EPISODE #12: How Cognitive Digital Twins May Soon Impact Everything
( Click here to listen to this podcast episode )
“May you live in interesting times” (a 19th-century English expression often mistaken as being of Chinese provenance), aptly describes today’s technology environment. Artificial intelligence, machine learning, data science, and the internet of things were already interesting enough. It turns out though that those were just prologue to the rise of something called a “Cognitive Digital Twin”.
To shed light on this and other advances, we turn to Dr. Ahmed El Adl, Principal Director of AI Consulting & Intelligent Solutions at Accenture. He shares his far-reaching insights with us, and clarifies whether AI will ultimately have a bigger impact as a tool for automation or worker augmentation, whether machine learning will ultimately eliminate the need for data science altogether, and the most important background to consider when hiring an AI professional.
Guy Nadivi: Welcome everyone! Our guest today on Intelligent Automation Radio is Dr. Ahmed El Adl, Principle Director of AI Consulting and Intelligent Solutions at Accenture. Dr. El Adl holds a Ph.D. in artificial intelligence and robotics and does a lot of public speaking about AI, machine learning, the technologies and standards surrounding the internet of things and something very intriguing called Cognitive Digital Twins. And with a background like that, we’re hoping he can help us make sense of the AI market’s direction for these innovations.
Dr. El Adl, welcome to Intelligent Automation Radio.
Dr. El Adl: Thank you, Guy. Thank you for having me.
Guy Nadivi: Let’s dive into it. Dr. El Adl, do you think artificial intelligence will ultimately have a bigger impact as a tool for automation or a tool for worker augmentation?
Dr. El Adl: Great question, Guy. I think it will play a major roll on both fronts. So, from the automation perspective, let us talk about two major categories, the process automation and machine or industrial automation. We see today one of the major areas of adopting even simple AI technologies is intelligent process automation. This is already going on and showing a lot of business values. The other one, which has maybe the number of implementation maybe is less than industrial automation but the impact is huge. On the other side what we call human worker with augmentation, so the humans will stay in the equation, will stay in everything we are doing. Maybe we will change. What we are doing will change. How we are doing it will change, but we’ll stay a measure part of the equation. Therefore I don’t see that one of them is going away, both, but we see the impact today.
Honestly it is equally, from the adoption perspective it is equal. If you see today the human worker augmentation take for example, call centers. They take virtual agents, customer interaction like chat bots and all of those. Adoption is very hard and it is rising every day. On both fronts, automation, whether process or machine automation, human augmentation, or also say not augmentation only, but also to replace humans in some areas that humans can make the best out of our biological intelligence, which is very valuable. So I see the adoption on both sides, Guy. It depends on the functional areas and the industry as well.
Guy Nadivi: So let’s talk about industry. You’ve stated that in your role at Accenture you are “driving the application of artificial intelligence into every industry, globally”. Given that lofty objective, from your vantage point, what industries and functional areas have you seen experience the biggest transformational impact from deployment of AI?
Dr. El Adl: Yes, just a small correction. My focus within Accenture is mainly on the fossil energy, renewable energy, utilities, chemicals, mining, which I will come back to this later on. So, three major areas I see the adoption is from the number of implementations and the added values to the business. Within a very short time there are three major areas. One, which is customer interaction. We know that one of the most mature areas in machine learning is natural language processing, natural language understanding, and we see all of those intelligent assistants, Alexa, Cortana, Siri and others. The adoption of those technologies is very, very high in customer interaction and customer understanding also.
This is one of the major fronts I see nearly across the board. Any industry which has intensive customer interaction, mainly B2C kind of interaction, I see major adoption whether it is virtual agents, chat bots, or software bots automating a lot of work, which takes a lot of time from humans while they are interacting with customers. The second area is, which has in my opinion the highest values — industries with critical physical assets. Take oil & gas industries, chemical industry, automotive industries, aviation industry. All of those industries have used machine learning or other AI technologies since actually a long, long time. Since I was myself in academia, you could mainly say aviation industries maybe automotive industrial machinery using embedded intelligence or embedded machine learning algorithms. Those industries today in my opinion, are leading the serious adoption of complex AI capabilities from the simple statistical machine learning all the way up to very sophisticated deep learning or deep artificial neural networks and knowledge representation capabilities.
The third area, which is R&D. Take for example pharmaceutical industry. Here I am sitting today in Boston. Across the street is MIT and Harvard. They are using very sophisticated machine learning approaches and other AI technologies and capabilities for drug discovery and showing results. It is showing real results especially when you combine sophisticated machine learning algorithms with quantum computing, delivering the required computing capabilities. This is a game changer. On the other side, even we at Accenture are working this field but also we have what we call digital chemists. It is exactly the application of complex AI or sophisticated machine learning and reference to help R&D people to discover in pharmaceutical, new drugs and chemicals or specialty chemicals, discover new chemical products, new materials, and so on. So those are the major three areas, Guy, where I see serious adoption. Not only in terms of the revenue but also in terms of the values which the business leaders are looking for.
Guy Nadivi: Now that serious adoption you refer to is driving up the demand for AI professionals, and given the critical talent shortage in AI, machine learning, data science, etc., should organizations rely on in-house staff or outsource when planning for their AI deployment?
Dr. El Adl: Yes, so I think to answer this question, Guy, I think let’s take a step back and now we have AI projects on the industry side, not anymore on the academic side. On the industry side, the structure of an AI project is as follows. You have data engineering team, you have data science team, you have machine learning team, and you have around them the IT infrastructure teams, UR/UX teams and all of those subgroups. One of the major misunderstandings we see today is you wanted to start an AI project, you have to hire 10 academics, have a Ph.D. from MIT or Stanford in AI. This is misunderstanding, this is not true. Actually if you have 10 people in your AI team maybe you need one or two good AI experts with strong AI academic background. For me personally, I like to hire for the machine learning part, I like to hire people with strong academic AI background even if they never applied machine learning to solve industry problems. But in this case, can every company hire those people? No. They don’t exist. Even if they exist they have some preference around the employer and so on, so what I see today, Guy, is those guys work mainly for consulting companies. I believe that this is the right approach, and consulting companies share those people across different projects. It doesn’t harm if you are a large company to hire one, two, three, or five people, but I think it should be a hybrid approach. Those very experts in AI, I think I hear some numbers around 10,000, 12,000 people across the globe, which is 1% of what we need today. What I want to say is let us use a hybrid approach where we can share those rare sources among different projects, and the 80% will come from your existing workforce.
Guy Nadivi: Welcome everyone! Our guest today on Intelligent Automation Radio is Dr. Ahmed El Adl, Principle Director of AI Consulting and Intelligent Solutions at Accenture. Dr. El Adl holds a Ph.D. in artificial intelligence and robotics and does a lot of public speaking about AI, machine learning, the technologies and standards surrounding the internet of things and something very intriguing called Cognitive Digital Twins. And with a background like that, we’re hoping he can help us make sense of the AI market’s direction for these innovations.
Dr. El Adl, welcome to Intelligent Automation Radio.
Dr. El Adl: Thank you, Guy. Thank you for having me.
Guy Nadivi: Let’s dive into it. Dr. El Adl, do you think artificial intelligence will ultimately have a bigger impact as a tool for automation or a tool for worker augmentation?
Dr. El Adl: Great question, Guy. I think it will play a major roll on both fronts. So, from the automation perspective, let us talk about two major categories, the process automation and machine or industrial automation. We see today one of the major areas of adopting even simple AI technologies is intelligent process automation. This is already going on and showing a lot of business values. The other one, which has maybe the number of implementation maybe is less than industrial automation but the impact is huge. On the other side what we call human worker with augmentation, so the humans will stay in the equation, will stay in everything we are doing. Maybe we will change. What we are doing will change. How we are doing it will change, but we’ll stay a measure part of the equation. Therefore I don’t see that one of them is going away, both, but we see the impact today.
Honestly it is equally, from the adoption perspective it is equal. If you see today the human worker augmentation take for example, call centers. They take virtual agents, customer interaction like chat bots and all of those. Adoption is very hard and it is rising every day. On both fronts, automation, whether process or machine automation, human augmentation, or also say not augmentation only, but also to replace humans in some areas that humans can make the best out of our biological intelligence, which is very valuable. So I see the adoption on both sides, Guy. It depends on the functional areas and the industry as well.
Guy Nadivi: So let’s talk about industry. You’ve stated that in your role at Accenture you are “driving the application of artificial intelligence into every industry, globally”. Given that lofty objective, from your vantage point, what industries and functional areas have you seen experience the biggest transformational impact from deployment of AI?
Dr. El Adl: Yes, just a small correction. My focus within Accenture is mainly on the fossil energy, renewable energy, utilities, chemicals, mining, which I will come back to this later on. So, three major areas I see the adoption is from the number of implementations and the added values to the business. Within a very short time there are three major areas. One, which is customer interaction. We know that one of the most mature areas in machine learning is natural language processing, natural language understanding, and we see all of those intelligent assistants, Alexa, Cortana, Siri and others. The adoption of those technologies is very, very high in customer interaction and customer understanding also.
This is one of the major fronts I see nearly across the board. Any industry which has intensive customer interaction, mainly B2C kind of interaction, I see major adoption whether it is virtual agents, chat bots, or software bots automating a lot of work, which takes a lot of time from humans while they are interacting with customers. The second area is, which has in my opinion the highest values — industries with critical physical assets. Take oil & gas industries, chemical industry, automotive industries, aviation industry. All of those industries have used machine learning or other AI technologies since actually a long, long time. Since I was myself in academia, you could mainly say aviation industries maybe automotive industrial machinery using embedded intelligence or embedded machine learning algorithms. Those industries today in my opinion, are leading the serious adoption of complex AI capabilities from the simple statistical machine learning all the way up to very sophisticated deep learning or deep artificial neural networks and knowledge representation capabilities.
The third area, which is R&D. Take for example pharmaceutical industry. Here I am sitting today in Boston. Across the street is MIT and Harvard. They are using very sophisticated machine learning approaches and other AI technologies and capabilities for drug discovery and showing results. It is showing real results especially when you combine sophisticated machine learning algorithms with quantum computing, delivering the required computing capabilities. This is a game changer. On the other side, even we at Accenture are working this field but also we have what we call digital chemists. It is exactly the application of complex AI or sophisticated machine learning and reference to help R&D people to discover in pharmaceutical, new drugs and chemicals or specialty chemicals, discover new chemical products, new materials, and so on. So those are the major three areas, Guy, where I see serious adoption. Not only in terms of the revenue but also in terms of the values which the business leaders are looking for.
Guy Nadivi: Now that serious adoption you refer to is driving up the demand for AI professionals, and given the critical talent shortage in AI, machine learning, data science, etc., should organizations rely on in-house staff or outsource when planning for their AI deployment?
Dr. El Adl: Yes, so I think to answer this question, Guy, I think let’s take a step back and now we have AI projects on the industry side, not anymore on the academic side. On the industry side, the structure of an AI project is as follows. You have data engineering team, you have data science team, you have machine learning team, and you have around them the IT infrastructure teams, UR/UX teams and all of those subgroups. One of the major misunderstandings we see today is you wanted to start an AI project, you have to hire 10 academics, have a Ph.D. from MIT or Stanford in AI. This is misunderstanding, this is not true. Actually if you have 10 people in your AI team maybe you need one or two good AI experts with strong AI academic background. For me personally, I like to hire for the machine learning part, I like to hire people with strong academic AI background even if they never applied machine learning to solve industry problems. But in this case, can every company hire those people? No. They don’t exist. Even if they exist they have some preference around the employer and so on, so what I see today, Guy, is those guys work mainly for consulting companies. I believe that this is the right approach, and consulting companies share those people across different projects. It doesn’t harm if you are a large company to hire one, two, three, or five people, but I think it should be a hybrid approach. Those very experts in AI, I think I hear some numbers around 10,000, 12,000 people across the globe, which is 1% of what we need today. What I want to say is let us use a hybrid approach where we can share those rare sources among different projects, and the 80% will come from your existing workforce.
Guy Nadivi: And so if I’m an IT executive thinking of doing an AI deployment and I’m gonna hire some people in addition to maybe outsourcing some work, what do you think would summarize the most important qualities an AI expert should have that I should look for?
Dr. El Adl: Yes, this is a very, very, very good and very important question, Guy, here. Two things I look for. One is strong academic background in machine learning. What I mean by that is, we see a lot of data scientists coming from data analytics background pretending to be machine learning experts. This is not true and this is one of the major reasons some POC or AI POC projects fail today, because we know that machine deep learning main goal is to eliminate data science. So if my job is to eliminate your job, you cannot take my job because you don’t have the qualification for that.
Let us go deeper a little bit, Guy. The mathematical background of machine learning is not or doesn’t exist on the data science or data analytic part, and when you are faced by real world problems you don’t have only to take the standard algorithms we have if we were the open source or platforms and just use it as it is. You will have to understand the industrial problem and mathematically tweak your algorithms to bring the best and most reliable results out of those algorithms. So this is the first quality. It is academic background.
Secondly at least you should be aware or familiar with at least one open source library and one enterprise-grade AI today. Take Google Cloud AI platform/TensorFlow, open source libraries, Azure, AWS, you name it, but those are the major two qualities. The academic part and the practical part in terms of libraries or enterprise AI platforms.
Guy Nadivi: Let’s switch gears a little bit Dr. El Adl. In late 2016, you coined the term “Cognitive Digital Twin”. Can you please explain to our audience, what is a cognitive digital twin and how will it potentially impact their organization?
Dr. El Adl: The digital twin? OK. Yes, I do believe, Guy, and maybe you read my article I wrote in 2016 as a response to a lot of noise at the time around the term “digital twin”. To explain the digital twin I also like to take a step back. What is the ultimate goal of digital twin today? Beyond the goals NASA had in late 90’s. In the late 90’s, NASA created the first digital twin for simulation for the spaceships, for the engineering and physics-driven simulations. Today actually the real rise of AI, the progresses we have on different areas like IOT, sensing, mobile communication network, and all of those technologies together actually give us a hope and I am one of the hopeful people that someday we’ll create machines independent like our human body. They can sense. They have edge intelligence. They have central intelligence. They have distributed intelligence. They can take smart decisions. They can take smart actions.
We have enough from those, sorry, stupid machines we have today. So the ultimate goal is to have a physical twin and digital twin first separated and the digital twin is not only a representation of the physical twin, this is not the right definition. It is partial representation, partial augmentation, and companion for its physical twin. Once you have the baseline of digital representation, you continue updating this new creature in the digital world, which will continuously represent the physical twin, continuously augment and also extend the capabilities of the physical twin and combine it across the life cycle.
One of the major problems, Guy, which unfortunately one of the major reasons that many digital twin initiatives failed, POCs failed, why? The initial initiatives to create digital twin mainly was kind of engineering digital twins. Some vendors connected the physical assets, retrofitted them with maybe some sensors, collected data, and you have real time data coming in, in a data lake, and these data lakes are growing by minute. You are dumping a lot of data in those data lakes and we know from the data science perspective, even from the data management perspective, this could be a nightmare. And I told some of my friends and colleagues in this industry after one or two years, you will drown in your data lakes and you will not be able to manage anything and unfortunately, without mentioning names, it happened to some major vendors in this field.
Why I coined the term “Cognitive Digital Twin”? It is exactly as I mentioned, biologically inspired. When you ask me a question now, I do not go and read everything I read before to answer your question because I read, I practiced, I memorized some parts, but other parts are knowledge represented somewhere, I don’t know how, in our human brain. The same here. I want that we don’t only collect data and have large data lakes or data oceans, I want that we continuously — the digital twin, have the capabilities to continuously convert the data to knowledge, which you need when you want to take a decision, whether the decision will be made by the digital twin or the physical twin or by a human, but if you just collect data this is not going anywhere and this is why I said, Guy, let us use AI capabilities like machine learning, knowledge representation, the simple reasoning capabilities we have today, to add a brain to the digital twin and instead of adding a large, large data lake underneath it where it will drown. This is the way I define and I hope will implement cognitive digital twins rather than just digital twins.
Guy Nadivi: What you said just now reminded me of a demonstration last year by Google’s CEO of something called Google Duplex and there was a video about it that quickly went viral. And for those in the audience that don’t remember, Google Duplex is a virtual assistant that can make phone calls on your behalf to schedule appointments, make reservations at restaurants, etc. Dr. El Adl, would Google Duplex be considered a type of cognitive digital twin?
Dr. El Adl: It is a basic component of a digital twin of a human and in my paper again in 2016 I had different categories of digital twin. One of them is a cognitive digital twin for a human, for us, and I am happy to see that some start-ups even here in Boston area started to create businesses or start-ups around creating a digital twin for humans. For me, for you, for everyone, collecting all the vital health data and other data and monitoring, continuously monitoring our health to of course have a better life.
Google Duplex, why though the video went viral? It was because the sound quality. The real-time intelligence behind Google Duplex, and this is why every one of us liked it. It looks real. It sounds real. It feels real. Everything is real and in real-time, and this is exactly, Guy, what I hope for the overall cognitive digital twin concept is to add real intelligence rather than data, and of course in real-time. By the way, one of the major promising areas of the concept of cognitive digital twin is in healthcare, and I am predicting now, I am not in the business of prediction now, but I am predicting that within three or five years you will see the rise and major investments in creating healthcare services based on the concept of cognitive digital twin. Maybe Google will use the core technology they implemented for the Google Duplex, maybe, but it will be a beautiful piece of technology to have it underneath the cognitive digital twin for a human.
Guy Nadivi: We’re gonna come back to some more predictions after the next question but first I wanted to ask you Dr. El Adl, a little bit about the internet of things or IOT. You’ve written that when it comes to IOT initiatives, many “failed miserably or were put on hold mainly because organizations underestimated the effect of the existing IT and OT infrastructure”. What are organizations underestimating about their infrastructure that’s derailing IOT initiatives?
Dr. El Adl: It’s exactly the same reasons we see today on the AI adoption site. IOT, when we started to talk about IOT or Internet of Everything in all of those terms. People rushed, “Oh IOT’s about connection”. Maybe, but it is not. IOT is about the solution, which can have the right data in the right time to take the right decision on time. This is the ultimate goal of IOT initiatives. When some clients started to implement IOT projects they discovered that they have all kinds of problems from the sensing. Sensing itself you need a lot of progress on the sensing technology side. Chemistry, biology, physics, all of those is the sensing hardware on all of this stuff.
You need it. You have a lot of problems around the communication and the network, whether mobile or line based, and again as you might remember a couple of years ago we didn’t have standards, protocols. We had to do a lot of work on the standards side. Standards and protocols and we see a lot of consortiums, like the industrial internet consortium here in Boston and similar started to work on even private standards or global standards to get things done.
Also again on the other side, this was on the infrastructure side, but also on the skills side. Again, having an IOT solution needs a lot of technology experience but actually like on AI side, you need to understand the problem you want to solve and in IOT sometimes you have to solve a problem with machines, nuclear power plants, chemical plants, manufacturing, factories. You have to understand a lot of things are on the process, the machines, the technology, the IT infrastructure. This was too much, and again as you know, a lot of bloggers promising a lot, over-promising everything and of course this resulted in failures, but the good side here, Guy, is we learned a lot. The standards and protocols people learned a lot. The hardware companies learned a lot. We as consulting companies learned lot, and I think I see now the number of failures is going down, number of successes is going up mainly because we learned, we failed, we learned, and I think we are doing better now.
Guy Nadivi: Let’s get back to predictions. In addition to your prediction for healthcare services, what are some of your other predictions for AI over the next three to five years?
Dr. El Adl: I would say I am still 30%-40% academic and the rest is maybe mainly industry and software guy. It is not honest, not professionally correct to predict. Why? Especially on the AI side. If you take this from the academic side, Guy, early 2000 Jeff Heaton published his first paper around “Convolutional Neural Network” or CNN. CNN was mathematically not very sophisticated work. It was results of a lot of work. I personally was part of it, but CNN alone enabled computer vision which we see today from simple computer vision, autonomous cars, robotics, video analytics, and so on. So one invention, one mathematical model solved the problem which opened the gates for a large number of AI applications.
Will we have similar inventions like CNN? Yes. It could happen today, it could happen tomorrow. Once such simple mathematical models or inventions happen on the academic side, you will see a lot of adoption. Therefore I wouldn’t predict anything, but I predict one thing, Guy, today. AI is here and here to stay and I always say, even if the academia ceased to deliver any new breakthrough inventions, what we have today in our open sources machine learning libraries or enterprise-grade platforms like Google, Microsoft, AWS, we have more than enough to deliver a lot of business values, solve problems, implement smart products for the next five years and we don’t need anything new from the academia at all. On the industrial side, I am predicting a second prediction, the industrial adoption of AI will rise like the IOT today. Once the business leaders and I’m not talking about IT leaders, the business leaders understood the real capabilities of AI, what AI can do, what AI cannot do, why AI can do this, why AI cannot do that, once they understand this, adoption will grow exponentially on the industrial side. Of course I’m predicting a lot of good things on the academic side as well.
Guy Nadivi: So getting back to today, what should enterprise IT managers who have never dealt with AI know before deploying it?
Dr. El Adl: First message I say always, some IT managers, their job is to take care of the IT infrastructure. Those guys will continue to be very, very important in the era of AI. We need them, whether you are SAP administrator or Oracle administrator, we need you guys. So therefore, but because we need you I want you to start reading about real AI. Don’t read books and all of this stuff, read serious materials, not mathematically sophisticated about AI. Inform yourself that you can help us create the new IT organization, which will build the intelligent business solutions and also service intelligent products. If you are today an IT organization of automotive company, tomorrow you will be servicing connected cars. Smart cars like Tesla. Tesla IT guys are not just taking care of salesforce platform or SAP platform. They are servicing the connected cars of Tesla.
The same applies to any IT organization virtually in any industry. So my message, inform yourself about AI and help us, help you to help us and stay relevant in the era of what I call intelligent enterprise.
Guy Nadivi: Sounds like some prudent advice. Alright, looks like that’s all the time we have for on this episode of Intelligent Automation Radio. Dr. El Adl, thank you very much for joining us today and sharing your thoughts about the current state of AI. It’s been great having you as our guest.
Dr. El Adl: Thank you, Guy, for having me. It was a pleasure talking to you. Thank you.
Guy Nadivi: Dr. Ahmed El Adl, principle director of AI consulting and intelligent solutions at Accenture. Be sure to visit their website to learn more about their services. Thank you for listening everyone and remember, don’t hesitate, automate.
DR. AHMED EL ADL
Principal Director of AI Consulting & Intelligent Solutions for Accenture.
Dr. Ahmed El Adl is Principal Director with Accenture, leading their AI Technology Consulting & Intelligent Solutions with a focus on understanding the recent advances in AI on the academic side, and applying them to solve real world industry problems. Dr. El Adl finished his Ph.D. and Post Grad in AI and mobile Robotics. He witnessed the last winter of AI, and is trying to avoid another one through a pragmatic and practical approach.
Dr. El Adl also published one of the most comprehensive architecture frameworks and overall visions for what he calls a “Cognitive Digital Twin”, and how it will influence everything we do soon.
Dr. El Adl can be found at:
E-Mail: eladl.ahmed@gmail.com
Twitter: @aeladl
LinkedIn: https://www.linkedin.com/in/ahmedeladl/
