be a little kinder to our audience. I could also see people just making a mistake, you know, rolling something out and not realizing how it was going to be used or not realizing some of the data that it was going to be encountering and giving the, you know, putting a tool into somebody's hands. That's going to make a mess of things. Is that, is that a risk as well? I think it is, but honestly, I see it kind of the flip way. So a lot of what I hear today when I talk to leaders kind of in data is they're like, I can't do my job that well, because it takes so long for people to do these

analysis. And they're often these silos so that by the time they bring something to me, right, it's almost impossible for me to kind of course correct, right? And go and be like, well, wait a second, what was the assumption there? And they're like, I'm fundamentally a very smart AI data professional, and they're sending me a PowerPoint slide. And I'm like, well, wait a second, right? You know, what happened here, right? And so we kind of need a way to share more often and actually to, you know, to catch those things, right? And kind of, you know, correct and kind of, you know, look at the data together and do that. And so I think you can have that risk in some ways in terms of saying like, oh, we sent this thing and now, you know, somebody's off on that. But I think one, you know, by, by keeping things more in code, right. And specifically code, you know, not necessarily

like a notebook where you could have, you know, errors and different types of things. Like it really helps with that, especially in terms of managers really understanding what's going on. But two, you know, a lot of what I also hear is because it's hard to share things. We often have people who are replicating work or doing the same work across things. Like we don't even know it. And you end up in this situation where it's like, we've recreated the same table four times and, you know, two people did it correctly and two other people did it in different ways. And so because of that, and because we're, you know, we're branching off and creating these, you know, two people did it correctly and two other people did it in different ways. And so because of that and because we're, you know, we're branching off and creating these, you know, giant tools, it's actually more likely that instead of just seeing something early at a stage, generally often where we could correct it, where we're having these other things kind of trickle out there and which are much harder to find and much harder to correct.

So I think you've kind of started to answer this question already, but I want to dig in a little bit deeper and just kind of get closer to the lens maybe, which is if the goal here is putting more data-driven decisions or, or if it's even a good idea, which is, can I create a way for the business person, the salesperson, the marketing person to just get access to the data and see the visualization themselves without having to run this through some process, or is that a bad idea? So I think it's a little bit of both, right? So, you know, data scientists in their job, you know, there's a number of things that they're doing. Right. You know, they're they're they're correlating data. They're making sure that the data is good. Right. They're they're asking different questions, you know, formulating things kind of in the right way. And, you know, data is constantly changing. Right. You know, and so, you know, we're changing up where we're getting data from. Right. And how things and so there's a need to keep on top of that. And I think that to the extent that we have, you know, let's say a static question, right. Something that's more of kind of like a dashboard, you know, absolutely. Right. You know, we really should just be shipping those kind of, you know, more often, you know,

things that are just like, look, this is just a SQL query, right. You know, we're going to look at it this way and we should let everybody kind of, you know, in the sales org, just be able to click into their customer and, you know, do some predictions and things, you know, kind of on top of that. But, you know, most businesses that I work with and talk to, there's just a constantly evolving, right, set of things that are going on, new questions that are being asked, new things that are happening in the marketplace, you know, new things that we need to know. And so that's where I think you meet this really tight loop with your data team, right, in order to say, okay, I've been seeing this with customers, but my intuition now says, right, it's not, you know,

what they do in the first seven days, it's actually the number of people that are added in the organization. And we don't have that built up and maybe we didn't even collect the data on that. Right. And so now there's a whole new conversation on that about like, Hey, can we, can we take this new intuition that's coming, you know, from the business team and work that back in. And that's where I think you really need that, that type loop is for, for all that kind of new stuff that's coming in. Yeah, because it seems too often that once you try to productize machine learning, you suddenly end up with this sort of giant, ungainly software development enterprise where people are putting together massive apps and it takes months and months for them to implement new features and all that, when really what we're trying to do here is we're trying to have a conversation.

We're trying to have a conversation with our data and with each other about what is this telling us and how is this? And so in a way, what you're saying is it's not about creating sort of the end all app. It's about getting it out there and getting it in people's hands so that we can play with it, right? Yeah, 100%. app. It's about getting it out there and getting it in people's hands so that we can play with it. Right. Yeah. A hundred percent. And I think we've all been in that situation where you, you, maybe you unloaded all of your requirements and you waited a number of months. Right. And then this giant tool rolled out and you were like, well, that's, that's not exactly what I wanted. Right. Or we have this new need, right. That we forgot about, but, but now we have this big kind of

monolith, right. That we have to wait for. And it's, you know, it's kind of, you know, too hard to change. You know, what we see at Streamlit is, you know, 90% of the apps that are getting created only live for a week. And that's actually great. It's a complete kind of, you know, upending, right? I think of how people think about what we should be doing with data and apps, which is you often just have a question, right, that needs to be answered. And you can go up and you can look at it. You can share what the share it with the data teammate, right? And just be like, does this, does this data look right? Can you check, you know, my assumptions on here, do things like that, right? And then boom, it's done. You move on, right? You can have a conversation with the

salesperson and say, you know, let's spin something up and, you know, see if this makes sense. And, and if it does, maybe you take some of that code and you put it into kind of a broader dashboard, but it allows for a flexibility, right? That you have in conversation to take the new things, right? That your people that are, that are talking to the customers, that are talking to the market are learning every single day, right, and bring that back into the data stack, right? And so that's really just a continuous flow, right, of information's assumption, really more iterative than we've generally seen to date. So that's wild. I mean, I hadn't thought of that. I didn't know that that was such a short lifespan on average there.

Is most of this like internal use for sales and marketing information or I don't know what I guess maybe what are the use cases that you're seeing most often for these kind of quick hit questions that are getting answered? Yeah. So, you know, we we really see kind of the full spectrum. Right. I mean, Streamlit's in use, I think, at over half the Fortune 50 and really every single kind of industry that you could mention. And so we see everything. If you really just mapped out kind of the ML process, right, every single stage, right, things are happening. So we have a lot of teams that use it mostly in R&D, right? And they're doing drug discovery.

They're predicting vehicle movements, right? They're doing these things and they're really just saying, hey, I need to visualize a part of it. I need to share an assumption with somebody. I need to say, hey, look at this thing that I did, right? And that's, you know, just become part of their rapid discovery process, right? That they're doing kind of in between, right? These data professionals, the machine learning professionals. And then you see it going, you know, all the way to like, we are shipping things to customers, right? You know, where you have, you know, a lot of, you know, consulting groups or, you know, media groups and stuff who are saying, Hey, I made this thing.

And I want you to understand how it works, right. I want you to be able to use it right. And they're creating things for them. And then in between, right. It's all of the things that you were mentioning, you know, Chris, where, you know, you're making things like, Hey, our product marketing team needs a way to kind of upload a list of customers and say, what coupons should we send them, right? You know, let's give kind of a recommendation, you know, engine for that. Our call center team needs to be able to see, you know, the metadata, right, of what's happening on the call and like a prediction, you know, of like, you know, what's happening, right? Managers need to be able to look at that

to understand, you know, which call center operators, right, are, you know, doing a good job. Hey, we need to look at our social media data, right, to understand, you know, which call center operators, right. Or, you know, doing a good job. Hey, we need to look at our social media data, right. And understand, you know, which campaigns, you know, are trending well or not trending well, or how our competition is doing. So, you know, basically I think everything that you can think of that AI touches, which is literally everything these days, right. You know, could have some application on it because almost invariably you have somebody else that somebody else might be the be the ML engineer who's sitting across the room for you or sitting off in Singapore. It could be a customer. It could be someone sitting in sales, but especially in this kind of distributed work environment, it's just so important to share what you're working on early

and often so other people can learn from it, give their own assumptions into it. And yeah, and if it only lives for a week, but it moved you forward, great. And if it lives for, you know, years, and that's the main thing that your company works on to make product recommendations, you know, that's great too. It's really about just creating what your company needs at that point in time. So, I mean, it's, it's just really, it's a whole different world of applications than people are used to, I think, because I think a lot of people really do. I don't know, they think, you know, you're going to kind of build an application, put a bow on it, and move on instead of, you know, let's have a sandbox, let's play, let's experiment and explore. And in a way, the software development angle kind of mirrors the interactions that we're trying to foster between

data science and the business. In other words, we would love to get data scientists more into the conversation with the business that they're trying to support, but it's hard to do that, just like it's hard to make software applications do that. And so what I'm hearing is that in a way, this is a metaphor for everything that we're talking about here. I mean, this is the application equivalent of MLOps in a way. I like to think so. And I do think it is just kind of an underserved part that we don't think about, right? When we just think about like, what are our companies actually need to move faster? What do the people, you know, who are making the decisions need to do that, right? And to do that today, right? And then how do we, how do we create those bonds,

right? You know, and I think that the way to do that is through these, you know, applications, you know, we often call them like micro apps or ephemeral apps, right? Things that you just create the same way that you might create a notion doc, right? Or a document to just write something down. And then that lives right for a week while people, you know, edit on it and try to get the messaging right or whatever it is. I mean, I think that's the attitude

that we also need to take, right? To data as well. Yeah, and I really liked that it can be kind of as simple as you want it to be. I mean, really this reminds me of, you know, my early scripting days or even current scripting days, but, you know, just kind of shoot something off, get the job done. Of course, this does it in a much more beautiful way where I can actually show it to people instead of having to

translate, you know, the gibberish that comes out of my script into something that can communicate with other folks. Yeah. Yeah. And I think it's, you know, well, I like to think that we have a very beautiful tool. It's actually something that we focus on kind of a lot. And part of it is because you are asking people to understand, to trust. And I think that, again, there are these kind of trust barriers, right? Sometimes between the data team, the ML team, right? And the rest of the org, because we don't understand where they got that number, how they're doing it. And so, you know, just exposing that, I think really often helps data teams show their impact, right? And say, this is what we did for you, right? You know, we have companies that make, you know, these giant, what we call them portals, right? Which are just basically like, hey, marketing, right? Here are like the 18 different things, right, you can look at, right, to kind of help self-serve some answers. And if it's not in there, right, give us a call, right? We'll have an office hours, you know, a lot of companies that we work with have started implementing hackathons where you have a chance, right. For business users

and data teams to come together over two to three days and say, Hey, I have this question, right. I have this hunch and I bet that we could answer it in data. Right. And then let's do it. Let's get out of the normal kind of business process and let's see if we can, you know, find some synergy, some innovation, just in terms of, you know, talking and working directly together. Yeah. And there's two pieces of that. I think too, some innovation, just in terms of, you know, talking and working directly together. Yeah. And there's two pieces of that, I think too, right? One is the trust aspect, which is, you know, to Steven's earlier point about Merlin, just coming back and saying, yes, it's really hard to believe him.

But for some reason, if he comes back with like a pie chart that shows you, yes, you know, at least you can see some of the data, you see some numbers and it becomes a little bit more trustworthy. And then the other piece is there are some data problems that just really need to be visualized. And looking at a screen full of numbers is just really hard to connect the dots. Whereas you put that data on a map or into a bar chart and all of a sudden, you know,

really things start to click and come together and the answer's there. Yeah, yeah. I mean, literally I was in a meeting yesterday where if our analyst had just told me that, I would have said, no, it's wrong, right? Like everything I'm hearing from the sales team basically says this, right? But they were able to present it to me and be like, look at these two histograms side

by side. And I was like, oh, actually, I believe that now. And I had to dig into, I was like, show me the actual data. I was looking line by line, right? And then I was like, okay, they're right, right? We've actually been wrong, right? In terms of how we were approaching that. Right. But I would not have believed it. I would have written this person off. Right. You know, if they had just told me,

you know, the answer is no. Yeah. And that's such a problem. I think with a lot of well, I think that, you know, it's right there in the name, you know, data science, they want to be, it should be a science. We should listen to the data. We should go where the data directs us. But without getting political, as many of us know, sometimes it's difficult to go where the data wants us to go if it contradicts our preconceived notions about things. And that's sometimes a challenge. Well, and, you know, data, you know, it can often tell us what, it can tell us where, but it very rarely tells us why or how, right? And so, you know, you can look at that and you can be like, okay, you know, 20% of our

leads are now coming from here or our bugs. But, you know, unless you're really talking to the people who are touching that, right, it's really hard sometimes to understand why, right? And the why is incredibly important because the why is when you get to strategic changes, when you get to, you know, you know, redoing your sales playbook, when you get to kind of new features and things that you, you know, you want to dig into. And so, you know, data is an incredibly important part of the story, right? That we need to move into more things. But, you know, if you're not talking to the people who know how to implement, who know what happened in the past and maybe why we collected that data

or did something in the first place, you're really only doing part of the story and therefore kind of only informing part of the decision. Yeah, and that reminds me of kind of my famous or my favorite drum to beat, which is this idea of putting data scientists together with domain experts. We often forget to connect them for some reason.

And it seems like this idea of kind of quick prototyping, put something together really quickly. Let's look at the data. Maybe allows that conversation to happen a lot faster than again, as you were saying, right. You talk to the expert, you go off for three months, you come back and they're like, no, no, no, this is all wrong. Go spend another three months fixing it instead of just being able to kind of, you know, actually work through it in almost in real time. Yeah. And you see a lot of kind of things it's like, well, let's, let's train our data scientists to be business people. Let's like basically get them MBAs or like, let's train our salespeople to understand data. Or I'm like, or let's let them focus on what they do best, but just get them in a room, allow them to kind of speak a

common language, right. So to look at things together, right. Understand kind of what's going on, right. And kind of, you know, let them do the, what their best at. Yeah. And I think that that's really the theme of this whole conversation right there in a nutshell, isn't it? I mean, the idea is to get back to the premise at the start there, we really want to try to get data scientists, data science into people's hands. And this is a way to do it literally, as well as figuratively. And by getting it into people's hands,

it will foster a better conversation, better interactions. So thank you so much. It's been a really great time. In fact, it's kind of been a whirlwind. It's kind of hard to believe that we've already reached the half hour mark and it's time for the next part of our podcast, which is our tradition of asking our guests three questions. So as a note to our listeners, our guest has not been prepared on these questions ahead of time. So we're going to get their off the cuff answers right now live. This season, we're also mixing things up a little bit. I'll ask a question and Chris will ask another. And then the third question actually comes from a previous podcast guest from season

three. So I'm really excited to be able to surprise you and also to hear what you might have to say for another guest in the future. So I'll kick things off. New question. Is AI just another aspect of data science or is this truly a unique field of study? I feel like you really threw me into the deep end here. There's, there's a lot of opinionated people on this. I'm going to offend some people kind of either way. I mean, I think it's, maybe this is just a very 2020 kind of style answer. It's a spectrum, right. You know, and that's kind of, you know, a lot of things. And so I think that, you know, AI is, you know, building off of data, but it is doing it in a different

way. And I think that, I guess the question is where you're, if the question is basically like, how do we operationalize these things, right? How do we build processes around it? Then I do think AI is different, right? It is handled in a different way. You need different tools. You need kind of different ways of doing things, right? The same kind of checks that you would write on software or certain data things are not going to be the same things you write on AI models. So I will say that they are different because of that. Awesome.

And this one, I'm really excited to ask you this question, actually, just based on your background. And I think you've probably seen some things and thought about some things in different ways. So can you think of an application for machine learning that has not yet been rolled out, but will make a major impact sometime in the future? Oh, interesting. Let me think about that one. That is a really, really tough question. I'm trying to come up with a good answer for it. You know, my brain wants to be like, say stuff on climate change or, you know, the truth of things like that. You know, I'll stick to my my guns in terms of kind of, you know, what I'm good at, which is more kind of the business side things.

I don't see kind of a lot of applications right now which are more about some of the core stuff kind of in marketing, in product, right? In terms of, let's kind of predict what's gonna be a good product mix, right? In some ways it's gonna appeal to people, right? Or what's gonna talk to them and what's gonna be appealing. I mean, that is still, I mean, for good reason,

maybe still kind of in very much the creative side of things, right? And people who have a kind of a lot of data. And so that's something I would probably expect to see emerging over the next decade or so is that, you know, some of the stuff that you probably get trained to do today as a product manager or product marketing manager, which is like, go interview 12 different users, right. You know, map that yourself, you know, figure out exactly what that looks like, you know, what that thing is, the stuff that we're kind of naturally doing

rapidly by like, we say one thing to one salesperson, then we say another thing to a different customer and another thing, right? I could see AI playing actually a really big role in that in terms of, you know, helping suggest what we should position, what we should say, what we should try. And as you said, at the end of the conversation as well, kind of busting some of those preconceived notions that we may have about what the correct answer is, and instead maybe listening to the data more. Yeah, absolutely. So as promised, our third question comes from a previous guest. The following question is brought to us by Leon Adato, host of the Technically Religious podcast. Leon, take it away. Hi, my name is Leon Adato. And as one of the hosts of the podcast,

Technically Religious, I thought I would ask something that has something to do with that area. I'm curious, what responsibility do you think IT folks have to ensure that the things we build are ethical? Oh, interesting. Well, let me say, I think we all have a responsibility, you know, in terms of kind of saying that, that something is ethical. And I think that when we look at a lot of cases, you know, where things have gotten very far down the line, it's because, you know, a lot of people didn't speak up sooner, right. And say, I'm not sure we should do this, or I'm not comfortable with this type of thing.

And I think that it's, it's very easy to get steamrolled when things are like, well, the product's about to ship, right. Or, or we have to do something. Right. And so the only way that we really kind of fix a lot of these things is, you know, for it professionals, but also everybody else who's kind of involved in some of these decisions stepping in a little bit earlier and saying, you know, let's, let's do a check on this. Let's slow this down by a few weeks, right? So we can do that. I don't think it takes one person or one group. I think it's a lot of people kind of raising their hands, asking questions and making sure that we're, you know,

we're staying on a path that we can all be proud of. Well, thank you very much for that, Amanda. And we're looking forward to hearing what question you might have for a future guest. And if you, our listeners, want to be part of this, you can just send us an email at host at utilizing-ai.com, and we'll record your question for a future guest. Thank you very much for joining us. Amanda, where can people connect with you and follow your thoughts on AI and other topics? I'm probably not like a lot of your other guests, and I'm not very active on social media or stuff. I will say a lot of my thoughts end up

getting kind of trickled out in the Streamlit blog. We're pretty active over there. So if you head over to Streamlit.io and click on the blog, you can might not have my byline on it, but it's probably got some of my thoughts in there. How about you, Chris? What's new with you? Everything new with me can be found at chrisgrundeman.com or you can follow me on Twitter at Chris Grundeman. And also LinkedIn is a great place to have conversations.

And as for me, I just hosted our Cloud Field Day event a couple weeks ago in California. It was our first hybrid event with some presenters and guests in person, as well as some connecting virtually. And of course, as you might expect, the subject of AI ops came up a few times, and I was very proud of the delegates making sure that we were not AI washing things. So if you go to techfieldday.com or if you go to YouTube slash techfieldday, you can see the videos from Cloud Field Day 12 in November. Thank you for listening to the Utilizing AI podcast. If you enjoyed this discussion, please do subscribe, rate, and review the show since it does help and share it with your friends. This podcast is brought to you by gestaltit.com,

your home for IT coverage from across the enterprise. For show notes and more episodes, go to utilizing-ai.com or connect with us on Twitter at utilizing underscore AI. Due to the Thanksgiving holiday in the US next week, we will not be publishing an episode, but be sure to tune back in on Tuesday, November 30th for the next episode of Utilizing AI.