EPISODE #72: Answering Turing: The AI Visionary On The Cusp Of Solving Computing’s Greatest Question (Part II)

Guy Nadivi
14 min readSep 2, 2021


( Click here to listen to this podcast episode )

In Part II of this 2-part episode, we continue our conversation with Chetan Dube, who for over 2 decades has been haunted by the seminal question Alan Turing posed in 1950 — “Can Machines Think?” Chetan’s years-long Odyssey in pursuit of an answer led him to found Amelia, one of the market leaders in enterprise AI software.

During this segment, we’ll learn how the government can significantly reduce its trade deficit by leveraging AI, when the majority of the workforce will be digital (hint: it’s sooner than you think), and Chetan’s tips on how to “hire” the right digital employee.

Guy Nadivi: Chetan there are some concerns about biases, intentional or otherwise, creeping into the algorithms that power things like conversational AI. In order to root out bias, Harvard Business School published an article not long ago, calling for the auditing of algorithms the same way companies are required to issue audited financial statements. Do you think AI algorithms should be audited in the same way financial statements are for publicly traded firms?

Chetan Dube: Brilliant question. How do you provide… Hawking was one of the profound thinkers of our times. If left, and even Musk and others agree, that while you will find Elon Musk saying that if your competition has AI and you don’t, you’re dead. I don’t think that the perceived hyperbole in the statement is really too far from the truth. It’s quite accurate. It’s just a question of whether it’s now or it’s a slow process that will last till 2025. The idea here is that, Hawking always advocated that, Musk has also advocated this, and I tend to agree that there should be some way for us to make sure that these are such powerful agents. They are so capable. Right now, we are at the tipping point. At least as a company, I can tell you that, when you find in over 210,000 calls yesterday that happened, you find the NPS characteristic of a digital agent, Amelia, exceeding that of human agents by 14 points, in the calls that were happening in this telco, you know that you are at the tipping point.

With the technological maturity of neocortical emulation, where you are able to deliver services that are equivalent, and in some cases even superior to what a human agent would be able to do because, the digital agent right away has access to all the information, and the digital transcripts, and everything about you. In the past, I had to take some time to read about. Well, if I were additional digital agent, I would have read all that and would have already known that, and you benefit from the quality of interaction there. So we know we are at the tipping point of this. So how do we ensure that this technology, which is so powerful, perhaps one of the most seismic things that has happened to our civilization, is used for gainful purposes? And how do we do that so that it does not impede the progress, and does not become onerous on the progress of the technology?

And so, there’s a balancing act between, yes, to ensure that it is conforming to being used for the good, and for efficiencies, for liberating man from common chores, but not for some derisive purposes, and not for some destructive purposes, and not for some cyberattack and other purposes. So those things, I think there has to be a balancing act between governance that needs to be exercised, without it becoming too onerous so that it starts impeding the growth of the technology.

Guy Nadivi: In 2020, Gartner estimated the worldwide conversational AI platform market is worth about $2.5 billion, and growing at a rate of 75% year over year. How did the COVID-19 crisis affect that growth rate?

Chetan Dube: If you asked me what has been the biggest catalyst in acceleration of the digital adoption? Unfortunately, that’s one silver lining of this very unfortunate pandemic. It has been COVID-19 that caused the world… The world went into COVID-19 as a manual-mostly world, with some experimentation with digital employees. That’s a fact. The world emerged out of COVID-19 right now, as clearly a digital first world, with humans in elevated positions educating the digital agents on how to be better at the jobs they perform for boosting up the customer satisfaction.

So COVID-19 has been, we have found it right across and in some verticals more than the others, like obviously in the case of healthcare, we have found a big acceleration happening due to COVID-19. And also in the case of some telcos or those we have found a big acceleration because, the customer care facilities itself could not be staffed adequately because of this unfortunate crisis. And particularly in a case in Europe also, we have found where GDPR and the data privacy was very strongly guarded, and you needed the data not to go out of the bounds of the customer care center, and now you had your employees that were unable to come to the customer care centers. And so we have found that in those cases also, there’s been a big acceleration to be able to get digital agents who will conform, and will be able to work within the bounds of GDPR and data privacy. And so, COVID-19 has been an accelerant for that transformation.

Guy Nadivi: As you alluded to earlier, Gartner estimates there are over 2,000 conversational AI platform providers currently, and they ranked Amelia, along with IBM, Microsoft, and Nuance at the very top of that landscape. Inevitably, there’s going to be a shakeout which will cause that vendor pool to decrease precipitously. Chetan, how will Amelia continue differentiating itself in order to remain a leader in conversational AI?

Chetan Dube: As long as Amelia is your best digital employee, she’ll continue to be hired. As long as Amelia is able to deliver superior containment, and a better NPS to your customers, she’ll continue to be hired. As long as Amelia provides better coverage and better ROI, she’ll continue to be hired. Despite the fact that others might have more bigger banners, the issue is, the math of it has to be irrefutable. And there is a great unequivocal assessment in this case, and it is based on realistically understanding, how do humans comprehend? Comprehension is a very tough nut to crack because, again, I mentioned this in the passing, and I would like to emphasize that, all those 2000 that you talked about, they do… Chris Fitzgerald, who used to be the CIO of NTT said, “All these chatbots are IVR 2.0. They’ve done a fantastic…”

In general, we have done a fantastic job of covering up that decision tree that you use to experience in IVR, where we said, press 17, press 11 and press 12. And you found your customers yelling, “Operator, representative,” getting aggravated with you because IVRs just didn’t understand the problem cases that the customers were trying to convey.

No? You have had your personal experience of that I’m sure, with IVR. Now, according to the CIO of NTT, he said, “You guys have done an exceptional job of putting a very thin veneer of what is a deep neural network, often a shallow network, that typically takes the input and classifies it into one of the buckets. So it appears to me that, I’m asking a question and the person, the other digital agent on the other side is understanding it, but all you continue to do is still classify me into a number, and play the song, whatever is at the back of that number.” Now, it works fantastically for shallow tasks. Book me a car, tell me a joke, find me a good French restaurant. When is the next train leaving? All those shallow quick tasks, administrative tasks.

But what percentage of a company’s payroll is administrative tasks? McKinsey estimates less than 1.8%. What is the majority of the payroll cost? Knowledge workers. That accounts for 35% to 50% of an enterprise costs. Now, those knowledge workers are not dumb doorknobs. The knowledge workers are not just doing atomic tasks. The knowledge workers are doing tasks of solving problems every day. And you cannot emulate their intelligence with simple decision trees that you could draw in IVR kind, after classifying inputs that are coming in. What you need to do is to really cross the chasm from classification to real comprehension. What is that based on? A big segment of that is based on not just using deep neural network, which are great connectionist engines, but fusing them with logic, which allows us to be able to form a connectionist model that is much stronger than what is commonly present in the industry.

So, how Amelia will continue to differentiate itself, we have a great leader and we have a great entity to study, and that’s the human brain. Our bot is not about, are we edging out this person or that person? Our bot is human levels of comprehension. I’ll tell you a personal story. It was about two and a half decades ago, I read that the Turing’s paper, the thesis that starts by saying, “I propose for you to consider the question, can machines think?” For two and a half decades, that question has haunted my sleep. I think we are approaching closer and closer to an area where we can, as the numbers illustrate, really start to develop thinking machines, comprehending machines. Not just classification machines, not just IVR machines, but machines that really understand the problem you are communicating, and are able to solve the problem that you are communicating.

And the pursuit has to be very purist about it. Can we really set our eyes on that human leadership that comes from human… Look, even the most common mammal has got a primal brain. This is about the new cortex that allows humans to be differentiated, that allows us to be able to think. And that’s what is our biggest source of inspiration, to be able to continue to study, how are humans and not… Not to be able to study a neuromorphic or neurosynaptic course, but to be able to study what the abstraction of the centers of human brain are. What is coming out of the semantic hippocampus? What is coming out of the episodic frontal event-based memory? What is coming out of the affected emotional memory? What is coming out of the analytic trending memory? And to be able to integrate all of that, to be able to generate human levels of conversational competence, or problem-solving competence.

Guy Nadivi: Chetan, PWC has estimated that by 2030, less than a decade away, China will be the biggest AI market, accounting for 26% of global AI market share. What can and should the US do to ensure it doesn’t fall so far behind China on AI, that it can never catch up?

Chetan Dube: China is obviously very serious about knowing that the competitive advantage for a company or a country will come from AI. That’s the biggest competitive differentiator. If you look at the amount of investments China has made, and the government of China has made in AI, it is staggering. And if you see the number of research articles coming out of China, they are now starting to outstrip the number of research articles coming out from United States on AI.

I still maintain that, the quality of AI research being done in this country is superior to just about anywhere else in the world. And that’s where it’s not just the volume or the quantity of research, I think it is the quality of research, and the quality of outcomes, that should become a big focus. And I think the US government has to realize that the biggest competitive differentiation in all aspects… And commerce is becoming global, consumer base is becoming global. A country is not operating with its profile of just 76% domestic consumption. It has to really think about like it’s… One of the biggest killers for deficit will be the ability of the government to be able to leverage AI for exporting its AI based services. Particularly as markets of the size of 2.9 trillion open up by 2025, in the automation of knowledge workers, as estimated by McKinsey.

Guy Nadivi: Chetan, automation is fundamentally different today than it was 20 years ago. What have been some of the biggest changes you’ve witnessed so far? And what do we need to pay more attention to?

Chetan Dube: Automation and AI, Guy, went through a winter where, you had John McCarthy and Minsky at one point assume that we feel that, and this was ’80s, the roaring ’80s, where we felt we would be able to just about clone everything with the human intelligence. And then we found out, even as the father of artificial intelligence, as my professor reminded me, found out that real artificial intelligence is very hard. The ability to clone human intellect turned out to be much harder than anticipated.

But since then, not only have deep neural networks evolved, we’ve had the privilege of working for some of the most distinguished minds in the world from Stanford and other universities, that actually have furthered the state of artificial intelligence. And this is not just deep neural networks, I emphasize. And also of course, compounded by the fact that you now have compute power, or accelerated by the fact that you have compute power that, earlier to compute these very intensive, whichever one, ELMo, or BERT, [inaudible] model that you are thinking about, these models are very compute and memory intensive.

Earlier, you could think about, if you wanted to have a learning engine that used these massive data models, it would take inordinate amount of time and compute power for you to be able to do that. Today, you have had the evolution of compute cycles, and obviously the dropping down of the cost of memory and cost of CPU, has allowed us to be able to provide such power computing, that can facilitate the generation of these machine learning models, that have furthered the state of AI much more than what we saw, a very gradual progression followed by a winter, and then a little hibernation. And then, now it has picked up where… The interesting part is, now there’s actual cases where people in banking, people in insurance, people in healthcare, retail, telco are starting to see the returns on their investment, which has caused more excitement in the market.

Guy Nadivi: The pace of innovation for digitally transforming technologies can leave your head spinning. Chetan, what do you envision will be some of the biggest disruptions we’ll see in the next one to three years, with respect to automation and conversational AI?

Chetan Dube: I have always maintained that, and I think our different experts are of similar mindset as I, but only thing that they don’t agree on is the timeframe. We feel strongly based on the research that we’re doing, and the progression in the thinking capabilities of machine. That by 2025, you’re going to pass someone in the hallway, and you will not be able to tell if it’s a human or an Android. You will have majority of the workforce in the world, and in your companies, will be digital by 2025, because you would have thrived by that time. 52% of your workforce, World Economic Forum agrees, by 2025, is going to be digital.

And the sooner you start getting there, the more competitive edge you will have because, we’re living in digital Darwinistic times where, you have on the left side of the curve, digital experimenters that have just done password resets, and active directory, and account and WiFi setup, and small little IT tasks. On the other side, you have digital front runners which have taken big parts of their supply chain, big parts of their customer relationship management, big parts of their origination, and big parts of their higher friction, low margin assets, and completely started to transform them into being rendered by digital agents. The differentiation and the returns that the people who are digital front runners and the digital laggards is so dramatic, and it is obviously progressively being realized in the marketplace because, the cost of goods services, and the cost of… And also the quantity of services, the NPS that they are able to drive is very different. That you are starting to see this chasm, a Darwinistic curve being created between the front runners and the laggards.

The biggest digital disruption that we are going to see is that, you will have a hybrid workforce. You will have a hybrid workforce which will be gainfully employed, digital employees. By 2025, majority of your workforce would’ve become digital.

Guy Nadivi: Amazing progress, almost unimaginable. Chetan, for the CIOs, CTOs and other IT executives listening in, what is the one big must have piece of advice you’d like them to take away from, our conversation, with regards to implementing automation and conversational AI for their business or IT operations?

Chetan Dube: Brilliant. Hire wisely. You spend so much amount of time interviewing and hiring the right human employee. No? Don’t you really know that the team with the best players wins? You take an incredible amount of time in making sure that you differentiate yourself by trying to get some of the best employees, human employees in your workforce. When the majority of your workforce is going to be digital in the next three and a half years, shouldn’t you be spending the same amount of attention to making sure that you are hiring a digital employee, which has got human equivalent capabilities? Don’t just hire the glossiest chatbot or something. Subject it to a human level of comprehension test. Ask it. If it’s going to do the task of a human agent, let me ask you the question about what my call center employees do really get. Ask it the simplest question that your call center employees get. Which is, “Hey, why was my policy canceled, when I just paid $68 last month?”

Ask it a question of the examples that I gave. Ask it, “Why is my bill 163 Soles when last month it was 158 Soles?” Ask those real questions. And it won’t take you too long. Don’t go by the brochure. Ask the questions. Ask the empirical, not just cookie cutter demos. But say, “Here are five different questions that came in, not just the password resets or account unlocks, or WiFi set up,” which will be just the rote part of the thing. But then you would have set your company up for getting the best digital employees, and a differentiated customer experience with a differentiated way. Cost of goods sold calls would have been reduced because, your digital delivery is going to be at least 35% to 40%, more effective than the human.

The fundamental platform for how services are being rendered is shifting. Last decade saw it being shifted from humans to more cost-effective, I’m talking about 2000–2010, to more cost-effective humans. We are now starting to see this fundamental platform for how services are rendered is shifting from humans to digital humans. Hire the right digital employee that can give you the NPS characteristic. That’s the closest to human levels of comprehension. And you will have set your company up for competitive differentiation, and climbed your company up in a digital Darwinistic curve.

Guy Nadivi: All right. Looks like that’s all the time we have for, on this episode of Intelligent Automation Radio. Chetan, despite artificial intelligence having plenty of naysayers and skeptics, you’ve grown Amelia into the largest privately held AI software company. So you’ve not only proven the viability of AI for business, you’ve also blazed a trail to follow, for the many other innovative AI companies out there, offering unique solutions to challenging problems. That’s something we all stand to benefit from, and you’re truly owed a debt of gratitude. Thank you for coming onto the podcast today and sharing your insights with us.

Chetan Dube: Thank you for the kind words, Guy. It’s my privilege. And thank you for this very interesting session.

Guy Nadivi: Chetan Dube, President, CEO and Founder of Amelia, an IPsoft company. Thank you for listening everyone. And remember, don’t hesitate, automate.


President, CEO, and Founder of Amelia, an IPsoft company

Chetan Dube has served as the President and CEO of Amelia, an IPsoft Company, since its founding in 1998. During his tenure, he has led the company to create a radical shift in the way IT is automated and managed, and introduced the industry to market-leading Conversational AI technology. Previously, Chetan served as an Assistant Professor at New York University, where his research was focused on deterministic finite-state computing engines. Chetan is a widely recognized and sought-after speaker on autonomics, Conversational AI, cognitive computing and AI’s impact on the Future of Work. He also serves on the board of numerous IT-related institutions.

Chetan can be reached at:

Conversations (Chetan’s personal site): https://chetandube.ai/

LinkedIn: https://www.linkedin.com/in/chetan-dube-13315940/