Mark and Brent catch back up with technology author Tom Taulli. Tom’s latest book “Implementing AI Systems” is a comprehensive and approachable guide to AI and a great companion to any automation practice. The conversation traverses the concepts in the book and how AI can make real world impacts for organizations.
Google Play / Apple Podcasts / Spotify
Tom Taulli, Mark Percival, Brent Sanders
Mark Percival 00:06
Well, Tom, thanks for coming on the podcast today, everyone, this is Tom Taulli. We've had him on before to talk about RPA. But obviously, he does a lot more than just, it's not just an RPA or UiPath expert, as you'll frequently see him on Forbes four. But he's also written several books in the programming space, more recently on artificial intelligence books on the basics of art of AI. And then most recently, that just came out, I think, in the past few weeks, which is implementing AI systems. And so we wanted to have Tom on to talk a little bit more about exactly that implementing AI systems. And we thought it'd be interesting for our listeners, because in the RPA, space, there is a lot of stuff going on with AI, there's a lot of it that might be, you know, oversold as AI. But there is still a lot of activity in the AI space. And there's a lot of opportunities that we see that Brent and I see with clients where AI actually is already there. So without further ado, let's jump in with Tom here.
Tom Taulli 01:02
Great, thanks. Thanks for having me.
Mark Percival 01:03
Yeah, thanks. maybe it'd be interesting, I think, to just go over the, you know, what brought you what, you know, got you to write this, this next set an AI set of books that you have, including AI systems. And we'll talk a little bit about that, if you could.
Tom Taulli 01:20
Yeah, it seems like what happens is one book leads to another, the AI basics came through my writings through forbes.com. And I realized that a lot of people I talked to founders or companies, and they didn't quite understand sometimes I could tell when I talked to them, maybe the difference between machine learning or deep learning. And I thought, if they're having problems with that, just have the average person in or in a, you know, business person or a manager that all of a sudden, they have to take on a project, and they don't know what's going on. So that was the thinking of AI basics. And then there, I had a chapter on RPA. And then my editor wanted to write a, you know, have someone write a book on RPA. And she noticed I had a chapter on RPA, my AI book and she said, Do you want to write a book about RPA? And I said, Actually, I actually have been interested in it, and I was going to actually talk to you about it. And funnily enough, you bring that up. So that led to the RPA book. And then through the RPA book and the AI book, and just through Forbes, I could just, you know, I go through all the IDC reports, and so forth, it's pretty common that AI projects fail. Lots of money go into it, and it's not such small companies, as big companies, as tech companies really doesn't matter. And I thought, if that's the case, maybe there should be a book on how to implement AI successfully. And so that's really the, you know, the pathway to this book. I don't know what this book is going to, if it leads to another book. But those are how it all came together.
Mark Percival 02:59
Yeah, that makes sense. I mean, the RPA space is definitely seeing a lot of issues. I think exactly that implementing AI and doing it successfully is not trivial. Right? Your book kind of has this, and Brent had brought this up earlier, right before we started recording, which was your book does a good job of he mentioned, essentially, on the technical side, and on the just higher level, ai side, everything from hiring, which I thought was interesting, you had an entire chapter, basically about building a team, to the more technical pieces, which are, you know, actually training a model on a data set. And it'd be interesting to start with just a high level, because I think our audience maybe is in that same position where they don't actually know the differences between these different terminologies, things like machine learning versus deep learning versus AI. Can you give us kind of a quick overview on that? I know there's a lot more to it.
Tom Taulli 03:50
Sure, yeah, I look at AI, there's a famous example, the circles example. So you have one big circle, and that's ai, ai is everything that has to do with taking data and getting insights out of that data. And then there are smaller circles within the big circle. One of the smaller circles is machine learning. And machine learning is more of a traditional approach to AI. And more about business almost more like business intelligence, where you look at some data, you say, you know, will I be reducing churn if I do XYZ? Or maybe I can, you know, you know, just just basically get better insight into my business through machine learning. A lot of it is more traditional statistics. And then, you know, there's the old joke about regression analysis. You know, people say, well, regression analysis, AI is AI. Well, it's kind of more of machine learning. But it's not the most sophisticated thing because regression has been around for a couple 100 years. So, you know, in machine learning is one of the early forms of Ai going back to the 1950s, right, so then you have another circle, which you call deep learning. And deep learning has actually been around for a long time, it's been around since the 50s, through neural networks. So the idea is you still taking data in, but when you take the data in each of those points of data have different weights and seals assign some random weighting to the different data. And then you there's this thing called a loss function that just kind of cranks through this data, constantly comparing to the end result and seeing if there's a match, if there is a match that there's a higher weight, if there's not a match, you know, you don't get the wait. And this just happens a lot with huge amounts of data in which you can get out of all that is, it could recognize my face and say, That's Tom, or it could look at a, you know, a pedestrian on the highway and says, you know, as a pedestrian right in front of the car move. So, you know, that's, that's really the more sophisticated AI is deep learning. And then you have NLP natural language processing, which is how to understand language that's been around for a long time as well. So AI, like I said, it's this huge category. And then there are these smaller categories. But they're not necessarily small either. Because there's a huge industry around deep learning, you know, just just do a Google search, and you'll see hundreds of millions of links when it comes to deep learning. So it's a big topic, but it's part of a bigger topic. If you want to even go more cosmetic, you can say that there's really two forms of artificial intelligence. There's the strong form, which is what we see in science fiction movies, where we have androids and we can talk to them. We don't know, you know, Philip K. Dick, and we don't know if we're talking to a person or not, or trying to kill us or, you know, take us over. You know, we're not, we're not there yet, of course, and then there's weak AI. That's where we're at right now. It's limited forms of AI, where we have discrete use cases for this technology, although it's getting more and more sophisticated as these models get bigger and bigger.
Mark Percival 07:06
Yeah, that's my guess that's, that's Alexa. Definitely a weak form AI for me.
Tom Taulli 07:10
Mark Percival 07:13
Yeah, there's the other piece, I thought I'd actually thought this was interesting. You went into the history of AI. And I'd heard a little in my past, you know, obviously, I've read a little bit about AI and heard these terms, and you had mentioned going back to the 50s. And there was also this thing that was you know, people talked about this, I think about a decade ago, the AI winter became a big term, they'll talk about this, this long period of time where we had a lot going on and in computers, but in computer science, but the AI seemed to just stagnate. And then it really started to kick off, you know, the 2000s. But really recently, it seems to, you know, obviously Siri and these things started to come and become more consumer friendly. And we started to see AI on a day to day basis. What's interesting now is it feels like we're still, you know, we bring up Alexa, it's not that great. But we also see a lot of things that are pretty amazing that they're doing with AI and one was computer vision. For me, that was the moment where I was like, oh, wow, this is unbelievable. I didn't know this is impossible, you know, AI has this problem of like, figuring out how to, like, get me to a location or to recognize my voice sometimes. But then in other cases, it's doing amazing things on computer vision. You know, you mentioned that in history. But going back to that, like where did where it really kicked off for you where you notice, like, Hey, this is AI is kind of becoming or it's, it's real, it's it's my day to day life?
Tom Taulli 08:31
Yeah, I've kind of a bit of a science fiction geek. And so since I was a kid, I've been fascinated with AI. But I also realized I needed to make a living. And, you know, when I got into computer business, I saw that there wasn't really a, you know, the technology was too, you know, too nascent. And the opportunity was not there to make a living out of that, to make a business out of that. And like you mentioned, last 10 years, what we've seen is some huge innovations of breakthroughs, multiple trends for that, you know, computing power has gotten stronger. We have seemingly unlimited amounts of storage. We have huge amounts of data from companies like Google and Facebook. We have innovations in theories around deep learning from Geoffrey Hinton, and people across the world. And it just kind of all came at once. And so it's probably the last 10 years. And I think a lot of it has to do with just the fact that we have, we have smartphones. You know, it's just Geoffrey Hinton said, you know, will there be another AI winter? And he said, Absolutely not. He says, it's because we have a smartphone. It's like everyone has a smartphone. All this technology has now become part of our daily lives is using data, getting insight out of that data. It's not going away. It's not going away. So for me, you know, once I saw companies like Sequoia Get into it, Google Facebook. And I thought, you know, this, it's it's big, it's important, it's finally hit its stride. And there's no turning back. I mean, it's just, there's no turning back.
Brent Sanders 10:13
I think the other interesting part of that is, to what you're sending around, you know, we all have a device in our pockets. So that just the supply of data feels like it's almost infinite. Now. It's our location, it's I mean, it was we're constantly sending signals, back and forth to these towers, and they're being collected. And, I mean, how many times people log into Facebook, what they're posting, what they're looking at how far they're scrolling, all of those companies that are making these investments also have these, you know, raw pipelines of data, which you mentioned your book, which I found to be a really great way to describe some of the pain, as somebody who has, I've taken a couple of courses in machine learning and AI, I went to a seminar here in there, and you quickly realize, for you to apply these techniques, in your own projects, you've got to have the data game figured out. And you have to have information in a format in a way that, again, you bring this all up in the book and do a really good job of detailing it of getting it into a format that you can work with and that you can either be training models on or dealing with. And sometimes that data is coming in so fast. It's user generated, it's, I think you use some of the examples of like Salesforce and YouTube of like, I mean, could you imagine the amount of information from just playback data comments? You know, all of that coming through? And then how do you like, it's almost the training of the models and building the AI piece of it is almost an afterthought to how do we even get this data in a single format in place? That makes sense. So I have to agree with you that the smartphone is almost like the realization of the data that then is generated, and it's like, okay, with all this data, what's important, what's not and what can be put into a model and be trained and what can yield? interesting insights?
Tom Taulli 12:14
Yeah, in and with the emergence of 5g, right, it's only going to accelerate, you know, we have much lower latency with 5g, bigger pipes. And it's not just for consumers, it's for businesses as well, IoT, I had a chance to talk to Deere, the tractor company, been around for 150 years. I haven't been in a tractor, I've had some family who were in the farming business years ago, that was like 2030 years ago, the tractors were a lot different than even then they're pretty sophisticated. But now these things are just completely connected. You know, data driven systems, Deere knows what, what's going on with every tractor it has. And they just bought spectrum, 5g spectrum, they're going to deploy their own LTE networks, at deer mile, they want to control it. And they want to have their own network where they can, you know, have robots in factories, build these machines in a customized fashion, you know, utilizing data and artificial intelligence and machine learning. So a lot of the companies you would not think who are at the cutting edge of AI, are, you know, they don't get you know, there's a lot of noise in the marketplace. It's hard for a company like Deere to say, Hey, we're the AI people, you know, people or whatever, wherever they are people. That's right. But these guys really are, you know, that's just you know, farming in tractors, just one one lm, you know, one part of this huge economy, you know, oil rigs. I talked to someone this morning from Juniper, he was talking about cruise ships, and you know, it's everywhere. And 5g is going to accelerate it. So it's that we're, we've just been warming up. Yeah, the next 10 years made me amazing.
Brent Sanders 14:11
Okay, you go through some really interesting case studies of applications of AI in the book. I’m curious in fact, in our prior conversation we talked about in the RPA book, similarly, I'm curious, like, what access Did you get with the companies that you talked about Intuit zillion Halliburton cadence? I mean, there's tons of them. Like, how much access Did you get for the book versus, you know, relying on public sources? How did that go?
Tom Taulli 14:38
Most of those are talking directly to the companies themselves. Most companies are open and willing to talk about it. They're excited to talk about it. They want to show their Cool, cool gadgets and systems and innovations. You know, we're going through a period of digital transformation companies are spending more on this no Andreessen said, we're all we're all becoming, you know, software's eating the world. And I think these bigger companies fortune 500 have gotten the religion. And they're, they're serious about it. And so yeah, I think some of the most interesting stories I've seen in AI have not been from tech companies, but companies like, like the deer or Yeah, you know, Halliburton, you know, you know, dealing with how to find, you know, better find oil, you know, or how to best maximize taking oil out of the ground. And I mean, this, there's so much that can be done and is being done. So, yeah, so yeah, I've had an opportunity to talk to some interesting companies directly.
Brent Sanders 15:42
And knowing and this may be covered in some of the aspects of the book, but in knowing that this can be somewhat of an imperfect science and requires some trial and error of the companies that have been successful with it, is there a common trait that you'll find where, you know, when I think of an experiment, or even like a software project, that's a failure? It's like the, the ones that realize, okay, this is the right time to stop and change gears or to abandon and readjust? And is there a common thread between those companies that are able to make those decisions?
Tom Taulli 16:17
I think that one common thread is in this is really the first, you know, part I talked about is, you know, what, what problem are you solving, I think the clearer you are in real, and the more realistic you are about what problem you're going to solve, it's going to make a life, it sounds so easy, but but in a company, you know, you have a CEO, your CFO who doesn't want to spend the money, and, you know, cio wants to everything, and CTO and all these see people, and then you have everyone underneath that don't want to lose their jobs, or you have data that's all in these different silos. And, and so it's hard to, to come up with that unified message. And everyone, you know, believes they have the problem that needs to be solved. And so it, you know, getting some clarity of focus on what the problem to be solved is, I think, is important. And that problem should not probably should be more back office problem, not a front office problem. You know, the worst thing you could do is start experimenting with your customers with AI. You know, all of a sudden, you have these horrible customer experiences, and then you've lost a customer and, and you don't get them back ever again. So start with the low, you know, the traditional low hanging fruit in the back office, you know, RPA is a good example of just, you got invoices, they come in, if you kind of screw up on the invoices, not the end of the world. And you probably should have a human in the loop anyway, with these systems. And so, start there, you know, me and machine vision with invoices, and try that out and see what see what you get. And then and then maybe a certain part of the organization, you start with that, and then you go from there. And as you get as you build your AI muscles in the back office, and get better at it, then maybe start experimenting, you know, from customer-facing parts of it. So it's a journey, it starts kind of with easier, easier first, but also clearly expressed problems to be solved. Like, you know, what, what is it really, what is the pain? You know, what is what is difficult for people to solve? You know, computers are really good at taking lots of information and doing something with it. People are not necessarily good at that, you know, and again, invoices is a great example of that. I mean, it's tedious, it's redundant, you know, it's boring and, you know, no one wants to spend their life looking at invoices. So that seems like a really good place to start, you know, or, you know, one of the top areas to start with, when it comes to AI.
Mark Percival 19:02
You have a chapter just on identifying the problem. Basically, we talk about a lot of this. And then one thing I thought that was interesting in there is, especially from the RPA spaces, you talk about evaluating the problem solution, or what is solving, and one metric A lot of times people use, in our case, our saved but interesting, you pointed out, you know, other KPIs, NPS score risk management, I think the risk management piece is one that's often overlooked because there is the piece that is yes, it saved X amount of hours. But there's a bigger piece, which is that just like that invoice problem, where a human is more likely to make a mistake than a well oiled RPA automation solution. So it's more than just thinking about it from that high level like how many hours Am I gonna save? There's so many other metrics there.
Tom Taulli 19:47
Yeah, no, I think compliance and governance is so important. And it's where automation can be a huge, huge factor, huge improvement over the old ways of doing things.
Mark Percival 19:58
One thing you weren't About i thought was interesting in that chapter was the where AI moonshots? Maybe you could talk a little bit about that.
Tom Taulli 20:05
Yeah, yeah, in the book, I talked about IBM. And by the way, I love IBM, nothing against IBM, they're an amazing company, historically a great company, and today are a great company, I think they're doing some very innovative things, but they're not perfect. And IBM has this tendency to, you know, want to put people on the moon and, you know, cure cancer. And literally, that's what they want to do is use AI to cure cancer. And, you know, talk about I mean, it's definitely a huge problem, no question about that. We need to solve it. Maybe I could do that. But boy, why maybe start a little less ambitious than me. So I think, you know, so in that book, I talk about boiling, you know, don't boil the ocean. And I use the example of IBM, where Watson was focused, invested heavily in essentially curing cancer solving that problem, and it didn't work. It just didn't work. And, and customers were not happy about it. And, and it's a real progress for the company. And I think, you know, the one thing about IBM is, you know, they helped build a Social Security system, the United States, you know, they've done some amazing things in the back office. Yeah. And why couldn't they just start there with Watson? And, you know, I think they could have, you know, really made some huge progress, and, and become a, you know, and then go to more ambitious programs over time as the technology gets better. And I think that's where I think don't boil the ocean, don't try to do everything all at once or else you're getting, you're pretty much almost guaranteed to fail. I would say.
Mark Percival 21:53
That's interesting, because I remember when Watson first came out, it was really heavily promoted because of jeopardy. Great for the average consumer hearing it.
Tom Taulli 22:02
Yeah. Yeah, it was. Yeah. It's all jeopardy. That's right. That's amazing. Yeah, I mean, that that's a huge accomplishment. But there's a difference between solving some trivia questions and curing cancer, right? Yeah. But I gotta tell you that what we've seen with the COVID vaccines, it's not necessarily an example of AI, per se, but it is an example of the use of data. Programming genetics. Yeah. And, you know, so it's maybe not strictly AI, but it does show what can be done with digital technologies across different industries. And, and when, when so many smart people are focused on put it this way, there was a problem to be solved, right? Yeah, this is totally ambitious, and probably should have failed. But it didn't. But there's kind of no choice. We got it. We got to figure this out. And it worked. So I think there are cases where it does make sense to do that. And in this case, it worked out with the COVID vaccines. But with IBM, I think it's a little different in that case.
Brent Sanders 23:09
Yeah. Going back to the, you know, like starting small, I mean, I think you bring up a good point with with invoice processing, it's it's company facing, it's something that we actually always throw out there is like an initial project is like, Hey, you know, how are you doing this? And the actually kind of dovetails with the question, something we talked about, the prior time you're on the podcast was about training, like pre trained models, or purpose built or purpose train models and the economy around that. And I think, you know, one thing that will work with is Azure has a great like, structured data endpoint, where you can give it a PDF or a file, some sort of, you know, give you back a table of some data or JSON file with like, Hey, here's what's in there. And it's a great controlled way, that even maybe without even much of a data science capability within your company, you can start to yield some of that, you know, yield some of the the sort of AI returns, which, you know, I think it's a good way to look at, as you mentioned, defining a, hey, let's define a real problem. versus a lot of the times, you know, the CEO comes in and says, the problem is we don't have AI, right? And it's like, well, alright, let's try to, you know, backfill our way into having a buzzword, which I think, at least what we're seeing, that's definitely going away, people are starting to understand what it is a little bit more and starting to see how it can be used and what problems are good and but in going back to that that same discussion around these sort of pre trained models, have you seen any developments in that economy, like I'm of the mindset that in five years from now, we're going to be, instead of purchasing software, we're going to be purchasing I mean, that's not going to go away, but we're going to be purchasing the models. And going between different brands. And that's really the differentiating factor in the future around AI. As you know, there have been companies that have invested tons and tons and tons of GPU time to, you know, build the best model. And I think we saw a taste of it a little bit with the GPT. Three, at least on Twitter was all the rage, and everyone was having it, write their own stories and make stock trades for them. But do you see that evolving since the last time we spoke that sort of economy?
Tom Taulli 25:31
Yeah, I mean, I think, you know, we talked about the API economy, you know, maybe the model economy is never thought of it that way. But I could, I could see that happening is the company's you know, these hyper scalar, cloud giants like Google and Microsoft, and IBM. And so, I mean, they have so many advantages of scale, economies of scale, access to data, smart data scientists. If you can find an off the shelf, pre trained model that works, then do it. You know, see, save your brain capability for other things that are more specific to your business. We're probably Google can't, doesn't want to play because it's maybe too small, too niche to edge for them. But yeah, so I yeah, I would, that's how I would start these resources are amazing as they are, right.
Brent Sanders 26:26
Yeah. I mean, you know, they're gonna get better. Yeah.
Tom Taulli 26:30
Yeah, yeah. And I, you know, I, there's a company called laserfiche, I'm gonna be doing and be part of their conference, virtual conference, everything's virtual now. And they showed me because they do document management they've been around for, since the 80s, one of the early players and document management, in case management, those kinds of things. And they're starting to get into RPA. And they should be one product that they have where you can, they just allow people, they're just testing out their product, but just allow people to upload an invoice into the cloud. And it'll, it'll map it, map it for you. And I know, Google has something similar to that, as well. And then you could eventually, you know, licensed that, and have it into your own implemented into your own system and maybe tweak it the way you want to tweak it, or it'll learn from the data typical to your to your business, but there's a lot of other things that it can it can detect. So definitely, I think that's, that's the way, you know, a great strategy to employ.
Brent Sanders 27:32
Yeah, yeah, I just think it'll, it'll get better, right? If we just follow this industry, like, all the others sort of technology industries that have sort of blossom before it, you would expect to start to see, you know, oh, there's, you know, you're going to use this product for this specific problem, because they've really focused in on that one space. And maybe somebody like a Google or Microsoft, or IBM, you know, focuses in on that, and, you know, offers more of a suite, and you bind a part of that sweet, but it's really interesting, because I think you, you do a really good job of setting expectation for your readers in this book around, hey, if you're going to spin up your HR department, here's what it looks like, here's what salaries typically look like, what job responsibilities, how these roles report through, I was curious, you know, how did you? How did you find out? And, you know, it's tough to publish a set, sort of here's the salary range, but I was curious, you know, how did you go about finding that information out? I mean, how did the information gathering part of that book?
Tom Taulli 28:36
Oh, it's actually not that hard. Okay, I'll give you the secret, indeed. A lot of information on this. And so what I did is, I took it from this, I mean, I talked to other companies, I knew that, you know, understood the different roles that you would have. And I did every role, I could think of a data engineer, data scientist, designer, which is often overlooked person, but you want to make a product that's easy to use. So you want to get a designer. And then I was looking at different companies. And then I would look at job ads for these different types of roles. And then see what kind of experience companies look for what kind of background, what kind of day to day activities that they engage in? So a lot of it is out there. Just this just takes time to go through and process that information. But that's how I went.
Mark Percival 29:33
I mean, this whole chapter, I thought was really interesting, because a lot of AI books are just in general, I think it's easy to overlook this piece, which is a team.
Tom Taulli 29:43
They, you know, it's they make it sound like there's magically these data scientists appear. And these data engineers show up one day, and then they do this all for you. You know, there's a really good another good book to take a look at Jeff Lawson from Twilio Hey, it's called Ask your developer, I got an advanced copy cut toxin, just recently about and I'll be writing a review of it in the next couple weeks. It comes out in a week or two. And it's about digital transformation. But you know, he's a, he's a coder by background. And he, I thought when I go, you know, when I read your book, I go, I've read lots of books about digital transformation. But you would never hear about, you know, developers, you know, that these people would write about them act like these developers don't exist, or, you know, they just magically show up. And in his book, he actually has names of developers, you know, that he has not just Twilio developers, but great developers at stripe or, yeah, Airbnb, and he writes profiles of these developers. And he said, if you go to like a, you know, a major company, you know, the CEO probably knows the top 10 salespeople, you know, I mean, they, they just know, you know, they probably couldn't tell you who their top coder is. Right. Which is kind of amazing. Because digital transformation is so important. Why don't you think about maybe knowing who your top coders are at your company? I don't think it could be just as important as your top salespeople.
Mark Percival 31:14
Yeah, yeah, for sure. I mean, I think that's Yeah, that's a really interesting point. I had not that interesting inclusion in a book.
Tom Taulli 31:22
Yeah, yeah. I mean, I'll be surprised at CEOs walking down the hall, and walking by a coder and not even recognizing that that person might be one of their most valuable employees. So, yeah, and probably is gotten pitched that day by Google to leave the company, you know, right. Right.
Mark Percival 31:40
Well, that's an interesting point that I bring this up, it's actually this is a heartspace, the higher end as well.
Tom Taulli 31:45
Yeah, very difficult.
Mark Percival 31:46
I mean, this is, you know, obviously, hiring in general, for developers, not easy today. But anytime you kind of have to hire somebody in a niche area that's in demand, it's hard. And then you look at the salaries being paid. And you make a point, pointing this out in the book, but the salaries listed are not necessarily inclusive of all the other things they get, like stock options and packages. Sure. And so as a company looking to kind of build something on AI, it's more than just, it's more than just, you know, going and picking someone, it's also convincing them to come work for you and finding the right person. But you also pointed out I thought, which was also just as important, which is, it's not just about hiring the AI person, there's all these other people involved in this, it's just like a software project that you would normally have you whatever project manager, or these other people, testers, and the data engineers, and I think it's very easy to look at this as Oh, I'm gonna hire this, you know, Uber, ai expert, to come in and just build everything I need. Then you pointed out, the one that I thought was, I think is often forgotten is there's a data scientist, but there's also this data engineer, just this person that goes out and finds the data that you're going to need to train all these models on and cleans it up and manipulates it and gets it into a state that it can constantly be, you know, brought on board and through the data warehouses. As the velocity grows, all that becomes a really challenging problem if you really are building an AI model out or some kind of new product out on AI. And so I think it's easy to overlook all those things. So I thought this chapter was really good, because it didn't just didn't just harp on the AI engineer experience, it was also like, you got to hire all these other people to make this work.
Tom Taulli 33:17
Yeah. And, you know, we've talked about Intuit as an example. You know, I talked to them. And when they created their chatbot, for QuickBooks, one of the persons they hired was a comedian. And who had this experience of, you know, making stuff fun, and making the conversation fun, instead of the typical chatbot discussion, which is not too fun. So, you know, I mean, some companies really, are at the cutting edge of this. And, you know, here's a tech company hiring a comedian. Yeah, this comedian probably thought the last job he did ever have is, you know, working as a, you know, Project Manager for a chatbot, you know, division. But that's the way it is. So,
Mark Percival 34:07
I mean, just basically your point about designers, right? I mean, if you're building something that's easy that people want to use, it's not just a function of solving the problem, but solving in a way that engages people.
Tom Taulli 34:15
Yeah, yeah. And I think that's the thing, too, is like, you know, you're Who are you building this for? You know, who is your end user? You know, is that person, a data scientist? Or is that person on the front line? Or who is that person, and that sometimes gets lost. And when that does, you could have a product that works. But it doesn't matter because you need change management in the organization. And if people just don't adopt it. It's not easy to use or, you know, it'll be a waste of wasted effort.
Mark Percival 34:46
Yeah. The other piece you brought up, which I thought is often really overlooked as rescaling. And Brent, you brought up you had taken these courses and I've done these courses as well on the Udacity and Coursera courses on this stuff. It is a thing that I think people often overlook, because I think there's this idea that if you're going to need if you're going to, you know, build an AI team, you're going to need them some, you know, grandiose AI expert that's from Carnegie Mellon, or MIT. But a lot of this goes back to a lot of it's understanding the tools that are out there and what it can and can't do. And you don't have to be an AI, you know, expert, expert, PhD expert, to get a lot of stuff done, you kind of just have to some extent, it's just like being a developer and going to learning a new framework, it's learning some new piece of technology that you might not have to have, you know, maybe you don't have to have a full understanding of how this neural network works under the hood. But you can get a lot done with just what you're reading or through these re-skilling courses. That's, I think, an interesting place that I think people overlook often.
Tom Taulli 35:44
Yeah, and in the book to talk about Bloomberg, and you know, how they invest a lot in training, video training. I know, I talked to the folks that Adobe is the same thing with them. Now, they, you know, so they'll, they'll use the Coursera has, but they'll create their own training systems as well. And, you know, if someone's good at mathematics and has the affinity for it, you know, they could be turned into a data scientist, maybe not the best one. But then again, that person is motivated. You know, what, why not? So, I think companies are realizing that, and the good news is with the, there's a lot of training material, a lot of it's gotta be careful. But there's a lot of good material out there. And, you know, in our education system, you know, does have its faults, because we don't, we're not necessarily training people for the jobs of the future. And if you know, it's going to, it's going to have to be more probably from industry to do that versus college systems, I would say.
Mark Percival 36:48
Yeah, yeah, this movie, it moves so fast, right? It's so hard. And it is a lot of it is understanding, I mean, just even on the regular development side, right, understanding front end frameworks can be just move so quickly, that you wind up constantly having to learn, I think the same thing can be said for AI as it's just a constant, it goes back to the ambition, if you are somebody who really wants to do this, you can get involved in it, and you can do it. Absolutely.
Tom Taulli 37:12
Absolutely. That is true. I mean, it's at that point where you don't have to show up with a PhD. I think if you show up, and also to, you know, this for career advice is that, you know, you get involved, you know, do something, create a project, you know, and put it on GitHub or whatever, just show that you've done something. Because there's a lot of people that come out of these educational programs that really haven't done a lot. You know, they've done their homework. But that's about, you know, they take some exams, and they pass them, but have they created anything? But to me that that's a true data scientists are true programmers, true developers, like you got to do something. And so just go out and try stuff out. And people will recognize that.
Mark Percival 38:01
Yeah, for me, it doesn't really stick unless I actually, you know, get hands on and do it. But it's been interesting, in the past few years, with AI, there have been all these tool sets that have come out that have made it a lot easier to get involved. Whereas I think before 5, 10 years ago was even more difficult. And you felt like, Hey, this is really hard. And now it's really interesting, with all this content out there, you can kind of spin something up pretty quickly that does stuff that you know, you kind of you're kind of impressed with what you can pull off with, with so little, you know, input on your own just by using what's off the shelf there. And a lot of open source, a lot of it's open source, that is not stuff that you have to spend a lot of money on.
Tom Taulli 38:37
Yeah, that's all free. Yeah. I mean, if you want to go crazy and get a GPU, that's Yeah. But, but you can work, you know, Anaconda, going, you can go download Anaconda for free. Yep. It'll give you the Jupyter Notebook where you can do some Python, there's courses, free courses out there and how to learn Python. Yeah. And then you can look at some of the frameworks. And then, I mean, I, you know, when I, when I went to school, you know, I took statistics, you know, I had a calculator to do all that. Today, I would just be a wizard statistics, regression analysis with no problem at all, but, you know, with just a few lines of Python, and I'm, I'm ready to go. So, you know, for career advice, you know, learn Python. Yeah. And learn Jupyter notebooks, and then, you know, learn some type of framework Kerris is usually the beginners AI framework, you could start with their tensor flows, deep learning, pretty sophisticated, but psychic learn and there's all these different frameworks. And so, you know, pick one, may the easier one, and then go from there, and then you'll just, you will probably won't take you along until you know, you'll be some hot shot, data scientists working at it. making tons of money with that. These great stock option packages.
Mark Percival 40:01
So well the pilot is interesting because it's showing up in more and more places. And it's not just data scientists, it's everything we're seeing in the RPA space. You know, the open source, startup, Robo core, released their tooling, and it's all based on Python. It's all under the hood. It's just a mini conda, which is Anaconda is just a smaller version of Anaconda same project. But yeah, it's all just, it's all just rooted in Python. And so I think if you're coming at it from any role and accompany learning Python, it's going to be useful. Even if you're not a data scientist or getting into machine learning. It's useful for a lot of different tasks.
Tom Taulli 40:35
Yeah, a friend of mine, she works at Twilio, all places, and she's in the finance department, no coding background whatsoever. And she had a problem she needed to solve with some of the reports, and she used Python for that.
Brent Sanders 40:47
Yeah. Yeah, it's a lot easier than forcing that through Excel or some VBA.
Tom Taulli 40:53
Brent Sanders 40:56
I have to plug your book though. Because even in the example that you gave, where you're taking some courses, and learning what I've found in, I've taken a handful of courses now, nobody really gives an overview. I mean, I'm sure there are books that have done it, but your book, I have to give you props, you go deep enough, where, you know, you can understand what's the use case, how it's used? What's the like, the high level math that's involved? And what are the problems that are good to use the technique against, right? So it's, how do you qualify, you know, regression versus something else. And so, hats off to you, the book does a really good job. And so I'd recommend people to get it mainly for that section, which is, to me, that's the meat of the book that was like the appetizers around the why and what does the team look like? But that's like the, the really kind of satisfying part of the book to me. And so I appreciated that a lot.
Tom Taulli 41:51
Cool. Well, and I did talk to, you know, people like at ServiceNow, and different companies and data scientists, you know, so when I look at a certain type of algorithm, I would bring in something they'll try to bring in, and maybe if I can a real real world example of how companies are using some of this technology, because he just there's so many different algorithms, so many different algorithms, and then it gets overwhelming. But the book, I try to look at the most common ones, and give us a sense of how those apply to certain use cases. To make it a little bit more understandable.
Brent Sanders 42:31
Yeah, a little context. Context is so important. Yeah. One question I had is, you mentioned it RPA, specifically around reinforcement learning. Can you expand on what you've seen in the space where companies have implemented reinforcement learning for RPA? I mean, I'm aware of the idea in general, but I was curious, have you seen any specific applications?
Tom Taulli 42:54
Yes, I think automation anywhere might be doing, to some extent, or, I mean, I don't know if it's officially reinforcement learning. But it seems like that could be going on. So reinforcement learning is, you know, Pavlov's dog, you know, you reward the dog, you know, the, for doing something, and it really usually works pretty well. These were for humans as well. So, and reinforcement learning is just that it's a reward punishment system to train a bot or whatever you want to call it. And, yeah, it did, the big use cases of it have come out of the gaming industry. So a deep mind is at the forefront of this. And they were able to beat the top player and go to the Chinese game. It's kind of a chess game, and it originated in China. No one thought it would be beat because I guess there's more potential loops and go than there are atoms in the universe. That's crazy like that. So, you know, obviously, you can't you can't have every potential possibility built into that. So the way it was able to beat the top player is it just simulated billions of different games. And through a reward, punishment rewards approach. And some people think that reinforcement learning could ultimately be a way to get towards strong AI. Yeah. So with RPA I could see it working with processes because, you know, if a process doesn't really work out too well, you can punish it. And if one one thing leads to a much more optimal process, you know, you reward it. So automation anywhere has this discovery bot, where they, they look at your people and they look at what the processes are and how they handle it. And it did look at what some looks like process mining to some extent, but then it looks at what works and what doesn't, it kind of automatically creates a bot out of that. So I think it's, you know, cutting edge, but I think there's some potential there for RPA and reinforcement learning, especially when it's because it's all about tasks and processes, and there's optimal ways of doing things. And it's something that can be fairly easily calculated, it's almost like a game, you know, getting through the maze, right? Getting more points, you know, getting more points in the game of RPA means, you know, you get from point A to point B with is less, you know, steps as you can and get the best outcome. So, kinda like a game. So I could see it now, I think, what a lot of the AI now, well, part of the suit process mining is doing a lot of that probably summit with RPA. And then you're seeing, you know, more computer vision, major use for our, which is more of a different concept.
Brent Sanders 46:10
And yeah, yeah, I was just, it was really interesting to see, in my mind, it's like, maybe it's not there yet. But in my mind, there's a lot of discussion in the space around these self healing bots. And so almost a similar concept of, you know, we're almost like automations, that are able to almost like grow an arm or grow a finger, where it's like, Hey, you know, one thing that RPA is really brittle. And, you know, are there ways that using reinforcement learning that you could potentially, like, in the development of a bot, potentially put some of these falls in front of it and say, Hey, if you notice, the process failing, here are some alternative ways almost like other moves in go where it's like, exactly, these five moves. And, you know, we can try this one first. And if that fails, you know, perhaps you try this one. And then eventually, if it fails enough, you decide, hey, I'm not going to go down that path, I'm going to go down this path, because it's more fortuitous. So just interesting things that, you know, we're thinking about as practitioners, and as we, you know, I just feel like all of the, every year, AI seems like it's kind of on the ceiling for me. And I feel like the ceilings are finally getting lower, and lower, and lower every year, we're okay, there are more as I was mentioning, API's that we've been using, and other tools that kind of have it built into it. And it's just, it's a good feeling, it makes it sound like the rooms are getting smaller, and we're gonna get squeezed like Indiana Jones, but in reality of that metaphor, is that I do feel like these tools are getting more and more attainable. And for even, you know, the, it doesn't seem like it's for Fortune 500 companies anymore. It's not only for them, you know, obviously, they will have the easiest time, you know, gaining access to these things. But I just feel like with relation to automation, everything's kind of getting democratized. And I always thought, and I do think after, especially after reading your book, it's like, I feel like automation. RPA is a fantastic way for companies to dive into AI. It's kind of hard and esoteric to me to understand and go into a logistics company and say, Hey, you guys should use AI and machine learning. And it's like, the first step is like high tier point invoice processing. Automation seems like a great gateway drug to get into this, you know, crazy world of AI. But that's kind of the thing, after reading your book I got really jazzed up about it's like, you know, we kind of don't sell or pitch the AI options, we kind of keep that a little bit under the cuff, because it's, it seems like a marketing buzzword, and we're pitching something that we're not going to do, but we are actually going to use it. And what's actually running behind the scenes is, in fact, a form of artificial intelligence. So I don't know it was it was a great read. I really enjoyed the book, it fired me up about doubling down on some of these techniques. And yeah, I just, I really enjoyed it. And I'm glad you put all that work into write it.
Tom Taulli 49:08
Thanks very much. Yeah, it's always a lot of work. But it's always satisfying in the end, and you learn a lot, a lot that I had the advantage of talking to a lot of smart people in companies and what they're doing and, you know, and you know, in this whole category, so that that helped a big deal too.
Brent Sanders 49:25
Fantastic. Mark, any other final thoughts you might have?
Mark Percival 49:30
No, this was great. I mean, I think, you know, anybody who's out there who's looking to kind of I think both these books are interesting, artificial intelligence basics. If you're kind of starting at New, I think implementing AI if you are kind of in this position where you're looking to actually start the project. I think this is really helpful. I think the team building alone chapter was really useful. And then yeah, dives into the technical aspects in a way that's still fairly approachable. So you don't have to be a super expert, but it really kind of gives you these pitfalls and things to watch out for. are very helpful. I think Brent and I will do our best to implement these as well as possible on our own practice. We'll see how it works. But yeah, thanks, Tom. Thanks for joining us. This was great.
Tom Taulli 50:12
Thank you. This was awesome, really appreciate it.