Artificial intelligence frightens some, excites others, and captivates just about everyone. Inflated expectations about AI’s potential for good & bad have accompanied its narrative since antiquity, when as American author Pamela McCorduck wrote, it originated with “an ancient wish to forge the gods.” Modern sentiments about AI continue to fuel some very basic misunderstandings about its nature, as well as what it can and can’t do (yet).
To help us analyze the seemingly concealed reality within this opacity of hype, we called upon Ron Schmelzer, Managing Partner and Principal Analyst for Cognilytica, an AI-focused analyst and advisory firm. In this absorbing episode, Ron guides us through a brief history of AI, its potential to augment & multiply human productivity, and why 3-5 years from now if a business doesn’t utilize AI bots it will be the equivalent of not having a website. Along the way we’ll learn what Amazon realized from deploying automation that caused it to hire more people than ever, which career offers the biggest opportunity to take advantage of the burgeoning growth in AI, and how C-suite executives should be thinking about this transformative technology for competitive advantage.
Guy Nadivi: Welcome everyone. Our guest today on Intelligent Automation Radio is Ron Schmelzer, Managing Partner and Principal Analyst for Cognilytica. Cognilytica is an AI-focused analyst and advisory firm that zeroes in on the usage of AI in the real world, not the buzz word hype. And Ron is also a serial entrepreneur with numerous successful exits. And as if all that wasn’t keeping him busy enough, he’s also the founder and chief organizer of TechBreakfast, the largest monthly morning tech meetup in the nation with over 50,000 members at locations around the world.
Ron, welcome to Intelligent Automation Radio.
Ron Schmelzer: Yes, hi. Welcome, thank you very much for having me on the podcast.
Guy Nadivi: Glad you’re here. Ron, you’ve speculated about whether the tech industry is heading towards an artificial intelligence winter, in other words and era of reduced interest, reduced funding, reduced research, et cetera in AI, following a period of vendors over promising and under delivering on AI’s capabilities. Should IT executives make like Ned Stark from Game of Thrones and brace themselves because AI winter is coming?
Ron Schmelzer: Well it’s a good question. Well I think, first of all thank you for having me on the podcast and really thrilled to be sharing the insights that we’ve produced here at Cognilytica and some of our research. And yeah, one of our more popular newsletters is the AI Winter Coming and we’re not the only ones who are talking about it. You know for those who are sort of familiar with the history of AI, I mean AI actually is an interesting paradox because artificial intelligence goes back farther than digital computing itself. Right, the first innovations in AI were in the ’40s and ’50s, right, with Alan Turing, you know, both the inventor of, you know, the developer of the modern computing, just the whole modern theory of computing, also the Turing Test, right? The same person.
So it’s interesting that as much as AI’s older than computing, we’re still trying to achieve the goals of AI. We haven’t quite gotten to where we wanted to. And AI, sort of the history of AI, the AI industry has gone through these fits and starts, right through the ’50s, through the ’70s were a real period of AI development, that was the heyday of that first wave of AI. Then it went through what’s called the AI winter in the mid ’70s, late ’70s when governments and corporations and the sources of money were no longer enthusiastic about AI and everybody pulled back from that. They kinda regained interest in the mid ’80s with expert systems and kinda through the early ’90s and, you know, with the desktop computing and a lot of corporate venture capital, that’s where the source of money and interest was for a VC in that second wave. But just like the first wave, mid ’90s people lost interest again. The promises of AI didn’t sort of meet the expectations and the world of AI went quiet.
It was only really in like the past decade, or less, that we’ve seen a resurgence of interest in artificial intelligence mainly because of big data and almost infinite computing in the form of Cloud computing. And of course the development of algorithms like deep learning and other techniques. Now that we just have almost an infinite amount of data and the ways to handle it, we can make some of the realities of AI happen that we weren’t able to before.
But of course, you know, people always seem to get overly excited about artificial intelligence and they want, they instantly think of the Terminator and talking machines and robots that will take over humanity. And meanwhile if you actually spend any time talking to Alexa or Siri you’ll know that these devices aren’t particularly intelligent. So once again, you know we’re hoping that the over inflation of expectations, you know the hype of what AI’s capable doesn’t hit up against that hard barrier of what’s actually possible. And you know the whole under deliver and over promise, that’s the risk of that.
But you know at the same time, companies are investing more money than ever in artificial intelligence. There’s a crazy amount of venture capital being thrown into AI all across the world too, not just here in the U.S., obviously. You know China has the most valued AI company with over $1.2 billion invested in the company with supposedly another one billion dollars can be put in anytime soon. And companies in like Germany and Japan and the United Kingdom and Canada and Russia, everybody’s putting money and effort into AI.
So that combined with the fact that people are becoming more and more familiar and more comfortable with the kind of technology seems to indicate that we won’t be having an AI winter anytime soon. But I think it’s always worth noting that history does like to repeat itself.
Guy Nadivi: Your firm Cognilytica is very focused on the actual usage and adoption of AI in the real world. What are currently the biggest bottlenecks you’re seeing that are preventing wider adoption of AI and how will those be resolved?
Ron Schmelzer: Yeah, so the interesting thing about AI, and we spend a lot of our time at Cognilytica with our customers which are mostly in the enterprise. So our customers are in insurance, banking, finance, healthcare, automotive, manufacturing, and pharmaceutical. We also are in the D.C./Baltimore region so we have a lot of clients in the government, various government agencies as well as consulting firms and professional services firms and government contractors that serve those agencies. And of course we spend a lot of our time talking to vendors as well.
And it’s interesting because if you take a look at what they’re saying about their adoption, AI products are different in that they’re not really development projects. Not like you really develop AI. Sure there’s coding and software that’s involved, but really what makes AI work is data. And as I say data is the oil of the new economy. And so we have this interesting challenge. On the one hand, you know huge platform vendors like Google, Facebook, Microsoft, Amazon, IBM, they have access to tons of data. And that’s part of the reason why they’re able to accelerate their AI efforts forward because AI algorithms are very data hungry. Especially deep learning algorithms and other approaches that just require a lot of data for training.
And I think what we find is that corporations who want to make use of AI for their own particular use cases or government agencies that wanna make use, you know they can’t use those same sources of information because they’re trying to train on their own data sets. And they don’t actually have enough data. So one of the challenges of AI adoption has a lot to do with managing and wrangling data. Cleaning data, getting great, good, trainable sets of data, having an adequate amount of data so that you don’t have systems that are trained and really are faulty. So the challenge for most organizations is having a ready supply and access to sufficient quantity of trainable data that is of clean nature so that the AI systems that are built off it aren’t trained on some wacky data sets that provide very unreliable results.
So that’s the primary challenge. It’s data.
Guy Nadivi: Ron, given all the organizations you speak with, what areas within the enterprise are you seeing that offer deployment of AI the biggest opportunities for transformational impact?
Ron Schmelzer: Yeah, so I think that’s a really good way of thinking about artificial intelligence in general and that is it’s transformational impact. Because a lot of technology ways that we’ve seen before, whether it was the internet or mobile or Cloud computing, they’ve been very broad movements. And you can’t say well what value does Cloud computing bring or the internet? It’s like well it changes the way that people have built and run and managed their businesses.
In the same way, if you think of AI as a collection of cognitive technologies that help you address some of the areas that used to require human labor or human capability or were just not doable at all, like facial recognition or pattern matching or natural language processing or any of the number of things that cognitive technologies can help with, with machine learning and pattern matching and all that, there’s a lot of things within the enterprise that are directly applicable with AI technologies. We call these augmented intelligence approaches because they’re not meant to be replacements of human capability or human labor, but they’re meant to basically be help humans and help us do either our tasks better or faster or do things that we may not be capable of on our own.
So what is augmented intelligence approaches involve whether it’s applications of AI in things like fraud management and cyber security or it could be enabling a much broader range of conversations through conversational interfaces and chat bots. Or highly … Or individual mass personalization where a company can actually have a very specific trained knowledge set on each customer, on each particular customer’s behavior so that it provides very personalized interactions. Or just giving people a much faster ability to query very large data sets, extract information that they would not be able to otherwise because you can’t just do a SQL query or an Excel search to find patterns like that. Or to be able to interact with what we call unstructured information, like text and images and video and extract information from that using AI capabilities that are able to identity things within that.
Those are all immediate opportunities. And you may not even think of them necessarily as AI because they don’t seem like robots that you can walk and talk with, but it’s every much the bit of artificial intelligence that companies are looking for.
Guy Nadivi: What is a cobot? And how does it have the potential to disrupt some of the concerns people have about artificial intelligence?
Ron Schmelzer: Yeah so it’s good that you bring up the term cobot. So cobot means collaborative robot. It’s actually kind of the physical version of the augmented intelligence thing that I was just talking about. So a cobot is basically a robot that’s meant to operate in very close proximity with a human operator and actually with them, right. The traditional challenge with industrial robots is they’ve been … they’re large and bulky and, therefore, they’re dangerous to be like in close proximity. You’re not meant to be anywhere near that swinging arm. Right. But a few decades ago actually companies in Germany and Japan and the U.S. came up with this idea of well what if you can sort of have a little personal robot with a single arm or two that you basically train by actually moving it around. So you move that cobot around in physical space and you tell it what to do and then it repeats that movement. It learns and it knows where the human operator is at all times.
Then basically a human can accomplish more things. If you’re trying to, if you’re working on some sort of assembly process rather than doing it one at a time, you and your little legion of cobots can multiply your activity by 20 or 30. So they’re all doing that at the same time. Or you can lift a 400 pound load by yourself when you couldn’t otherwise. And it really, a cobot is this augmented intelligence device. Here we have the example of robots being used in a manufacturing situation, but they’re not replacing humans, which you’d think would be the role. Oh we’re gonna replace a human with a robot. And certainly that does happen, but in this situation it’s not happening. What you’re really doing is your multiplying human productivity.
And that’s the interesting paradox. You know if you take a look at Amazon, as you know they fully automated their warehouses. You know those tiny little Kiva bots are messing around those warehouse facilities and picking and packing and doing all sorts of things so that humans don’t have to go in there with a forklift or ladder or whatever they have to do and pick that stuff up. But the interesting paradox is that Amazon will tell you that they’re hiring more people than ever at their warehouses, which is a really strange thing. How is that they’ve increased automation and they’re using intelligent robotics, but they’re also increasing human labor, the actual employment?
That’s because what Amazon realizes is that you can make these humans do higher value activity rather than just sitting there and pulling something off a shelf and shoving it in a box, you can have them deliver stuff within one hour. You know deliver, do the sorts of things that they promised Amazon Prime provide value that none of their competitors that can provide. And so now instead of just doing whatever everybody else is doing, just maybe with lower costs or higher efficiency, they’re actually doing things that their competitors are not doing and sort of forcing everybody to play the Amazon way, which I think they’ve proven now as one of the most valuable companies in the world, maybe I think it is now, this is what the power of taking the robot out of the human as one person put it and allowing the human to do what humans do best.
Guy Nadivi: Do you think AI will ultimately have a bigger impact as a tool for automation or a tool for worker augmentation with things like cobots?
Ron Schmelzer: Oh yes. Seems like more it’s both. Right so. At the same time, you know, one of the great things that AI does, just because it provides cognitive technology that allows us to automate things that were previously not automatable. You know for example, it would’ve been impossible if you had hours of surveillance camera footage to say go through all this footage and find every instance of a white box truck that shows up through the frame. It’s like you would have to have a whole army of people just going through and looking through that video and it would’ve taken a long time. Now, as you know, with the power of image recognition and video … and facial recognition, all that stuff you can just train a computer with AI to recognize what a white box truck looks like and it can zoom through all those pieces of video footage and find those things almost instantaneously once it’s been properly trained.
So this is clearly a task that was not automatable before that becomes automatable, it’s not replacing human labor except in the situations where you might actually want humans to do all that sort of stuff. So in that case, yeah, cognitive technology is automating things that have not been automated before, filling in those gaps.
On the other hand, to the situations where we’re not trying to automate we’re trying to increase productivity, as I mentioned with the cobot example or the other augmented intelligence examples, or AI enhanced search or natural language generation, people who are using AI to create stories, you know summarizing quantitative data in a qualitative way. These are things we’ve never done before. And so AI’s really helping us augment sorts of things we’re doing, creating all sorts of new opportunities across a very wide range of industries from education and finance and manufacturing to retail to healthcare to automotive to E-commerce, it’s just AI has a very broad application.
Guy Nadivi: You half jokingly mentioned the Terminator earlier which makes me wonder what are some of the most unrealistic expectations currently plaguing AI?
Ron Schmelzer: Yeah so, I mean, there’s a lot of I’d say unusual concern. I don’t wanna call it fear mongering, but it’s like a lot of concern about so called AI super intelligence. And I don’t wanna downplay it because there are folks that say well there’s a possibility if these systems truly, you know, if we can reach certain breakthroughs in learning and these systems can achieve some sort of behavior then they can have runaway levels of control and we should be scared of that.
But I’m like, really … First of all, these systems are not evolving the way they would like to evolve them. I think it’s much harder than people expect just to have them do simple tasks. As I said before, if you spend any time talking to any of your voice assistants you realize they’re pretty dumb, they’re not really learning at this exponential rate that you would expect. So I think unfounded concerns about super intelligence are just getting in the way. And I find it odd that folks like Elon Musk have these concerns, especially when he’s using AI you know in the products that they’re building. He has Autopilot and then this Tesla. Clearly if he was concerned then the Tesla Autopilot would at this point be driving people around without any human intervention. But as we can see by all the accidents that it’s having, well or just a few, that it’s clearly not at that level where even Tesla Autopilot can be trusted to drive somebody from point A to point B. So I think the “robots are taking over” scenario is quite a bit over blown.
So that’s on the fear mongering side. I think on the other expectations is that there has been some conversation about so called pseudo-AI. We actually wrote a good piece about this in our newsletter. We also did a podcast in our AI Today Podcast that we have. By the way, Cognilytica we produce a podcast of our own and I encourage all of your listeners to subscribe and listen where we talk about adoption of AI in different industries. And we talk a little bit about pseudo-AI where companies who are usually venture funded, who are tackling some problem with AI. For example let’s say expense report processing or calendar scheduling and saying their using an AI system. A couple, I think in 2016, Bloomberg and The Guardian did a little exposé and found that some of these companies are actually using humans to do that activity, they’re using some Mechanical Turk or their own system to … So these are humans pretending to be machines pretending to be humans. That’s not a good thing.
First of all, it may violate some privacy laws and other regulations. But I think it’s showcasing that maybe the technology, even some of these very specific use cases, like invoice processing or scheduling, it may be even in those cases AI may be challenged to do it 100% flawlessly. So I think organizations and technology companies need to have some expectation that things may not be as perfect as they may, or as advanced as they may think it is. That the AI technology you know while it may be more advanced than it was before, you know it’s not flawless and there may still need to be a human in the loop. And they should probably count on it.
So those are some of the concerns right now for what people are … unrealistic expectations.
Guy Nadivi: Your assessment of the current state of AI brings to mind a quote from Einstein which may or may not be anecdotal, but artificial intelligence is still no match for natural stupidity. So we may have a ways to go yet with AI.
Ron, what are some of your predictions for AI over the next three to five years?
Ron Schmelzer: Well I mean if the current trends continue, I mean if you look at sort of the past three to five years and you kind of extrapolate and make a straight line, maybe not an exponential line, you’ll see that … Well first of all there’s a lot of data and I think systems are becoming a lot more powerful. Computing capabilities are just getting that much stronger and more efficient. And I think within three to five years, I think most organizations if they don’t already have some AI in use in practice will have it. It’s just becoming an inevitability that in order to compete in many of these industries, you know we just mentioned retail, but you know the same thing is happening in finance and in hospitality and in travel, in all these industries, I think most companies are gonna have to adopt AI as part of their core IT strategy and budget and plan.
The other thing is that it looks like customers are becoming more and more willing to, and preferring, to talk to chat bots. I know I recently had an experience where I just moved to a new house not too long ago and had to get my internet installed and it was like 3:00 in the morning or whatever and I messaged the internet provider, not expecting anybody. But I had a conversation with a chat bot, scheduled the install, verified prices, did all this sort of stuff. And I was pretty happy. I wasn’t 100% sure if it was completely automated or if it was a human behind it, and I think that was part of the interesting thing is that they abstracted the human, if there was one.
So I think in the next three to five years, most organizations are gonna be expected to have some sort of 24/7 conversational agent chat bot, some sort of assistant. Because to not have it is like the same disadvantages to not having a website or not having mobile app or some access or whatever you need. If you’re a bank for example, you don’t have a website or a mobile app, what century are you in? I think if you’re a bank three to five years ago and you don’t have … sorry, three to five years from now, and you don’t have like a chat bot or a conversational agent, I think people are gonna also be wondering what century you’re living in.
So those are some of the predictions. There’s other, but, I think at the high level.
Guy Nadivi: I guess if your ISP provided a chat bot that you couldn’t tell if it was human or not, they passed the Turing Test, right?
Ron Schmelzer: Yeah, there’s actually some discussion about that. Not too long ago Google had an interesting demo of something called Duplex which was this voice assistant, it was like a demo, it wasn’t meant to be a product. But they demoed it and it caused a lot of chatter because it called and made a hair appointment, a hair salon appointment for somebody. And it was a voice assistant, so it was completely fake, well not fake, but it was like it was completely AI driven, conversational. And it had all the conversational mannerisms of a person. It would say “hmm” and “ahh” and like go “well, I think”. So the person on the other end of the line, the person who was booking the appointment, didn’t know if it was a person or a bot and they were having a conversation as if they were having a conversation with a person.
So of course the first thought was like “hey, did this just pass the Turing Test?” Did this just, you know, did we just inadvertently cross some sort of boundary? And there’s been a lot of chatter as to what the Turing Test really means and is that a good test of intelligence and was this really scripted. And is setting an appointment really the same kind of random conversation that if the person on the other line decides to veer off script and talk about the fires in California for example, what would the bot have done? So, you know, there’s been a lot of conversation about what this really means.
We actually, we wrote a piece on the whole…what are the implications of the Google Duplex demo.
Guy Nadivi: If you’re in IT now, or thinking of going into IT, what areas of AI offer the greatest career growth potential?
Ron Schmelzer: Data science. So if you’re thinking about, if you’re in the … If you’re kinda looking for a career jump, a career change, or you’re just entering the workforce or you’re thinking about going to school, the biggest demand right now is for data science. And data science records all the things, it’s not even AI machine learning-specific. It’s just dealing with big data and understanding how to give organizations the power to use that data independent of the systems from where that data came from. So that’s the biggest opportunity right now if you were gonna pursue some opportunity.
The other thing there’s a lot happening of course in autonomous. So if you’re sort of hardware oriented, if you’re technology oriented, think about what’s all the various different autonomous systems. It’s not just autonomous cars, there’s a lot happening everywhere else in autonomous whether it’s trains and ships and boats and planes and drones, a lot’s happening in drones. But there’s also autonomous on the software side, right, what we so call intelligent processes, intelligent process automation. Something that we’re covering in our research increasingly called autonomous business process, so we really do mean autonomous, as in self driving. You basically can create technology that will identify on its own what the business processes are and know what those flows are like and know what’s normal and know what’s not normal without you having to explicitly go in there and either record about or code something, or something. So that’s sort of a desired end state goal that we’re spending more of our time looking at.
But these are all great opportunities. Of course there’s a lot happening in conversational interfaces as well. The voice assistants and chat bots. So if you’re looking at AI as a career opportunity, start with the data first, then look at some of the applications of AI and figure out where you can help fill some of those gaps. It helps to be a PhD researcher, but that’s a whole other story.
Guy Nadivi: Last question, Ron, what one piece of advice do you have for CIO’s, CTO’s, and other IT executives considering or already deploying artificial intelligence in their environments?
Ron Schmelzer: Well if you’re a CIO, you’re a C level and you’re in charge of whatever the mission is for your organization, obviously you have to start with what’s the business problem you’re trying to solve, right? And so the question is, just like every technology way, whether it was the internet or the mobile or Cloud computing, you know technology for technology’s sake rarely makes anybody happy.
So you have to figure out what is your job. Is your job, if you’re a bank, are you trying to … you’re trying to satisfy your customers better. So you have to think, well what cognitive technologies can we put in place that will basically satisfy our customers better, provide better return to all the stakeholders, lower our risk, increase our efficiency? And then you can look and say okay, well we can use chat bots this way, we can use predictive modeling for fraud this way, we can expand our features set, maybe do somethings for our customers we’ve never done before. That’s the best way to think about it. Is like to really look at the problem areas, gaps you’re trying to fill, and how this technology either enhances what people are doing now to basically make what people are doing more effective, more efficient, more productive. Or are filling gaps that have not been doable before. Either because it would’ve been too high labor to do it, or because it was just not technically possible.
That’s a great way of looking at any transformative technology, whether it’s AI, emerging block chain, whatever that’s gonna be, any new forms of internet, mobile, and Cloud computing technology. Autonomous vehicles is another thing. So I think just going from the problem and going from the opportunity and working your way back to the technology I think for all good C levels is just the best place to start.
Guy Nadivi: Alright. Looks like that’s all the time we have for on this episode of Intelligent Automation Radio. Ron, thank you very much for joining us today and sharing your thoughts on the current state of AI. I’ve really enjoyed having you as our guest.
Ron Schmelzer: Well, thank you, I’m thrilled to be part of it and hope all the listeners got a lot of good value out of this one.
Guy Nadivi: Ron Schmelzer, Managing Partner and Principal Analyst for Cognilytica. By the way his firm puts out some really outstanding podcasts of their own, as he alluded to earlier, as well as other content like great infographics about AI in different industries. Be sure to visit their website to check them out.
Thank you for listening everyone. And remember, don’t hesitate, automate.