Agile Podcast: AI Live Unbiased

Ep. 2

AI Podcast Ep. 2: AI Digital Transformation with Dr. Jerry

AI Podcast

Episode Description

Dr. Jerry Smith welcomes you to another episode of AI Live and Unbiased to explore the breadth and depth of Artificial Intelligence and to encourage you to change the world, not just observe it!

In today’s episode, Dr. Jerry is going to talk about AI digital transformation, a process that has evolved into some capabilities that are believed to be transformational. Dr. Jerry is going through the story circle, outlining the reasons why digital transformation is important and its several components.

Key Takeaways

  • Challenges in traditional digital transformation:
    • Transforming the company digitally (which is not the same as digitally transforming you)
    • Digital transformation is a continuous adapting process
  • First problems that all organizations have:
    • Organizations collect information and store it without seeing the immediate value in storing data
    • What can data tell me? Does any of this date tell me about the behavior or the customer?
  • The cognitive phase is about the human cognition applied to data; Dr. Jerry explains how it is done through our five senses:
    • Vision
    • Voice (we can tell how someone is feeling by their different tones)
    • Language processing
    • Sentiment analysis
    • Cognitive search on decision-making
  • Causality: All data is interesting, but it is not all important
    • Certain kinds of data are causal to certain kinds of problems
    • Find what is causal to your business concern
  • Levers of change:
    • If you can find a way to change a variable to go from A to B then you are going to change the output
    • We move from a passive understanding of the world (collecting data) to a deep insight phase where we can now change the variables and learn deeply
  • A digital surrogate is a predictor with causal inputs, it is a digital representation of the entity that is important to us
    • We can predict how the world is going to change
  • The Optimization Stage:
    • Evolutionary computing is an optimization technique that allows us to seek in the complete search base in a very time-friendly way, and find the optimal  result
    • We apply evolutionary computing to figure out how to best treat inputs to achieve an optimal outcome
  • The Implementation Stage:
    • We take these inputs and transition them into real-life programs in the real world
    • If you apply this program, you will create change; it is unavoidable

Transcript [This transcript is auto-generated and may not be completely accurate in its depiction of the English language or rules of grammar.]

Intro: [00:04] You’re listening to AI Live and Unbiased. The podcast where we explore how to apply artificial intelligence and the impact it has on enterprises and society. Now, here’s your host, Dr. Jerry Smith. Dr. Jerry Smith: [00:19] Welcome to AgileThought, Live and Unbiased. This is your host, Dr. Jerry Smith. It’s a pleasure to be with you today. Today is going to be a special day. We’re going to talk in general about our AI digital transformation. It is a process that has evolved over time into some capabilities that we believe will be transformational in the industry. At least I believe it’ll be transformation. If you haven’t followed me before, please go on to LinkedIn. You’ll see a lot of our writing specifically in terms of how we apply this into everyday problems, whether it’s healthcare or retail or professional services like audit and materiality. Lots of content on LinkedIn. We’ll leave you a link below on where you can find that stuff. But for today, what I’d like to do is just go through the story circle, right? Go through why this is important and what the components are. And then over the next series of podcasts, we’ll get into each one or combinations of one individually, so you can get the details behind it. So for now, if you’re interested in this journey, stick with us for the next few minutes, if you’re not interested, swipe left on the box and go on to the next podcast. But if you stick with this, I promise, you know, we’ll give you something of interest. And, maybe even at the end there something, a little secret tip or two on how you can get this stuff going in your organization. All right, with that said. Again, Dr. Jerry Smith, it’s great to be here today. One of the challenges we see in traditional digital transformation activities that is transforming a company digitally, which is not the same as digitally transforming you, right? You think about the two differences digitally transforming you is just moving to a cloud, right? Microsoft, AWS, Google, getting you to consume cloud-based storage, or cloud-based compute super important, right? If you’re not applying those capabilities to your organization, you’re missing out on a lot of things, but you just doing that alone is not going to make you different in the world, right? It’s not going to digitally transform you. Your behaviors are going to be the same. 69% of CIOs, CEOs, CFOs, and the CEEIO over all believes that they have not seen a return on their investment from this activity, right? When they spend millions of dollars into this world and then they turn around at the end of the year and say, Hey, what did we get for it? Did our revenue grow? Did our margins improve? Were our customers happier? They can’t go back to that investment and contribute it to the success of any program they have. Let alone actually showing improvements in those areas. We believe one of the fundamental reasons why that is happening is because they’re dealing with it in what we would call in the engineering world in terms of an open loop system, right? I study something, I propose something I implement, I go away. What you’re going to see here. And the reason why we have this in a circle, the way it is is this is a closed loop process. By the way, little technical thing here, everything on the diagram you see here has meaning, right? This is not a marketing slide. I actually use this to help people who are having problems diagnose their problems. So this is a diagnostic tool that we have out there. So everything has meaning on this slide. All the meanings have reasons why they’re there. We’ll get into that over time. But right now the very first thing is this continuous adaptive process, which is probably one of the most complex activities you’ll do in terms of DevOps, machine learning Ops, AI Ops that are out there. Again, another episode. So for right now, I just want you to observe that it’s a continuous process. We start with the world, at the very bottom, right? In that world is the real world, right? It’s your hospital where you have patients in under healthcare, heart patients that are struggling to have their heart work. And so you’re wondering, Hey why are my patients not recovering as well as it should be. How do I help my patients not have a heart attack and how do I help them recover from a heart attack, right? The whys and the hows of any executive kind of conversation. You’re in retail, classic problems, right? In retail, you’re sitting there saying my machines are failing. I’m sorry. In the retail area, a better example is I’ve got these products and service our products on the shelf. Why aren’t they selling? Right. We’ve heard the classic analysis of men that buy beer, buy diapers at the same time while that’s not technically true in the sense that that studied actually didn’t quite happen that way. It is important to understand when people buy something, why aren’t they buying another? Or how do we help them buy another? So that’s in retail. When we think about manufacturing is just right for this sort of closed loop system centered around why are my machines failing? It’s just not important to understand why they’re failing. You have to understand, can you predict in the future? And then what do you do about that? So that’s this closed loop process. So we start with the real world. And in that real world sense, we have to instrument that world and collect information, right? And when we collect that information, we store it in our data stores. Now, here in lies, one of the very first problems that most organizations have, and that is they’ve collected information, right? They’ve gone out and they’ve logged stuff. They’ve logged both the spinning platers on their disc, old school, right? They’ve logged the database interactions in logs. They’ve logged the interactions with people on a web. They’ve done point of sale solutions, right? They’ve taken doctor’s notes and transcribed it through a scribe process. So they have digital content. There. All that stuff goes into storage. CIOs today are and CFOs today are going, that’s a cost, I don’t see immediate value from storing data, right? There is no value in storing data. There’s a regulatory responsibility, there’s a transactional point of sales responsibility, there’s an operational responsibility, but the value intrinsic in storing data, there is none. And so why we store it is for these follow on activities. We’re going to get into data later on, but for now we have that data from the real world, right? We now move into a very important phase, a next phase in this, which is to say, now that I have that data, what can it tell me? Right? If I have doctor’s notes, you know, doctor examines you, he takes notes, he has a scribe in the room. That’s a pre-medical student in the room and he is taking notes and he is popping all that stuff down. He’s taking your family history, your case history, your current state history. And he’s noting things in there. Is there anything in those notes that can tell us something intrinsically about the behavior? Right? We go take a look at a retail store and we see people who buy product A product B and product C, and they check out, right? Is there any behavioral insights in the nature of the products that we buy, the timing that we buy them in and the amount that we’re willing to pay for them. We’re in manufacturing, right? Manufacturing, we have a machine and we brought all the logs and we see a power surge one day and a machine failure the next day, is there any behavioral insights we can get from that data? So this next phase, the cognitive phase of work is all about the human cognition applied to data, right? And what are the three principle out of the five that are in there? we have vision, right? We’ve all seen it, right? You’ve seen the boxes centered around objects that say, Hey, this is an apple, this is an orange, right? So we can tell the difference visually by applying human cognition to be able to discern the difference between an apple and orange, between a red light to stop and a green light to go, right? We also have voice, right? Can we do voice to text, right. Is there a way to take my voice, my deep voice up close to this microphone speaking, slow and learn something from what I was doing right there, or my happy voice. Hey, it’s really good to be here today. Right? So those subtle cues that come from the vocalization of thoughts have deep meaning in them. Right? We all know that, right. We can tell when somebody’s sad, by the way they speak until when somebody’s happy, by the way they speak. Same words, different tones, different meanings. So, and then the last area is in text, right? The the language portion, natural language processing, which is taking voice to text and then text to meaning. Is there something in the words we have a classic example I’d love to use in understanding the behaviors that went into words, right? Sentiment analysis, who hasn’t heard of that these days. But sentiment analysis fails fundamentally because it just looks at the characteristic belief of the emotion behind a word, right? Here’s a good example that was written in an analysis of a restaurant and the person wrote, this is a great restaurant, if you like cold food. Think about that. It’s a great restaurant. If you like cold food. Do a sentiment analysis on that, for all our data science friends out there. What you’re going to come up with is probably a positive ranking, certainly a positive ranking. Thank you very much. Or this is a great restaurant, that’s positive and then cold food, right? Which is just if you like cold food, this is a great restaurant, but you and I know what I was saying, right? This is an awful restaurant because they serve their food cold, right? We call that satire, right? In that space. So the cognitive session in here is to derive the deep meaning the behavioral insights from the content that we collected in data, whether it’s visual, auditory, textual, we don’t do a factory yet smells, but that’s coming down specifically in the world of pharmaceuticals. That world is coming in that space. There’s a couple other cognitive services we’ll get into later on, not today, but I just want you to focus in things like, for example, cognitive search, right? And decision making. Those will be separate things we’ll talk about later on. But for today, I just want you to sort of think I have data. Can I make any human sense out of it? Right? Is there meaning in those doctors notes, is there meaning in those logs, is there meaning in that voice right? So we do that work. Now that is the first process where we are digitally transforming the way we think, right? We’re just not relying upon the content we have. We’re now saying I am looking for behavior in that content. And we store that back in where? Store that back into data. So our data grows. When our data grows, our CFOs going to go, why am I spending more on data and process? And we can now say, because we actually derive deep meaning from that, that we can use every day. We can actually go in and look at the, the behavioral insights that we’ve gotten. And we can just share them with our marketing team. We can share them with our sales team. We can share them with our operations. Native in and of itself, that has value. But we continue the journey, right? We’re not done with the journey yet. We’re going to continue up with this causally based predictive predictions in here. And the very first word I’d like us to focus on is that causality piece, which is. All data is interesting, but it’s not all important. Right? We know that there are certain kinds of data that are causal to certain kinds of problems. Right? If you think about the refrigerator failing again in a large retail shop, right? Was that power search the day before a causal contributor to that failing? Or was age a causal contributor it? You think about the heart patient who just had a major heart failure. What did it have to do with his demographics? The kind of the genetic predisposition he has in that space? Was it because of his lifestyle choices of smoking and drinking excessively? Or was it just life, right? Certain things we still can’t predict. Not all things have noble causes, but all things have causes that could be noble down the road. So we spend our first bit of time. Once we get data, once we understand it’s cognitive characteristics, it’s behavioral characteristics. We start to ask the question of this data. What is causal to our business concern? Right? Our business concern is patient health. Our business concern is operational equipment. Our business concern is selling more products and services. That causal characteristic is the second most important thing that’s going to come out of this process, right? The first being the behaviors, the second being, you now have a map. And by the way, when we get into the causal based predictive world, we’ll show you some actual causality maps, so you can do the search yourself, look up causal maps, look up, Judea Pearl, folks like that. You’ll see some really brilliant work that’ll hopefully blow your mind, right? Cause you to think differently. So now we have the causality, right? We know the things that are important to us. And why is causality important? There are levers of change, levers of change. And what we mean by levers of change is because it’s causal, If I can find a way to force that variable to go from state A to B, to go from one condition to another, I know that I will change the output. And there comes one of our first observable statements. We are here to change the world, not observe it. Our job should not be one based upon just writing reports and having other human beings look at them and then trying to figure out what, what can be done our job in this space, in the digital transformation space, not the transforming you digitally, not moving you to the cloud, but actually making you do things differently that contribute to your outcome is a deep understanding that causality. And now that we know that these specific variables, if I change them, I will affect an outcome. I am now in a position. So we move from this first passive understanding of our world, which is deep insights, right? Grabbing data, understanding it cognitively, looking at causality. I’m now in that deep insights phase, I am now transitioning to the other side. I’m now moving from passive world to, Hey, if I know the variables that is going to change the world, how do I change them? Now, you’re actually interested in moving to the active side of the curve, the side of the curve where we go from deep insights to deep learning, right? So then, that deep learning for us, the very first thing we do is we build digital surrogates, right? A lot of folks out there build digital surrogates. These are predictors, right? These are logistic regression algorithms. These are your trees. These are your, we got to say it today, because everybody’s doing it right. Your deep neural networks, your convolutional neural networks, your temporal neural networks, your hybrids like your spatial temporal networks that combine both a spatial understanding. This is an apple and an orange versus this is a time understanding this apple is this apple is not ready to be picked and this apple is ready to be picked, same object, different time space. They combine them together to form a spatial temporal understanding, really kind of cool part of that space. So we build those digital surrogates, but we build those predictors, I should say. They’re digital surrogates for a specific reason. Again, every word in this chart has meaning. I could have easily said, we just build predictors. A digital surrogate is by definition a predictor with causal inputs, right? So where do we get the causals elements? Previous stage. So we know the things that change the outcome now, what are we doing? Anytime we are trying to change the world, well, we have to predict where that world’s going to be, either in a moment or a day or a week or month or a year. We have to predict something down there. And that’s what a lot of managers do in life, right? They look at reports that are given to them and they go, I think next week I’m going to be out of stock on this product, I think in a month or two given your lifestyle, I think you’re going to have a heart event. Right? I think given the fact that, you know what, several years ago I saw a power surge in the same area and we lost a couple of our machines. We may want to think about, you know, replacing a couple of these older machines in there. So we actually had human beings do this work based upon the passive nature of our reports. So what do we do? We take those causal predictors, those causal elements of our world, and we shove them into the inputs of our predictor model, our deep neural system. And we give it an output and we say heart attack, machine failure, product purchase. And what do we produce? We produce a digital surrogate and that digital surrogate is a digital representation of the entity that is important to us. Guess what we’ve done here? We’ve moved from a passive to an active world. We’ve moved from a world of transforming you digitally to the world where you’re digitally transformed. Now, I don’t have to have my executives try to figure out from rows and rows of Excel spreadsheets from previous history, what tomorrow will look like? Why? because we can predict it ourselves. It’s hard, but we can do it. More importantly, because the inputs are the causal factors, what else can we do here? Well, we can play the what if game? What if this particular characteristic of a person would change? What if he lost weight? What if he stopped smoking? What if he limited his drinking? Right? What if we can apply this sort of pharmaceutical to him, right? Or nutraceutical? What if he can provide him with medicine or nutrients? How would that affect his health condition? What about the case with the retail service? What if we put product A next to product B, right? What if we reduce the price of a product? would people buy them more? We can now do what ifs, right. And that is super important because now we actually can predict tomorrow, not just observe it, but if we can predict it, what can we do with that digital surrogate? We can optimize it. And that’s that next phase that optimization, that’s the prescriptive side of the house. Rather than have a human being go through the complete list of elements in that space, we have a machine do it. Now think about this, if I have three inputs to a box with an output and each input can change two different ways. There’s roughly eight states that that box can go through, right? That’s pretty easy for a person to deal, right. He starts off with input A and changes it from a one to a zero and goes to B and changes it from a one to zero and goes to C and changes it from one to zero, whatever the range is, it goes through both states. You can get that done in pretty short order. What if he had 10 inputs? Two to the 10th minus one, that’s a lot of work by a person, right? There’s no chance that person’s actually going to find the optimal. Well, There’s limited chance that that person will find an optimal solution to that. So what do we do in that particular case? What do we do when there’s not just a fidelity of two in our inputs, but it’s a dynamic range. It’s a linear range. It’s a numerical range. What do we do when there’s a lot of inputs in it? Well, we use something called evolutionary computing. It’s an optimization technique. And what it does is this, it allows us to search the complete search space in a very time friendly way. Think of it as a functional quantum computer in this case. And that is by having that two to 10th space, those 10 elements on the inputs and an output, we can actually use evolutionary computing to search that space and find the optimal resolve result. That’s super important. Now what didn’t I say here? I didn’t say oh, in this optimized based prescription space, this house space, we’re going to build a new model. There’s only one model, it’s actually technically only two models that we’re building here. The first model being the causal one and the second one being the digital surrogate, but we’re not building a prescriptive model. And that’s a fallacy today that you see a lot in the machine learning world. They think they have to build a new model that’s prescriptive. We don’t do that. You’ve already built a vision of tomorrow based upon causal inputs of today, we apply evolutionary computing to optimize that, to figure out how you have to set those inputs to achieve your ultimate outcome. So we do that. And what do we get? We get a state. We say, Hey, if you can drive these causal inputs of this date, you’ll likely achieve this outcome. Well, that is the stepping stone back to reality, right? So as we move from that optimized state, we then sit down, who do we sit down with? We sit down with marketing, Hey, if I want to get a person to think this way about a product, what can we do in marketing? What about sales people? If I want to produce this behavioral care input characteristic in? What is the sales program to do that operations, right? predictive doctoring, right? If you want a doctor can now sit down and say, well, you know what? Here’s things I can do. I can prescribe this medicine for him every time I see this case. What about therapeutic mental therapy, right? Psychological issues. We can help him discover these components in order to relieve these pressures. So the good news is that we take those inputs and we transition them back to real live programs in the real world. And what do we do with those? We go implement them. We take those programs now. And we say, okay, this is our best educated guess at a process that will achieve the state that we identified to be causal in the output. And we go apply them. And what do we do? We wait for a day, week, month, year, right? We wait for those systems to take hold, those process to be implemented, the behaviors to change in people and what happens? Their behaviors change, we collect the new data, we go through the cognitive processes, we check to see that the causality is still there, we build the digital surrogates, we optimize it and we update our programs, field implement them. Wash, rinse, and repeat. By doing this, you will change tomorrow. It’s guaranteed, right? There is no way tomorrow will not change, if you apply this problem or process to this, right? That’s the neat thing about a closed loop system. Now, you could argue, well, it didn’t happen as fast as I wanted. Got it. It will change. You can argue another point, which is a classic one. I don’t have all my data yet. In a different session we’re going to talk about that one in detail, because that is such a common one “Well, my data Dr. Jerry is man, it’s just not that good of shape, right?” It’s not that good of shape, we’ll wait. As it turns out, it doesn’t matter, why? Because this system is closed loop and it’s optimized. It’s two things come out of this process when we apply. And this will blow your mind in a future episode, which is number one, will identify what data actually does need to be change. So don’t worry about trying to upgrade all your data. Some data you don’t even need to worry about right. Causal map. But the other piece is you can start with just random data, just start with random data. And we have shown this to be true. Just start with the fields you think are important, stick random numbers in there, create your digital surrogate, go implement it, and then watch what happens and collect new data. Now, what’ll end up happening in the beginning is you get a negative response, right? Your performance curve drops. Why it was random, right? I guessed that stuff by the way. Most human beings in the real world guess at what they’re doing. So you wonder why it takes a long time to change your company, you wonder why things fail often in the beginning. And then projects are abandoned. It’s because we guess, right? Human beings are really bad at one predicting tomorrow, and then once we predict tomorrow, optimizing our organization around achieving that outcome. And when things go wrong, what happens? Programs are stopped. People are fired, new people are brought in a different direction. So we never give the time. So we can actually show, even if you start at a random process, you’ll eventually dip and self correct up, your data will change, the behaviors will be identified. You’ll identify which ones are really causal in the process. You’ll optimize. You’ll go through them. Now we obviously don’t want to do this for healthcare, right? You don’t want to start by saying, Hey, I don’t know anything about heart failure. So I’m just going to randomize that. But in terms of sales, that is a different conversation we’ll get into. There may be a sense where even if you don’t have data starting off with just random variables in there and comparing east coast to west coast, comparing north to south, these set of stories of those set of stories, you’ll be able to very, very quickly identify the kind of things that you need to be analyzing in your go forward process of digital transformation. So for today, I think that’s pretty good place to stop, right? We had a chance to talk briefly about the continuous adaptive circle, right? Anytime we implement a continuous adaptive circle, what happens? You’re right, e get a chance to change tomorrow, not observe it. We talked about the elements of it, which is data, cognitive causality, digital surrogates, digital surrogates, being what a predictor with causal inputs, right? An optimized prescription. What is the prescriptive optimization? We use things like evolutionary computing, particle swarm optimization. We I like personally evolutionary computing because it fixes a lot of other problems we’ll get into later on just not about optimizing input and output. You can hyper tune your parameters. You can actually change the structure of your deep learning or your learning system at the same time. We’ll get into that later on. And then finally you transform those things into some sort of field plan and you go out and deliver it and you sit back and, you know, have a couple of margaritas. You take a vacation and you wait for the data to change and you wash, rinse and repeat on that sort of process. The first part of the process is very passive. It’s about insights. Right? Second part of the process is about being very active, where we take action. It’s about changing the world in that particular space. I hope you found this value. I hope you see something in here that if you haven’t followed us before you haven’t seen our writings before that you say, Hey, that’s, that’s different. And I hope this gives you the courage to maybe come back and spend more time with us down the road as we delve into each of these areas. And as we delve into these areas, we’ll bring in customers and we’ll bring in other folks that we’ll talk about their problems and stuff like that. So while today was a generalized view of this around the circle, you know, tomorrow when we get into data and how we do that and cognitive stuff, how we get into that, we’ll bring in customers now and then, and we’ll talk through those things or other team members or other smart people in the industry, but for now that’s it. AI driven digital transformation. I am Dr. Jerry, your host on this journey as we go through the space of all things artificially intelligent. And I look forward to seeing you next time.

Outro:[28:39] This has been the AI Live and Unbiased podcast brought to you by AgileThought. The views, opinions and information expressed in this podcast are solely those of the hosts and the guests, and do not necessarily represent those of AgileThought. Get the show notes and other helpful tips for this episode and other episodes at agilethought.com/podcast.

Share this content

Transcript

Speakers

Dr. Jerry Smith

(Panelist), Managing Director, Global Analytics & Data Insights, AgileThought

Dr. Smith is a practicing AI & Data Scientist, Thought Leader, Innovator, Speaker, Author, and Philanthropist dedicated to advancing and transforming businesses through evolutionary computing, enterprise AI and data sciences, machine learning, and causal AI.

He’s presented at Gartner Conferences, CIO, SalesForce, DreamForce, and the World Pharma Congress, and is often who leaders turn to for help with developing practical methods for unifying insights and data to solve real problems facing today’s businesses.

Related