Human Centric Automation

Super.AI partnered with Techopian to put on Human Centric Automation, a conversation that explores the two most critical success factors in any automation program:

1. How can organizations leverage unstructured data in automation programs?
2. Why are humans the most important automation tool?

Full Transcript:

Rob Hughes [00:00:03] And we're like. So thanks, everyone. Thanks for joining human centric automation. This webinar is hosted by Techopian and along with our colleagues at super.AI. I'm going to hand over to our editor in chief, Mr. Michael Baxter, and he's going to take it from here and introduce everybody. But thanks very much. And we'll we'll kick it off from here. Michael, where are you go. 

Michael Baxter [00:00:37] Thank you very much, Rob. Well, good morning and good afternoon and good evening to everybody wherever you are in the world. So my name is Mike, says Rob just set the Ed ticker. And today, I'm joined by Brad Cordova over a man of three from Super.AI And my business partner, Robert Hughes. And we will be finding out why humans are getting the secrets. We're going to be talking a little bit about the wisdom of crowds and the wisdom of crowds when those crowds involve an entrance. And we'll be talking about the combination of humans and technology and why it is better than one or the other. You know, it's better to have humans and technology than just humans or just technology. And how this combined with processing power, how this combined processing power can reduce risks and increase potential breakthroughs. And also, we told something about string theory as well. So, Brad, maybe you could take this opportunity to introduce yourself and tell us all about string theory while you're at it. 

Brad Cordova [00:01:46] Yeah, that may take another few months to talk through, but yeah, I actually started as a as a physicist and mathematician doing string theory. I figured out it's it's kind of a fraud. I don't actually think that's what explains the universe. It's very elegant mathematics and the most brilliant people in the entire planet are working on it. And so there's a lot of great stuff going on, but I wanted to do something more practical. So I went to MIT to do my Ph.D. in artificial intelligence, and my research involves amalgamating symbolic AI and machine learning. And then during my Ph.D., I founded a company called True Motion. We built the back end for some of the biggest mobility companies around the world, like Lyft and and Progressive Insurance. We scaled that to a unicorn and recently merged with another company. And then I founded Dupri AI and a lot of the confluence ideas that came from my research at MIT in True Motion and were kind of the genesis of Super AI. So excited to chat with you guys today. 

Michael Baxter [00:03:01] Thank you very much, Brad and Manish. Could you say a little bit about your good self? 

Manish Rai [00:03:06] Yeah, my self manager. I have been in Silicon Valley for over 20 years in various product and marketing roads. The last five years I've been living in the automation world to work closely with your CEO Rob at automation anywhere where, among other things, I launched two IDP product IQ bot, which had a really good run and it's become one of the leaders in that space. And I'm really the reason I'm here at Super High is because I saw in these last five years the massive amount of unstructured data that enterprises are sitting, sitting on. I mean, I was just looking at a stop there in 2020. 200 zettabytes of data is expected to be created and we are creating a ton of data. But the key is to leveraging this data to deliver better customer experience, to further drive down cost and to delight your customers. And I think that's where moving to a human centric or automation, which can can really help enterprises go and transform themselves. 

Michael Baxter [00:04:33] Thank you, Manish. So if we could just sell the need for you. 

Rob Hughes [00:04:41] That's okay. That's okay. Hi, I'm Rob Hughes. I co-founded two coping with Michael Baxter, who actually does all the work and all the writing and all the research on this stuff. And I'm lucky enough to get to speak to very small people who've done amazing things as a result of it. And prior to that, as many said, he and I worked together at a company called Automation Anywhere, and we were involved in the space. Just more structured information, I suppose, movement across businesses. And prior to that, I worked in the analyst community for around 15 years and again surrounded by smart people and just sat and listened for a while. So that's me. 

Michael Baxter [00:05:23] Okay. So first of all, shall we identify the problem or at least one of the main challenges that businesses need to solve with technology and which is the problem of unstructured data? And it's been estimated that around 80% of all data is unstructured. And right now, the way we deal with unstructured data is use those things like equipment, human beings. So human beings, they they have to manually type in the data. And that can be a problem because people get some what is it not? What is it people say, oh, that's right, they lose concentration and they make mistakes. So. Right, how big a problem is this? 

Brad Cordova [00:06:07] So, Maneesh, and you made some stats, but the problem statement is simple, as you said, is most data in the world you mentioned 80% of the data in the world is unstructured data. And I think what's even more profound is the growth rate. If you look at the growth of unstructured data, because we all have cell phones and take pictures, there's Iot devices, it's growing exponentially where the amount of structured data, data that can fit into a spreadsheet that machines can understand by default or is if you put it on the same graph, it almost looks flat, but it's it's growing linearly. And so the problem stem in simple if you most data is unstructured data but but the problem with unstructured data is you can only use it if you give it structure because most of the automation tools out there like robotic process automation, robotic desktop automation, intelligent document can only function unstructured data, some things can function in some way structured data. So you have this bifurcation of data in unusable data and usable data. And the problem is most data is unusable. And, and so so that's what we we solve a super but that's really the problem here it's a simple statement really difficult to to actually solve. 

Michael Baxter [00:07:25] I can manage. And Rob. Do you have anything to add to that? 

Manish Rai [00:07:29] So I was just looking at this stat that every day we create 2.5 quintillion bytes of data. That's just staggering. And, you know, as Brad pointed out, right, with to a camera and a video recorder in our hands in enterprises via collecting a lot more data and bringing in in all sorts of formats, audio, video, images, enterprises spend $20 billion a year on paper, in documents. Right. So these are all massive and staggering numbers out there. So what I like to call is since I live in Silicon Valley and close to a fault line, I think enterprises today are living on a data faultline and the pressure of this data is building. And and they are one massive event away from being, you know, marginalized. A startup that figures out how to make that sense of data better, to deliver a better experience to their customers or at a lower cost. And that's that's the challenge facing enterprises. So they need to be moving now. It's not sufficient for them to just make sense of structured data. People may be feeling happy that I've deployed 100 bots. I've automated my processes with RPA too, to streamline my processes. But imagine the art of the possible, what you could do if you make similar sense of unstructured data. How much of a transformational experience you can deliver to your customers? And that's the opportunity out there. 

Rob Hughes [00:09:12] Yeah. I think that's a that's a really good point. And I think when we think about automation and RPA specifically, I see it as the plumbing. You know, it's there to be able to move information from disparate systems. Basically, you know, one system doesn't talk to the other. They don't speak the same language. It has a good linkage, a good pipe between those two disparate systems. But it's all the data that's just lost in the ether that we're not learning from. And I think that's the real opportunity, is what could we really do with all of this information if we could process it faster if quantum computing takes off? Then again, Brad, you'll know more about this than me as to whether or not there's some truth in that. But I'm hoping that our processing power starts to speed up again as Moore's Law starts to slow down. And what we do with all of that data is much more important than just having the plumbing in place. The infrastructure is, and the potential to build that infrastructure around businesses is is becoming apparent and it's being built. It's just how to how to access all of the hidden gems that are still in them hills, as it were, when it comes to, you know, nuggets of data. Gold, I suppose. So that's that's my contribution, Michael. 

Michael Baxter [00:10:28] Okay. So there's a secret there's a little bit of a secret, which I'm going to let you all into today, whisper it that that that air has a secret as it secrets and it's human beings. So, Brad, why are humans always dirty little secret? 

Brad Cordova [00:10:49] Yeah. So there is there's fundamental reasons, sociological reasons. But I think fundamentally, if you look at any production today that's actually doing any value, it was trained by a human. Even all of Google's search algorithms, there's a lot of marketing that we use, these massive deep neural networks. But behind the scenes, if you look at any A.I. system, it's humans teaching the machines. And the reason is simple. We as humans, where we're engineered or evolved, whatever your is to to structure unstructured data. So so our eyes, our retinas, they have rods and cones and they take incoming photons and epithelial cells and they convert it into electrical signals which our brains can operate on. So they structure this unstructured data or our ears. They have Cochlear, which essentially perform a 48 transform on sound waves and convert them into electrical signals. So we are the ideal creature so far to structure unstructured data because every, every sense organ we have converts into structured data. And so the the challenge is, is how do you do something similar for AI? How do you create these quote unquote sense organs to convert unstructured data into structured data so that then you can use them in effective way? So that's the core of why people use humans. And also the human brain is the most one of the most impressive things we've ever known about in the universe. And so the challenges is how do you take all these functionalities and be able to allow artificial intelligence to do that? And we we have a solution to that as well. I am happy to have you to give you a little demo of your, if you want. 

Michael Baxter [00:12:41] This place. 

Brad Cordova [00:12:43] Okay. Great. So I will I will share my screen. And I. 

Michael Baxter [00:12:50] Will. 

Brad Cordova [00:12:51] Do a demo on Undocumented since that's what we were talking about. I assume you can see this. 

Michael Baxter [00:12:59] Yes. Just people watching this. If you could maximize your screen, you'll probably see that much more clearly. So that's the obvious. 

Brad Cordova [00:13:10] So just as a bit of context, what we did at Superior is we built a protocol. And what's a protocol? Well, a protocol is a language or a set of rules that different entities that use a protocol need to follow. So, for example, the Internet has a protocol, HTTP, UDP, all these things that allow computers to talk to each other, which make the Internet. So we built a protocol which allows us to talk together. And so as a result, we can build generic applications and one of these is processing documents. And so every, every document processing application starts with the customization phase. And so we have pre-trained filled detectors and pre-trained document detectors out of the box that you can use. So for example, if this data program classified the document as a bill of lading will then it would automatically extract things that company agent numbers, signature, etc.. But if if it was a warrant, we work with some government agencies, it would extract the warrant number. There's many of these. If it's a claim form, it would if it's an invoice, it would extract things like the invoice ID. In this case, if it works correctly, it would classify it as an ID and extract things like the country. And of course, you can you can delete these or add fields or add completely new fields. And so the first thing here, which is really important, is we don't just take a machine learning model and try to make the best quality. We we take A.I. models and build applications. And what's important is to optimize for this triangle of of quality, cost and speed, this triad. So, so maybe in order to get a business ROI, which in the end is most important, well, I need to hit a cost. Let's say that's that's $0.09 or even less. Or maybe I need it to be a certain speed, or maybe I need it for some reason to be really, really high quality. And so each of these selections will allow you to choose different workers to work on this. And, and so today, in the world of AI, if you if you want to train and able, you need to go hire a bunch of humans to label it and then train in the AI. These are these are disparate processes, but it doesn't make sense because a human's taking some input to an output and AI's taking an input to an output. Even bots like regular expressions are taking inputs, outputs. And so we developed this mathematical model that they all work together in the same place and I'll show you how that works. So let me just process this idea. So the first step here is that you take the input from some database with this RPA system, your on prem database and API, and you input it into what we call this data program. And what it data program is, is you can think of it like an assembly line. So Henry Ford taught us, hey, don't build the car. And one step have have one simple task where you put a wheel in the car, have another simple task where you put a mirror on the car. And this turns out to be really powerful. And so we do a same thing. We decompose a complex task into simpler tasks. For example, classification detection, transcription and fuzzy matching, which allows you to match dates to like, okay, this day a birthday is an expiration date. And so now let's, let's execute this workflow. So the first step is a classification. And what happens first is you take the simple task of classification and you route it to different workers. In this case, it's it's routing to a human, a BPO, an AI and a software bot. And then what happens is we wait for these three workers to process the output. And of course, the AI comes back pretty quickly. Next, the human does and then and then the software. And in this case, they all said, hey, this is an idea, which is correct, but they don't all need to give the same answer. And oftentimes, let's say you're using 20 different machine learning models. They may all give different answers. And so what we have is a combining to combine all these outputs of the workers into a single output. So that you can and it does it in a smart way that if you trust the worker more, it'll wear higher. But at this stage you have a single output, which in this case, unsurprisingly, is an idea for the classification. But we another value here is that you need to guarantee the quality. It's 2022. We expect clear otherwise from our applications, not just science projects that we write on our website. Oh, when a company. And so there's over 150 different quality assurance checks that we look at the quality of the data, for example, maybe you upload it really noisy data. We look at the quality of the workers, the task we have an anomaly detector, so maybe a human answer too quickly or too slowly doesn't mean they're wrong, but that that's surprising if they just maybe they're just clicking through it or they're confused. So it took them a long time. And then the final step of this task is we train an eye on this classification task so that eventually you don't need to use humans at all. You you can just use A.I.. So I will fast go fast through some of these. Now there's detection, transcription, fuzzy matching and so that that finish. But if we if we look just on this task, it started out as around 13% automated, meaning 13% of the tasks were handled by AI or software. And then each, each round of of training here increased automation. So now even just after this single data point, it's already 60% automated and similar with the the cost because we're going to route to to more now to A.I. in the future in the time that decreases but also the quality increases. And so this was a brief look at the platform. I don't know, there's any questions, but that's that's roughly how this this platform works and how we implement the protocol. 

Rob Hughes [00:19:25] That's I mean, that's amazing. And I think, you know, it's I keep thinking of the term travel adapter, right, where you need electricity wherever you go and you need good quality electricity wherever you go. And this allows you almost to plug in electricity and allows you to plug in the best electricity available at the time because. This world is moving so quickly with an asset I'm talking about. I'm talking about our technology world is moving very, very quickly. And small companies just don't get the opportunity to go to market either. A lot of technology companies with great ideas and great tech just disappear because they're not go to market people or they don't have those skills. That's almost allows them a way to get to market without having that go to market function. And I suppose that the risks from a purchasing point of view, because their relationship, I take it, is with super A.I., it's not with the end user. So it's almost that platform layer in between that allows you to buy best in breed technology, but doesn't have the risk of saying, I don't know if this guy, this company will be around or this tech will be around in two years time. And I've implemented it because you're using it when you need it and as a service rather than a, you know, an implemented tool. So really it really interesting. And you mentioned 150 points that you have to check on these companies or to check on the tech. Does that happen? Is that a standard process? Do you work with companies like did you go out and look for them or do they how do they apply to you and get in touch? How do you find this this new tech to solve these these complex problems? 

Brad Cordova [00:21:04] Yeah. So I think just to underscore what you mentioned, we call this unified A.I. platform. And and we're seeing this trend a lot. So, I mean, even just with our with our iPhones, all you need to do is invest in an iPhone in it. And there's new apps developed every day. There's there's so many different apps, whereas before we'd have to go invest in a calculator and now I just have a calculator on my phone or a camera. I just use the camera on my phone. I don't need to buy a separate camera. So or if you see like even the Tesla cars, as long as you have the hardware for for self-driving, eventually they release updates every week or every month. And so the power is also now with this unified air platform, you can have any application running on this. All you need to do is buy the iPhone and now you can get any application running on there. So it's kind of the iPhone. And so you before you needed to buy document extraction from a certain company or reduction reclassification. Now this can all happen in a single place and then then it's future proof because it's easy just to make an update or whatever. And then as you saw, you can use any worker, so you can use AI from from Amazon or Google or Microsoft or open source any human. And so as we've seen in the fields, if you walk into a vendor which is using their own proprietary AI model, will probably tomorrow or the next month, there's going to be someone smarter who came along, who built a better model. And we're seeing this all the time. And so that's why we built it this way. So you can plug in any model you want at any time. And so you're always using the best. And a consequence of that is you really future proofing yourself? 

Rob Hughes [00:22:46] Absolutely. And with regards to the people, I suppose companies can use their own people to to plug in too, because it's a task with internally. But I suppose there's almost an opportunity. I'm kind of thinking outside the box here a little bit, but, you know, there's almost an opportunity to just crowdsource skills from areas that, you know, high, high level skills to be able to to to contribute and just increase the accuracy of the AI. And the more you do that, the better the AI gets, the more it learns as a result. And yeah, it's a it seems to be very fulfilling if, if, you know, if for both users and buyers, if know if we can get that right. Really interesting combination. 

Brad Cordova [00:23:28] Yeah, that's exactly what we do. We have a hierarchy of of of humans. Like maybe there's a lawyer or a doctor we have to use every once in a while, but they're going to be expensive. So that's what the router does, is if the task is simple, well, either out to an AI or software or let's say a cheaper human. And if we need something that's an expert or you can use a single expert, you can use many experts and then combine them into a single output, which gives you the highest accuracy. But of course, it's the most expensive. 

Manish Rai [00:23:58] Yeah. I mean I as I'm thinking broad and you've been talking and Rob you have been talking is you know, when you look back in the journey five years ago when a lot of IDPs started, we were looking at one piece of unstructured data, which was documents, even though it's massive, 40 billion opportunity, it's like right below the tip of the iceberg, right. You are handling and mostly semi-structured, unstructured documents for most parts, which is a very small piece of the problem. And what the solutions ended up doing is having either the proprietary oceans underneath or thick set. Of course, here's the human part of it was an afterthought, right? So you have to source the humans from in-house, put them, and then you have to tune the systems and most will say, hey, you can define the confidence level at the field level, but a business user wants outcome. They don't want to. What does 80% confidence level at a field level mean to a business user? Nothing is said what is going to be the outcome for me and nobody does that. And what happens if the human is out on vacation for a week? Like, are you going to have escalation routing? What if I don't have the skill in-house? So there was a massive problem over there. We have solved this by building sort of think about a mechanical turk of users, but focused on business by not just generic any skill but these people who can solve look at the business data and help you and could it be a specialized lawyer and we'll help you source these. And then we built all the gamification, all of the things to engage the human workforce, as Brad pointed out to me the other day, that humans need to be motivated, they need encouragement, they take vacations, they they get distracted. And so we we make errors also because we are not designed to do repetitive work every day. We like creativity and that. And so we've built all those mechanisms to measure the human quality, keep them engaged, source them to solve the human part of the problem. And, and, and then that's how we can guarantee quality and make setup seamless. And a lot of easy has gone into how you combined results from multiple AI and human to guarantee that quality. It's not a trivial problem and others will say, hey, yeah, we'll just combine and we'll work on on those model. Simple simple you know, average out the performance but but that's not doing justice to the work and we build all those underpinnings and create an open platform where we can be the best model, best humans and guarantee quality and give you a business ready enterprise scale application and days. It's a promise a lot of people make, but I think we are delivering it today. That's makes it very exciting. 

Rob Hughes [00:27:04] Yeah, I think I think the platform approach that you've taken is unique. I haven't seen that before, there's no doubt about that. And you're right about the human component of being missed out, right? We kind of thought, oh, we're going to automate all of this rubbish work away from everybody. Now, I'm still a firm believer that there's a lot of rubbish work that we do, and I think a lot of people listening in would agree, right? We do loads of rubbish work that doesn't add any value is just getting data. All we're doing is winning the night around. So and we've got a problem because we've got massive skill shortages, right? And we have to give people the tools to do the best job that they can rather than doing the best job that they can with the existing tools set and more work and less people to do it. So we have to use technology to do that. And I think the approach of combining the two together is really powerful. You know, reminds me, Garry Kasparov said, you know, the only reason we around to complain about technology is because of technology, you know, and, you know, he he when he lost to Big Blue back in the day, he started studying, you know, why did I lose? And he said that technology is just amazing, that computer, you know, you've got more power on your iPhone. We've all heard that story. But he ran a few chess tournaments afterwards where it was combined human and technology working together. And it was a group of he said they weren't very good. But, you know, Garry Kasparov is not very good at choices. It's probably brilliant for most of us, but it was a group of three guys. We just used the average computers of the average computing power that you get from a a chess computer for free on the Internet. And they won the tournament because it was three minds versus three artificial minds combined actually produces better results. So yeah, there's a lot of anecdotes to say that this is a good thing to do and it's it's the right thing to do for us. But I think one of the challenges has always been how we combine this together and how we. Take people on that journey and let them know that this is a tool. It's not replacing urine, right? It's not a human brain. And although we're all we all think it's I really I you know, we're still we're still a little bit way off before that before the Terminator said. 

Manish Rai [00:29:13] And another thing sorry, I wanted to talk about this that yeah, the biggest problem we have, the documents is invoices. And people have been tackling it for 30 plus years trying to solve it. And Utah, when I talked to most of the people getting to 60% automation is very easy because there's some high volume invoices that have certain format you can configure. You can either create templates or train to do that. But what overwhelms people after that is all the new invoices that keep coming in, new formats that you have to keep tuning the system to accommodate for these changes. And EIA is a complex beast right now. Today, if you don't design it correctly, your models drift your your your models. Actually, many times the performance goes down over time as the new formats come. And you take them to don't take into account all the variants and outliers properly in how you're building the model. So, so, so. So my point is when you have automated 60%, you still have 40%. The humans are looking at it and how you efficiently you use the humans and leverage the learnings from these humans to get to 80%, 85% becomes critical to how much you can automate. 

Rob Hughes [00:30:40] Mm. Yeah, absolutely. So back to Michael. We, we, we, we went into a bit of a hole, but it's an interesting hall and thanks very much for the for the time. I think it's really you've got a really interesting proposition going on. 

Michael Baxter [00:30:55] Okay. Well, thank you very much for that. That was very interesting. So I have a confession to make. I can't I am. And I don't think I'm unusual. I think a lot of people can't sing. And that's something that's always puzzled me, because whenever you listen to a crowd singing at a rock concert or crowd singing at a football match, they tend to sing and cheering. Although that's very strange. Not because I don't think many people can sing. And and there's a reason for that. And I think the explanation goes back to 1906 when a chap when a chap called Francis Gordon, who is a mathematician, around 5:00 with some dodgy views on eugenics, and he visited a livestock fair and there was, I guess, the weight of an ox competition at that fair. And I'm just checking here. It turns out that the average weight of people what sorry. The weight of the ox was £1,198 and something like 800 people had to go guessing it. No one got it right, but Galton took the average of all those guesses, and it turns out that the average was less than a pound away from the correct amount and it was by far and away the most accurate estimate. And that's. With Democrats. That's why football crowds appeared to sing in shame, because, well, some people might think too high. Some people might seem too low. On average, they're about what they're about. Right. I think this is a similar kind of principle applying here to what you are doing. Right. 

‍Brad Cordova [00:32:36] Yeah, definitely. So we see we see this phenomenon in a lot of different areas. The classic one is measuring the weight, but people have also used that to find submarines. The there is a scorpion submarine that that was lost and they took the greatest experts, the greatest physicists to find it. And no one can no one could find it. But what they did is when when they did a Bayesian puts theory or when they did all this be more simple. So when they just took the average of of all other estimates, it was like exactly where where the scorpion has sunk. And so this works really well under. But there's four assumptions that that it needs to work well. So there needs to be a diversity of opinion. So everyone needs to have kind of local information. They need to be independent and needs to be decentralized. And then there needs to kind of be an aggregation mechanism in trust to do that, because along with wisdom of the crowds is like madness of the crowds and groupthink. And so when you don't have these assumptions satisfied, then instead of wisdom, it turns into madness. And so there's all these stories. But to make it work, you need to a lot of the work is in making sure that that these assumptions are satisfied. But it is a very powerful thing. And also speaking about mathematicians, there was Sean d'Orsay during the French Revolution. He he formed his jury theorem because they were trying to figure out what is fair. Like if a jury says someone's guilty, is it worth to put an innocent person in jail or to let a guilty person free? And so they were they were using a lot of math to solve this. And he found a similar conclusion, injuries that if you if you have people who are independent, again, that's a similar assumption and they're above random as a number of people approaches infinity, their accuracy approaches 100% accuracy. And so we just took this idea, which appears in many different forms, and we took it to the next level and made it very practical. And we've also actually created some some theorems around it as well. 

Rob Hughes [00:35:00] So I'm just going to jump in, Michael, if you don't. That's really interesting. The jury piece and I hadn't thought of that before, that you've you've almost going to come to a conclusion of fairness at some stage. So juries that makes juries much more much fairer that it say, than a single judge with with a bias or whatever the case might be. Okay. And that, of course, returns with the more data that comes into the system, the more accurate you can become because you have more data points to be able to to reference outputs. 

‍

Brad Cordova [00:35:29] So yeah, for example, if you are going to create a new country and you had new rules of law, you can you can make the outcomes of courts way more accurate, way more accurate. But yeah, that's a whole other way to begin. 

Rob Hughes [00:35:44] To change the judicial system in a way. 

Michael Baxter [00:35:47] Which I guess is the kind of diversity argument as well. You know, the more diverse across the board, the is the wiser it is, I suppose point plays into it. I bias as well. Okay. Is that what you mean by a unified platform? 

Brad Cordova [00:36:07] It's a piece of it, but there's kind of two. So I mentioned this earlier, but there's two pieces of this. One is you can use any worker so you can use humans, I, RPA, software bots and I can come from wherever it can. You can use Google's Amazon's open source your own, you can use statistics, you can use expert systems or you can use any human, for example, so you can use BPOs, you can use your own humans. We have a crowd of humans around the world with varying skill sets, or you can use any software functions, whether it's regular expressions, RPA, bots. So that's the half of it. You can use any worker. I mean, it unifies all that. And the second is you can use any application so you can extract documents, you can redact people's faces out of videos, you can classify products and create taxonomies all in a single platform. And, and it allows you to start with a high level of automation, but then in addition, customize that to you and then then it's easier to integrate. We don't want we don't want people, even though it's really powerful and we can improve that super easily. But sometimes the practical piece of just integrating it or adopting something new is a lot of work. And so we make it also easy to integrate into your existing systems, use your existing humans, use your existing RPA or whatever you have. So that's that's what we mean by, by the unified platform. 

Michael Baxter [00:37:39] Okay. So this is a more general question. So where's automation technology going in the next few years? And if you could get your kids to pull those out maybe a little bit further beyond that, there is ideas. Right? 

Brad Cordova [00:37:59] Yeah. So. Well, and in a lot of ways, we there's automation all around us, right? Like, my like we also have different parts of our brains that you could even consider automation. So I have my neocortex, which is my higher level cognition, but then I have the limbic system, my brain, which like I don't need to think about my heart beating or like it controls my muscles or my spinal cord. So a lot of it is automated. Like there's, there's a lot of stuff in my body that feels effortless. And in a lot of ways these are kind of different computers in my head. They just have a high bandwidth of communication and we just kind of see this thing as as outside of us. But, but in the end, it's also a part of us. And and the bandwidth is the bandwidth. I can use my thumbs to connect to that and it's a lower bandwidth. And so I see the future as essentially just increasing the bandwidth between these different systems, like increasing the bandwidth between our phones, increasing the bandwidth to our eye, increasing the bandwidth to our computers. And, and and I think that's a long term view, but that's what I see as the future of automation. And we wrote a piece on where we see 2022 automation and we can go to a lot of detail there. But but high level. That's right here. 

Manish Rai [00:39:23] If you look at the automation world, we've been talking for a few years, the Holy Grail is that you have AI that's broadly observing what's happening inside the enterprise and piecing it together to find how the business processes work and how how they deviate from the norm and what instances. And that's what we call process mining or Das mining or process discovery. And, and where we want to be is that have a closed loop. We are constantly observing the business processes, finding ops opportunities to optimize and then with creating a platform that can automatically create an application to automate that business process. Today, the vision is that that application is going to be an RPA bot which handles structured data. Tomorrow I think it'll be, you know, that applications combine the best of human AI and bots to automate that business processes and really streamline your costs and, and delight your customers. And that's where it needs to move in the future, where it's a seamless process today. It requires a lot of hand-holding, a lot of work on everybody to make it possible. But I think in ten years, hopefully it'll be a lot more automated that the process of automating processes will become a lot more automated and streamline automation. 

Rob Hughes [00:41:01] Yeah, I think yeah. I think you hit the nail on the head there. I think what's going to happen is we're going to see more automation, creating automation. So understanding where, you know, where the blockages are and saying, hey, we need to, you know, there's a better way to fix this as a new way to fix this. I also see an acceptance of standardization. You know, for many years, I think companies created lots of customizable linkages and codes and whatever else to make things work. And then ERP came along and you know, you've really got the process in a box to a large degree, and that's compliance. So you get a stamp for it. The problem with that is everybody's in the same box and it doesn't. Your operations are not a competitive advantage and that you look at the, you know, the best companies in the world now, every part of their business is optimized, you know, data flows and it's very efficient. Things are automated as best they can. And I think of Amazon and, you know, they they treat the supply chain as a competitive advantage. You know, it's just and that's really all it is. Every part of that business is optimized. And as Brad pointed out, we're really we really just are extensions of our technology. Our technology is extensions of what we want to try and do and how we want to process stuff. And as we start to integrate with that more and there's more demand and and we get more comfortable with it. Right. You know, I think people are much more comfortable with AI today than they were five years ago, if you'd mentioned that. You know, I know. And when I first started looking at this, you know, that that picture of the Terminator was flying around the whole Internet saying the robots are coming in. Now, people look at it, laugh. Right. You know, they think because they know what reality is, I think they've come to terms with it. So I think that's going to happen. And yeah, I'm very excited for it because I do think we've been lumped with a lot of really rubbish work and a lot of people take the short end of the stick for that and end up doing work that we could have automated years ago and we should have automated years ago. And we I'm hoping that we uncover the next Einstein as a result of that as well, because I think there's lots of lost potential just the way that we're set up and we lose a lot of potential and, and and brainpower and and how we think about things. Michael, what do you think? Because I know you're you're very keen on the autonomous vehicles and the whole idea of, you know, an automated travel system across the world. I suppose that would be your ideal. 

Michael Baxter [00:43:36] I think that the autonomous cars take off and I believe they will probably later this decade and next. Should it be the most profound impact on the global economy that it will do to the global economy? What would cause it to the ports? Because, you know, to the charity, of course. And that's going to disrupt the most important sectors. How many factories stream? I think it's going to be a maybe. But on the on the subject of what we find to be at any point, would it help us for the next day, one story, maybe he'll be able to explain to us a bit more about string theory if we do that. But um, I, um, I think I know how government you have got to know the ring. And we had this practice, you know, my practice and I kind of feel like that is not my, you know, it's sort of like it's approach. And, and I think someone said that we are like cyborgs now. We are half we're half human, half slow. And it's that the is nature of A.I. that that will you know, we'll become much more integrated with with our associates anyway. That's going on, right? Yeah. 

Brad Cordova [00:45:08] Yeah. So I mean, yeah, I think even like I kind of mentioned my, my view on maybe I'll take a different view and and kind of start with Adam Smith about he analyzed what made countries successful and what why some countries were poor. And one of his big conclusions was, was specialization of labor. You can try to do everything yourself. You can go get water yourself, go hunting. But but if I specialize and then I trade with you, well, everybody is better off. And so when when we tamed plants and had agriculture, well, this you just look at the spike in in every measurable quality of life metric when we when we tamed animals to the farm and similar thing when we tamed electrons so that we can do electricity. It's the same thing. And so I see the taming of AI artificial intelligence as another step in. And you just see this sociologically. Like a few hundred years ago, we would all be farmers, right? Like 95% of people were farmers. There was no weekends. And now today, I think it's only like three or 4% of the economy is our farmers and now we can do higher level things. And so I think the taming and merging of every species, whether the species or software animals or plants, it allows us to be more human in a way that's special to us. And I think we as humans love being around other humans. We love having empathy, creativity and and we don't we can outsource the things that plants are good at. We can all sorts of things that A.I. is good at. And, and it allows us to all be more specialized. And and I think the kind of root word or etymology of specialize is special, which, in my opinion, is is a great thing. And of course, it's not without challenges. But but I think long term, it's where we should move as a species. 

Michael Baxter [00:47:12] Yeah, very interesting. So I think Adam Smith, I was thinking of The Wealth of Nations and you wrote that in 1776, if I remember correctly. And one of the examples he gave was of a pin factory. If an individual is making the pin, they get absolutely nowhere. If you look if you get different people specializing in different parts of the process, then, you know, you can you can send out the pin. But by the thousands today and and that was 240, 20, 40 years ago, I suppose. And maybe we're taking that argument to its logical conclusion with with alien technology. I agree. Okay. Just one more point about future proofing. Back in the 1980s, there was a computer called the Enterprise Zone. As you remember it quite well. And it was it was a really impressive piece of kit, which never went anywhere. No one's ever heard of it now. I would imagine that it was my work computer and it had the slogan obsolescence obsolescence built out. Quite ironic, because I think that computer is completely obsolete today. But. I think that one of the things that super irony talks about is future proofing or at least de-risking the adoption of payola. Is that right? And can you expand upon that? Yeah. 

Brad Cordova [00:48:44] So one of our assumptions of of why we why we built the platform the way we did is we we view the actual A.I. model as as a commodity. And we we see it changing almost every month. I don't think people realize how quickly the field of AI is, is, is growing and how many innovations or it's exponential in our human brains work on a on a linear scale. And so our view, as if we spent all of our effort trying to make the best AI model today. Well, tomorrow we're going to be out of business. And then if you couple that with the with Google, Amazon, and they're spending $4 billion every year doing this. And so we realize that if we're having that problem, everyone's going to have that problem. And so we decided to build the protocol and the platform to take models and build applications because that is the thing that's going to be valuable long term. And as a side effect that if future proofs you allows you to always have the best and no matter what technology comes out, whether there's some new symbolic A.I. or new machine learning model. Well, because we we built the protocol in the right level of abstraction, you can use all of this, and you don't need to change your system. And this is only possible because of the because of AI and the properties of AI. If we try to do this with software, we wouldn't work. And and just like when the Internet came out, there was new business models like platforms and network effects and new jobs like the chief information officer. The same thing is true in the world of AI. There's new business models, for example, the one we're pursuing new roles, Chief Data Officer All these things. And so that's what allows this to be possible. 

Manish Rai [00:50:41] Yeah. I mean that today I'm beginning to see roles of Chief Automation Officer or Chief Intelligent Automation Officer. Sure. When we started, Rob, in this industry almost to the GBS use used to report most of the time and to the CFO or CIO or some other function. And I saw the Cicilline report that came out that 30% of them report to the CEO directly. So the role of intelligent automation and how you harness I humans and software to build a more agile organization and digitize your processes is becoming much more strategic and important. And and that's why leaders are reporting directly to CEO today compared to where they were maybe five years or a decade ago. 

Rob Hughes [00:51:39] Mm hmm. Yeah, that's. And I think, you know, I think it's actually table stakes now. You know, I still hear the term digital transformation, and I'm scratching my head thinking, you know, when were you created and where have you been? Because I think all companies are digital to a large degree in certain aspects of their delivery or their their structure is not digital. And then they've got to step back and say and understand what is the potential for all of this? Like what? Because I have all of this stuff connected, because I have all of this data that I can't capture if I captured it, what's the potential? What's the opportunity? And I think yeah, I think companies just have to do this nowadays, right? They have to modernize really quickly. If you are building a business today, you would build it very, very differently than you would ten years ago or 20 years ago, 30 years ago. But they're still stuck with a lot of that infrastructure. And the quicker they can move away from that, the more competitive advantage they gain. And we see that across every industry and every company that's that's using technology for this type of work. Right. 

Brad Cordova [00:52:40] And your. I agree. And Michael, your quote reminded me of of Michael Levitt. I think he said something like, to survive, companies have to plot the obsolescence of what currently provides their likelihood. And so I think planning out what could kill you tomorrow and integrating that is really powerful. 

Rob Hughes [00:53:02] But yeah, on the rolls, actually the best role I've heard is chief disruption officer because they of disrupt yourself because if you don't somebody else is going to. Right. So staying ahead of your own disruption is is a good idea to say that's not good enough. There's a different way to do it so try to break everything is sounds like my kind of job actually. Yeah. 

Brad Cordova [00:53:23] I think that should be the CEO. Yeah. I feel if you have a role for it, no one's going to listen to that. 

Rob Hughes [00:53:30] It's true. That's a good point. Yeah, yeah, yeah. They need an enforcement title. Yeah. 

Michael Baxter [00:53:34] That's quite interesting. One of the phrases I heard that in order to resist the dangers of disruptive technology is to cannibalize it. So, you know, robustness might have said, no, we don't want to have some videos for which we don't collect fines because then would be characterizing or in products. But actually that was their own downfall and there are many examples of that. And I suppose a less glamorous fashion is to disrupt or be disrupted. Or if somebody once said to me, if it's not broken, break it. 

Rob Hughes [00:54:12] And that's the one. Yeah. 

‍Michael Baxter [00:54:15] Okay. Well, that's been really very, very interesting. And so I think we've heard about why Siemens eyes dirty little secret. I think we've heard about the unified A.I. platform and how the wisdom of crowds when those crowds includes AI algorithms and when they improve humans can be very, very powerful. And we talked a little bit about, you know, the future as well. And unstructured data is perhaps one of the big challenges of the next decade or two. And I know you both have backgrounds working, so which I think overlap. We're talking quite a lot. Do you have anything else to add in that context at all? 

Manish Rai [00:55:09] I mean, I would say one thing and Salesforce was started in 1999 and they came out with a campaign in 2000 and saying, end of software. And at the end of that year they did four 5 million in revenue out of that. And that was the beginning of the SAS revolution. I'd like to posit that we are entering the phase of end of unstructured data. So as I matures over the next ten years, the line between structured and unstructured data will blur and we won't think of them as different because humans don't make that distress distinction structured, unstructured. It's all data and we process it. And ten years from now, we won't even think about this. And so we're starting with then disruptive journey of processing all kinds of data in a single platform. 

Michael Baxter [00:56:09] So this unsent unstructured data to send, does that. Does that work out? I think it might be. 

Manish Rai [00:56:17] Yeah. 

Michael Baxter [00:56:20] Okay. Well, that's been very, very interesting. Anybody got anything else to ask before we close? 

Rob Hughes [00:56:26] I just want to say, look, thanks very much, Manish and Bradford, for joining us. This has been really interesting. I love the take on string theory, as I'm sure maybe you don't understand that either. Right. So, you know, my best reference is the Big Bang Theory, and that's it. But yeah, it's been really, really interesting. I think you guys are doing something really exciting. And I'm just wondering, I'm just going to see if anybody wants to get in touch with super. I you're going to have these, Manish, as your as your point of contact. But I've seen this quite a few comments and so on about how we get in touch and whether or not certain data pools can be accessed and how you can integrate with you. So I'm just sharing the details now. And if not, please get in touch with us. It's Gopi and I. We can. We can marry you up pretty much, really, and really enjoyed it. And yeah. 

Michael Baxter [00:57:22] I think we will. There was a recording of this which we will be putting on YouTube and on a website. I will be doing an article based on this as well. So there will be more information to follow. 

Manish Rai [00:57:39] Great. 

Brad Cordova [00:57:40] That was fun, guys. And if anybody watching has any questions, reach out to manage the email me because my email never gets checked. 

Manish Rai [00:57:52] And also on our website, you can book a demo and we we can definitely have our people get on and and give you a more detailed insight on onto our platform. 

Michael Baxter [00:58:08] Thank you very much. Thank you very much, Manish. 

Rob Hughes [00:58:11] Thanks, everyone. We'll see you guys. Everyone for Darlene in just about.