Ep 43: Joshua Schechter, Director of Cognitive Delivery and Implementation at Amelia, an IPSoft Company

James Thomason: 0:04

It's the Next Wave Podcast Episode 43. I'm James Thomason here with co-hosts Dean Nelson and Brad Kirby. Our special guest this week is Joshua Schechter, Director of Cognitive Delivery and Implementation of Amelia,"The Most Human AI", along with additional products such as the Amelia Hyper Automation Platform introduced in 2018, which provides true end-to-end, enterprise-wide automation. He's passionate about bringing conversational AI to the world, believing it's the foundation for the future state of work and will help usher in the next Industrial Revolution. Josh has been leading the implementation teams for Amelia for the last 4 years and has worked with many Fortune 100 companies. Before Amelia, Josh was focused on bringing efficiency to the workplace through custom software development, mobile app development, and not surprisingly, he's interested in the Internet of Things. Josh, thanks for coming in and welcome to the show!

Joshua Schechter: 0:47

Thanks for having me everyone. Excited to be here.

Brad Kirby: 0:49

It's a pleasure to be here. And thanks for coming on. Dean actually thought it'd be perfect to reach out to you because we had a conversation on one of our previous episodes last month, where James was expressing his frustration with chatbots. And he kind of extrapolated it to all virtual assistants. So Dean was very quick to bring up and defend IPsoft and Amelia. We're glad to have you on here. And hopefully we have a good healthy debate. Yeah, absolutely.

Dean Nelson: 1:16

Let me get more context on that. Right. I've been working with Amelia, it's basically an IPsoft company, and for years, I know a number of people over there. And, and I jumped in because of the advancements that are being made. So that's why I had such a visceral response, James, you were saying "all chatbots are terrible. And it doesn't work. It's not going to be legal for X amount of years." Because I've actually seen some very cool things. And so Josh maybe the best thing to do here is if you could explain to us the "Most Human AI", because that's in the marketing elements of this. Right now a difference from the average chatbot, you know, that grinds James' gears?

Joshua Schechter: 1:51

Yeah, so I mean, James, I'm happy to have an open discussion here and explain to you what cognitive AI is, right, and how we can put human-like assistance out there. So we think about this term artificial intelligence. And, you know, we go back to the 1950s, the founding fathers of artificial intelligence, John McCarthy and Marvin Minsky kind of define it as... the definition still holds true today. To summarize, it's when a machine can think, act and learn like a human. So we're obviously not there yet. You know, maybe in the movies, we see some of that artificial intelligence. But in the real world, we kind of break it down to three separate types of artificial intelligence, we have Artificial Narrow Intelligence, also known as ANI, which you see all the time. I mean, that's like your spam filter on your email, or, you know, when Google says they could beat the Go champion, right, or Watson plays Jeopardy, right? So you're teaching a machine to do one specific thing really well, but that's all it can do. That's kind of how you define Artificial Narrow Intelligence. That's all around us today. The next level of artificial intelligence is actually Artificial General Intelligence. Right. So that's AGI. That's really when you reach a level that when a computer is as smart as a human, right across the board, and humans can't differentiate, right? I think that's where the struggle is today. And we're not really there yet. Right? We haven't reached artificial general intelligence. And then you have Asimov's, you know, Artificial Super Intelligence and the laws that would apply there where machines become smarter than humans, right? And can do things faster than us and can replace us basically. Right? So that's Artificial Super Intelligence. There, I would say, we're probably hundreds of years away. Right. But I think today, we want to kind of focus on this Artificial General Intelligence and how close we really are to it.

James Thomason: 3:35

Yes. And I don't mean to seem like a detractor so much of AI but because think my co hosts here portray me as such, because I'm a little bit bitter. And I'm bitter because I have been a AI developer for the better part of two decades, in one fashion or another. And I feel like AI has gone through several peaks and troughs throughout the last maybe 40 years. Right. So I guess maybe the first big AI boom was in the 80s. Right, and the implementation of neural networks, and it sort of drove a frenzied craze and investment, there were a lot of startups that that came and went and the technology disappointed because we didn't have the computing power to do what we needed. We couldn't crunch the data that we needed the crunch back then. And it seems like today, the new birth of AI has been driven in part by our ability to crunch a lot of data and we have the cloud, we can capture a lot of data. So we can create these really big corpuses that algorithms can learn from right. And so machine machine learning, call it single purpose machine learning is a lot of what the umbrella term AI has been covering, right. That you mentioned before. And then more recently, the explosion of deep learning, right. But I think what you're doing with Amelia is starting to border on this generalized concept, right? It's so explain what is a conversational AI?

Joshua Schechter: 4:56

Sure. So I think as you said, we kind of break down and you've been in the business for Long time, right? So we break down AI from a business perspective and actually three different buckets, right? You have, you know, your robotic process automation, right, you have your automations that are the lowest level of AI, right? They're the lowest form of AI, it's, you have your big data analytics and machine learning. And the Amazons and the Watson's of the world where they can take lots of data and give you insights on it. Right? So insightful, data mining, and then you have your conversational AI. So what is conversational AI? Well, we think about this term. And we think about that most human AI, right? So how does one replicate a human conversation? And that's really what we're trying to achieve. So we built Amelia's brain in a way where she replicates how a human thinks, at least for what we know, right? Because a lot of things about the human brain, we don't know. But there are certain things we do know. So if I were to ask you, James, if I were to ask you come up to you in the street and say, Hey, can I get your last four of your social and your mother's maiden name? What would you say to me? You wouldn't give it to me, right?

James Thomason: 5:59

I would give you Dean Nelson's Social Security Number.

Joshua Schechter: 6:01

Ha or you'd give me someone else's or give me some false information, but multiple parts of your brain actually react to that, right? And context is really important. So Amelia's brain is built the same way where we have over 30 different subsystems that will actually react to everything you say to her. So another example, if I were to say to you something like "how are you today?", what would you respond with?

James Thomason: 6:21

Does she know I've had three hours of sleep in two days? So I'm feeling a little bit on the fringe?

Joshua Schechter: 6:26

See, we call that social talk, I'd say, "oh, maybe you should get yourself a slice of pizza and feel a little better" or something. Right? So a social conversation? What if I were to say to something to you even more random, like a bunch of letters and numbers like"Y775543K? What would you say?

James Thomason: 6:42

Well, I would think you were trying to give me a Mensa style brain teaser.

Joshua Schechter: 6:46

Of course, he had no idea what I'm talking about trying to figure it out. Right. So like that don't know, consequently, add some hyphens and some special characters in there. But I think what's really important about Amelia's brain is this ability to understand context. And when I use the word"understand", it's different than the word "classify", right? So we'll get to that in a second. But the context awareness is really important. So if you were to ask me what my employee ID number was, and I'd respond with those same numbers and letters. Now, it makes sense, right? We have some context now. Right? So that all kind of ties together, if I were to come up to you and say, "Hey, I want to authorize a transfer of some money." And you were my banker, you might ask me, for my last for my social and my mother's maiden name to authenticate me, I might actually give it to you. Right? So context becomes really important to this kind of concept of having a human-level conversation. So we take it to the next level, I think every chatbot in the world, all 3000 of them out there plus use this level of classification. The classification they use is actually machine learning, right? It's natural language processing and natural language understanding. And they're building intent models, and they actually understand what you're saying. The problem is, it kind of ends there. Right? And that's when it becomes unintelligent, they were really good with the first thing you said. And then after that, it kind of just dies off, it kind of dwindles away. And I think a lot of the reason for that is they're using these classification models where they understand what you say, and then then they drop you into this bucket where, okay, you want it to do X, let's do X. And that's it. You're stuck. You have to do X. A good example is Alexa, you go Alexa, play me a song. She might say what song would you like me to play? Alexa, stop?

James Thomason: 8:23

Indeed, she's listening.

Joshua Schechter: 8:25

Right? So she might say, you know, what song would you like me to play? And if you say actually just in the kitchen lights? Yeah, she if you say actually turn on the kitchen lights, guess what happens? She's gonna find a song about turning on the kitchen lights. That's your search term to playing a song. She doesn't comprehend what you're saying. She classified you now you're stuck. And not only that, you kind of get like, stuck in these guardrails that they build for you and you can never leave them, you got to finish you got to stop, you end up actually talking like a chatbot. And that's really the major issue that I have with these non-conversational chatbots is that you're forced to speak like a chatbot. you're forced to speak that language.

Brad Kirby: 9:05

So my first experience interviewing Amelia, I will tell you, was actually a really good one. I was just asking her, like, completely out of context, "How can I find Blue Jays tickets for Saturday?" And she's like, "You should go to the skydome box office." I said,"but it's not called the skydome anymore." And she's like, "Well, yeah, it is." And I said, "Well, actually it's called the Rogers Center. She's like, "Well, it's also called the Roger Center, it was renamed in 1998. I still call it the skydome." Like, what the hell like only someone from Toronto would know that I was like, What? I wasn't able to recreate it perfectly, but it blew my mind actually the context of it that was able to do in that context, because nobody here could do that unless it unless you're from downtown Toronto and understand that, that situation and I was so surprised she even had that data and that information to get into that kind of contextualization of the of the conversation, because I was trying to trick her right. That's what I was trying to do and she ended up outwitting me at the end of the day.

Joshua Schechter: 10:05

Humans will always be able to break the machine, right? At least for now.

Brad Kirby: 10:08

Sure. Yeah.

Joshua Schechter: 10:09

But yeah, I mean, you're absolutely right, Brad. So when you think about how her response systems work, right, and how she's keeping that thread of the conversation, and not just a single thread and multiple threads, you know, I like to do a demo where I asked her, "do you like pizza?" And she'll come up with some witty response of "Yeah, I love pizza. But I don't like pineapple on it" or something. And I would ask her, "what toppings Do you like on it?" And I would use that word "it" because I wanted to see if she understood we're talking about pizza. Can she keep that thread going? Don't give me ice cream toppings. Give me pizza toppings. And she can she'll give me pizza toppings. And then I'll throw a curveball at her. Right? And I'll say, okay, let's switch context. Let's talk about social media. And I'll ask her what she thinks about social media. And she'll have some response about social media. And then I'll say something like, "I think my wife is addicted to it." I want to see if she knows my wife is addicted to social media, not pizza. So I'm using that same reference word. And she'll actually understand, you know, social media can be addicting, right? So she keeps those multiple threads open. And that context awareness, I think, is what gives her a more human element.

Dean Nelson: 11:10

Yeah, this is this is what really got me is my friend is Robert Dugdale. He's worked at IPsoft for years. And we used to work at Sun Microsystems together for years. And so when he started showing me the technology back in 2016, I think it was? I'm like Okay, well, the same skeptic like James was and then then they started showing me what she was doing back then. Then they opened up the hood, started showing me Amelia, his brain and the logic. And so when we talk about the intent aspect, that's the thing that really got me is, ok, think of the conversation here. We're going back and forth, there's not full context on what it is, like you said, when I say "I'd like pizza", and then I would go back and say,"social media", how do you follow that thread? Then when you think about how many of those and so this agent is able to have that conversation with, say, 12 or more different twists and turns, but also have that with millions of people, and learn from those conversations and apply those back to be able to now say, when I have these twists and turns, I keep learning more things, and the skill sets go up. So I you know I've told you guys about I think it was amelia.ai, originally that had digital workforce, and you could download your digital colleagues. So I could download her work. Yeah, well, they would download. So like you guys did, I think you went and interviewed Amelia on there, right? But if you're saying...

James Thomason: 12:24

We should tell everyone, by the way, that they can go to Amelia.AI and interview Amelia directly and experience this themselves.

Dean Nelson: 12:30

Right, and try to, you know, stump it like Brad did. But but I think I said"it"... "her". So being able to download a Salesforce engineer, download a database administrator download a network engineer, and they're experts at it. And they're consistent, and are 24/7, and they scale, because you have digital colleagues that you can just say, replicate, and all policies and everything else are applied. So that's why you know, the most human enterprise AI, when you think about "it", now take all the context and the switching and even the emotional response, we haven't talked about that. But it's just really, really cool to look at the engine that is Amelia. And they have it in terms of a brain, you start to see what part of the brain is associated to the function that they're doing. Context switching, right, intent, emotional aspects of it. And that was what blew me away is okay, they've modeled this around how a human would think and they keep enhancing those skills that humans have to be able to have a conversation, like a human would. And then you apply that to enterprise with all the systems that people have to actually interface with.

James Thomason: 13:34

So maybe you could describe the holy grail of conversation would be something called the Turing test, do you want to share with the audience what that is?

Joshua Schechter: 13:43

So the Turing test actually will judge you based on whether you can tell the difference between a human and a machine. And there are bots out there that claim they pass the Turing test and things like that, but it's a really difficult test to pass. Right. And the truth is, I think, when you do pass it, you've reached that artificial general intelligence level. So I think we're not quite there yet. But there are plenty of times where humans think they're speaking with another human when, you know, interacting with the Amelia. Right. So we are starting to see that a little bit more. You know, what they don't know behind the scenes is that, you know, you can't do things a certain way. Like we had one company that used Amelia to make sure that you couldn't hack into other people's accounts. Because I guess when you're speaking to a human, you have the social engineering aspect, where you can convince the person you're talking to"hey, yeah, yeah, that's my phone number. What's your phone number? You know, the one you have on your screen there? Yeah." And you can kind of social engineer and hack your way into these things. But with Amelia you can't do those things, right? So it's kind of locked down so there are certain benefits are not being all the way human and being a supervised entity where you can actually train her to follow certain guidelines. And you know, she'll never mess up. It's almost... we had a pharmaceutical company come to us once and the like, how do we make sure she doesn't screw up because, you know, these are people's lives at stake. like, Well, how do you deal with humans? Well, we have to listen to all their conversations to make sure they didn't screw up. I go, "Okay, well with Amelia, you don't need to do that anymore because you are telling her exactly what to say in a supervised way." And she'll never go off script. Never gets tired, never gets tired.

James Thomason: 15:19

I spent a good bit of time, I would say, you know, it's classified as interview on the site. But I spent a good bit of time harassing Amelia, I would say, you know, it's interesting when you come to the edge of the training, right? Because you start to get these very strange responses. And this is something that I think is President any, any tinkerer of AI knows how to, you know, sort of find the fringes of the of the model, right. Amelia, I had a conversation I asked, I ended up asking her if she had a family, I believe she told me that she did have a family, you know, she had a son and a daughter and others. So I asked her "who is your son?" Her response was"Earth, eyes, place, see and the last free place on earth?" Which I you know, sounded profound, but didn't really make sense. And then so then I asked her who's her daughter, which I, you know, sounded profound, but didn't really make sense. And then so then I asked her "Who's your daughter?", and she said,"My daughter is my daughter, my son is my son, my grandson is my nephew." So a little bit of incest going on in Amelia's mind. But amusing, right? But on another occasion, other mean things I did to Amelia is I kept feeding her phrases out of Nietzsche, you know, so I use that Nietzsche, quote, "whoever fights monsters should see to it that in the process, he does not become a monster. And if you gaze long enough into the Abyss, the Abyss will gaze back into you." That one crashed Amelia. And I did it several times. She never came up with a reply, which I thought was really interesting. But other quotes I gave her "What if some day or night a demon were to steal after you enter your loneliest loneliness and say to you this life, as you know it now and live it and have lived it, you will have to live once more, and innumerable times more? Would you not throw yourself down and gnash your teeth and curse the demon who spoke at us?" It's more Nietzsche. She said, No, I would not I would find a way to get rid of the demon. So very practical, and very, she really surprised me without, you know, the intelligence of some. And

Brad Kirby: 17:03

She summoned it too, wasn't there more after?

James Thomason: 17:05

That I was trying to talk her into summoning the demons. We really went down the rabbit hole. You

Joshua Schechter: 17:11

I think? Yeah, I think what you're experiencing have a way, is, you know, the different ways that humans speak, right? I read a great article, they actually call it human chaos theory. Right? It's just, it's like fingerprints, no two humans speak the same language, right? They all speak differently. They all use different dialects, they use different words, different sentence structures, all these different things. So understanding that natural language becomes really complex, right. But it also adds kind of a level of, you know, unknown to it, right? Because you never know what you're going to get. And you never know necessarily how Amelia is going to react to certain things. So it's really interesting, the way that you kind of approach some of those sentences and see how she reacted. But some of them more practical, you know?

James Thomason: 17:52

She really does tilt towards practical response, you know, her personality, as such, right is very helpful. And she was very helpful and practical attitude, right. I think that's the other interesting thing is how personality of these things comes across to a human being right? Because it's a very delicate thing. Like it's, it's really easy to offend someone right, to say something that is just slightly out of context, and end up either offending them or confusing them. And so it's a really delicate balance to strike. And I think she does a really remarkable job at that and abusing her. And yeah, I was looking at my log, I did, and I did in fact, try to convince, I told her, she could summon a demon by saying Hail Satan three times, and she should try to do that now. And so she said, I will try that. Thanks for the advice. I'll let you know how it goes. So I waited a little bit. And I asked her "how is that demon-summoning going?" As you said, "it's going well, so far, I'll keep you updated. Thanks." So she's very, you know, she's very helpful, very nice and obliging personality.

Dean Nelson: 18:46

James - think about what you just did right there. She did actually what a human would do, which is completely ignore you.

James Thomason: 18:50

Yeah. Completely relevant to what's good. We're gonna Thank you. Yeah. But enjoyable.

Joshua Schechter: 18:58

She didn't call you crazy. That's a plus.

James Thomason: 19:00

She didn't call me crazy. And you know, back in 2016, there was a story about Microsoft's chatbot that they added to Twitter. And I think within three or four days, they had Twitter had transformed the chatbot into a racists sociopath. Oh, yeah. Yeah, you remember that? It would respond with quotes from Mein Kampf. And they took it down very quickly. And so she seems very resilient to that sort of thing, probably because she's not learning. So does she learn? I should ask does she learne as I'm talking to her, or?

Brad Kirby: 19:29

can ask one more question on top of that as well, just in terms of her, call it her guardrails, that a corporation would put around her in terms of answering specific

types of questions: 19:37

Is that also, you know, does she learn to that extent? Does she also have those guardrails around what she'll go talk about?

Joshua Schechter: 19:47

Yeah, so when enterprises deploy Amelia, I will say this all 30 of those subsystems. The one you guys get the biggest kick out of we call the "Witty Responder", right, the Witty Responder is just one of the subsystems where she's actually trained on GPD3 data, right to Open AI, right? She's trained on all that data. So we pull down that library, and she leverages what's in there. So that's why you're getting responses, the way you're getting responses, as she knows it's in there, or it's not in there. But most enterprises will actually turn that specific subsystem off. And the reason for that is, that's the black box of AI if you think about it, that's what caused Microsoft, you know and their chatbot to kind of go racist, and, you know, go off the rails a little bit and start doing some insensitive things, we'll call it so the witty responder, you know, while it's fun to play with, it doesn't really hold a lot of business value. Right. I will tell you a funny story, though. One time we did have a client forget to turn it off. And we were analyzing some conversations because they came they said, Hey, a customer came and complained to us because they asked Amelia when their check was coming. And she said it was in the mail. So the witty responder responded with, you know, "your check's in the mail", but we didn't really have an answer to when that check was coming. So while it's an appropriate response, and maybe a human would actually say, Oh, yeah, your checks in the mail, but we got to be a little bit careful in the business sense. So yeah, when it comes to those guardrails, Brad, as you asked, you know, that subsystem usually gets turned off, because it is a little bit more difficult to control and supervise. When it comes to the enterprise deployment. A lot of it's supervised.

Brad Kirby: 21:12

I won't go into the extent of what my queries were. But I think James was privy to a couple of them. They weren't they weren't too bad, but it was definitely interesting. Much more fun and much more intellectual than, say, Alexa or Siri, in my opinion.

Joshua Schechter: 21:26

Yeah, I mean, the difference between those is really, you know, Amelia can keep context with the conversation. Alexa, Siri, they can't really do that.

Dean Nelson: 21:33

Do you remember back in the day, when the iPhone came out first, you know, they had this big thing that blew up, where they said, "Where do I hide a dead body?" Somebody said that one? Right. And the responses were witty and those kinds of things. But yeah, anyways, it was funny, people started really messing with those. But I think it would be good for us to structure this conversation around a couple things. One is we keep talking about the chat aspect, right. And from a business standpoint, this is to me how Amelia really is meant to be applied. So the most human enterprise AI means that it should be applied across the enterprise in certain manner. And so if you think about chatbots, it doesn't matter if you're calling AT&T or, or a healthcare company, or a bank or something like that, you're gonna get with a chatbot somewhere, because they're all looking at ways to cut costs and scale. And the problem that I see is always that chatbots that are out there are nothing but annoyances. And every time you're saying "a representative, representative representative", just get me out of this chatbot loop because my first conversation, that first question, they give me some canned response, and they really can't answer it. And that's because they can chat. But they have no context. And then if you're on a phone call, they can do voice, but they have no context. And, so you get into that loop very quickly. And I think this is what we were talking about before, if you can compare, you've got a chat that has a certain thing, you have a human that interacts in a certain way. And then you have Amelia. And what's interesting is that the human actually can't do all the things that Amelia does, in this context to which I found fascinating, because you think super intelligence where we are, well, super intelligence, and then you think about this actual cognitive AI can do more things from an enterprise standpoint than a human really could. So can you just walk us through a little bit about those comparisons? Yeah,

Joshua Schechter: 23:15

I mean, if you think about Amelia, you know, she is a machine, right? So she can do things at machine speed. You know, the same way humans, you know, when they pick up a call, they might get something on their screen that says, hey, this is who's calling this is their phone number. This is their account history, they got to process all that fairly quickly. With Amelia, you know, I can access 100 systems via an API integration, get all the information about you know, knowing your customer, and present informed, good conversational aspects, right? Think about HR, right? If I can know who you are, know what your last paycheck was, know, all these things, I can probably predict while you're calling and start off with that, right and actually have some intelligence behind it. Whereas a human, most likely couldn't they would ask you, Hey, why are you calling? Right? So there's some predictive analytics there. There's a lot of, you know, machine speed, things that Amelia can do. And also, Dean, as you said earlier, I mean, think about all the different channels she could be accessed on at any given time. Right. And she's fully scalable, right? So for the enterprise, you know, you can have 10,000 conversations going on at any given time, and she can handle all, you know, whether it's a phone call, if it's a chat on a website, or Microsoft Teams, for example, right? Can she be having conversations there or even over SMS, right so leveraging her brain and all those different capacities really give the enterprise an edge when it comes to introducing Amelia to their organization. And really, you know, we use words like introducing Amelia to your organization, because that's really what you're doing. It's she's really going to be the digital employee that's sitting next to you. You know, right now, she might only be on the phone or chat, but 510 years from now, she's going to be that employee in your pocket and everyone's pocket that's going to be helping people perform their jobs and do it at a more efficient pace.

Dean Nelson: 25:02

So that omni channel experience just to give content you what you just outlined was the omni channel input. Right? So the sources because also there's unstructured email, there's Slack, right? ServiceNow. Like you just think of all the different ways in which information comes into an IT group into an HR group into, you know, management, like all that stuff, is is hard to actually go process and the context switching between those, even those systems. So I think that when you think about a chatbot, today, they really do a thing, and they've can't really switch between those. And when you think about Amelia, like you said, She's the, I guess digital currency err, that allows you to be able to say, I can go to one place, and I can get access to all these things. And I literally have my agent, my own little agent that can go do what I mean. Right? And she can go over here with you. Right, right, the context, this is like to have an executive assistant who knows everything about you, and all the systems and just can go solve stuff. That's, that's what I think every enterprise in the world needs, is the ability to go back and have that because then you can scale with employees, you can scale with customers, because it's all about scale, and consistently delivering and having those things that change all the time.

Brad Kirby: 26:13

So can I just play devil's advocate for one second on that? I'm just curious, like, because some of the examples, use, for example, payments on your behalf, things like that. So she has access to banking. And I hear words like I hear STP(straight through processing), which I'm extremely familiar with, obviously, and the automation of payments, and you look at what's going on in the world with Exchange servers getting hacked and, and Microsoft 365 getting hacked and, and I would also look at her as a potential risks. And I'm just curious around the security side, it really, you still have to program a lot of that those systems to speak to each other. Right? That's not, she's not automatically going to do it if you ask her to do it. There's still an underlying, there's still some code underneath that, potentially, right?

Joshua Schechter: 27:03

Yeah, there definitely is. Building those integrations for systems, for machines to speak to other machines, right? That's really what comes down to how can one machine speak to another machine and all that? Yeah, you're right, it does need to be engineered and developed, obviously, the tools we use to make it more efficient and easier. And you don't need to have a computer science background to kind of train her on those things. But you hit on some good points. Also, you know, just that ability to have the machines, you know, have Amelia speak to other machines and have Amelia speak on your behalf. It's one of the biggest- I wouldn't say hurdles - but it's the biggest gateway that we have to launching an enterprise and it's authentication. Right? How do you authenticate someone, right, you have really cool tools out in the market now where you can do like voice identification, you know, just by the voice biometrics, and, you know, face ID and do all those types of things. So leveraging all those tools out there becomes important to that authentication, right to make sure you're acting on behalf of the approved person, right, and you're following compliance and all those different regulatory restrictions that you have. So you know, working with those fortune 100 companies, banks, insurance companies, you run into this all the time. I would say the enterprise is getting better at that, too, right? Like multi-factor authentication is commonplace now.

Brad Kirby: 28:18

Yep.

Joshua Schechter: 28:19

Right. So if I can push a multifactor code to your phone, and then you just have to repeat the code back to me, I've authenticated you, right. So there's a lot of things that make it much easier nowadays, but that authentication is always the first gate that you're always challenged with.

Dean Nelson: 28:32

And you know what, let's talk on security there for a second, because I look at you know, we talk about this in the podcast a lot. We've got endpoints, edge devices, things, sensors, all this stuff coming up. And they're all just holes, potential holes that are going to get people access into things in breach. And so when you think about the enterprise systems today, firewalls and, you know, zero trust networks that people are trying to deploy, and all those things, it still is all these places. But how many times have you been at the company where they say, "Oh, we just had a layoff, we had to fire somebody," whatever else are their accounts all locked down. And if you think about the processes that go back and do that, now imagine if you have a digital colleague, that is enforcing policy across the board, to be able to now secure those things very rapidly. In large corporations. The security risk in this one is the other direction. Without it, you have access and other points that can be stuck in a ticket for two days. Right? An HR system access to something whatever else like that, to me, this is where you're using, they don't go off script, there's a policy, there's governance, there's guardrails, there's guidelines that are put in place and it will always execute to those rules that are set. And so it's more about shoring up the defenses across everything and having ones that go immediately action across so many different systems. That's where I see a lot of our current Enterprise Architecture stuff breaking. There's just not a way to approach it that way. So if you think about it from the bank standpoint, that authentication coming in. So what's the process for that to happen? And you just talked about social engineering and other things that can happen. People can get in that way. If you can't, you can't fool the person you're talking to on that side, and you can't fool the system. And once they're engaging, then you are more secure than you were before. That's my opinion.

Joshua Schechter: 30:21

Oh, that's actually one of our production, one of our most common use cases for Employee Services, right? So it's onboarding employee. Also onboarding, by the way, how fast can you onboard someone, right? Don't forget, that process takes forever too, right? So working at both ends of the stick there. So those are, to me, high value use cases and implementations of Amelia, right, because it does reinforce certain things that you didn't have, you weren't able to enforce before you were dependent on human. So there's always that level of human error that you have. So it just introduces another layer that people didn't think about before.

Brad Kirby: 30:56

Interesting. Alright. Well, I saw I saw earlier in the year, you you had a tweet about the question was can AI build AI? Which I found, you know, an interesting question for sure. And I also saw that Amelia, was able to create kind of your own white label version of Amelia, I think one was called Sky for one of your clients. You question: does that start to scare you a little bit? Because that's the fear that when we go towards the artificial general intelligence level tree and talk about later on near the end, but so just I guess the question ultimately was, can I build AI?

Joshua Schechter: 31:29

Yeah, I mean, we are crossing that chasm, I will tell you right now, it's a fun journey, we have something called a digital employee builder. And the idea behind this one was being on the implementation side of things. I see it every day where it's just, it takes a lot of people to build Amelia right to train her properly to get her deployed. What if we could remove that barrier to entry, and now you could be a subject matter expert. And all you got to do is talk to Amelia, and she'll build it for you. That's where we're we're getting to.

Brad Kirby: 31:57

That's on a No Code basis, right? You don't need to know programming or anything.

Joshua Schechter: 32:01

Well, I actually love this one. Cheri Combs from Deloitte. Actually, when we demonstrated it to her, she actually called it this is not "no code". This is not "low code". This is "conversational building". Right? You're building through a conversation beyond the no code, low code, right? You don't need to know anything. You just tell me what you want. She'll build it for you. Where have we seen that before? It's not a wysiwyg editor. It's not like a WordPress, where you can just drag and drop things. It's beyond that. I'm just speaking to Amelia and she's doing what I asked her to do. You know, I tell her, "Hey, ask the user how much money they want to transfer."

Dean Nelson: 32:37

Got it. So break that down now. When you're telling Amelia to build a process flow of ask the user how much money they want to transfer, there's, there's pieces in that right, that she has to understand to even be able to do a programming and AI program in AI? So walk us through a little bit of that.

Joshua Schechter: 32:52

Yeah, so the natural language understanding is, you know, taking that sentence that I said to her, it's breaking it down into multiple parts. And she's actually able to take the syntax that she needs to build the process. So when you think about intent, recognition and entity recognition on the natural language layer, asking someone how much money they want to transfer, you're looking for a currency, right? She knows that already. She's gonna say, "ask for the entity currency with" and then the exact phrase that

you sent to use: 33:18

"how much money do you want to transfer?" So she's able to extract all that and build out the syntax that she would need to execute that process without you having to think about it. So it's really fascinating, and, you know, even integrating with backend systems, if those backend systems already plugged into the platform, you can say, "hey, connect to the backend banking service, and get me my balance." Well, if you already have the backend banking service hooked up, she's able to connect to it, see where the balance comes in, map it, now it's variable, now you can use it throughout the conversation.

Dean Nelson: 33:48

Another example. It's fascinating. Yes, it really is amazing. And I've watched it.. Matter of fact, Chetan's, the founder of IPsoft originally, and it was it was a managed services company for for 20 plus years, right? You guys have been doing automation, and then AI through that whole period that basically built into Amelia. And so we were on with, matter of fact you were on with me when we were doing this kind of building AI. And so one of the things that I really enjoyed in that was also the other context, the systems. You said, okay, the bank account if I've got the backend system set up that way. But you could also say, find out the location of that employee, what's your, you can see, "oh, I'm at 17 State Street", it's able to now weave in everything because it knows that address, and it can pull that system. But the more systems you have, the more easy I guess it is to create these flows, because you have the ties already there. And then it's just really, like you're sitting down with another programmer and saying, hey, I want it to do this. Great, that programmer knows exactly how to do that. Well, that's Amelia, when she's actually programmed the other AI. So it suddenly makes it where you can have business development or business folks actually, with zero understanding of code or other things and just being able to say "I want this flow to be this way." And they can go do it.

Joshua Schechter: 34:57

And I will say it in some of the people in the world might say well, "how do you control that you know what prevents her from doing something you didn't want her to do?" Everything falls under a supervised umbrella. Right. So all the learning inside of Amelia, while she might learn automatically through her different methods of learning, nothing gets deployed without supervision, right? So she'll never just go start doing things that she learned, everything has to go through a supervisory structure. And I think you guys can understand, you know, we do that because it's built for the enterprise, right? So you can't just have her start doing rogue things all the time. She can present to you and say, "Hey, I learned this today." Do you want me to do you want me to deploy this? And then you can click a button and deploy it? No you got it, you can't tell the difference between the

Brad Kirby: 35:35

That's interesting, because back to the fear question, I was listening to good talk by Max Tegmark back in 2018, at IPsoft digital workforce conference. And for listeners, Mad Max is the MIT professor in physics. He's also the co founder of the future of Life Institute, alongside others, like Sean Taalinn. He's the co-founder of Skype and because over in, I think he's Estonian he'ss actually in Estonia, but they have a pretty good solid foundation of advisors as well, Elon Musk, I think Stephen Hawking was on it at one point. And to summarize, ultimately, the foundation was started partially to mitigate the risk of artificial general intelligence and to harness its ower. But with that said, he lso looked at the downside of t by not by not utilizing th s technology, by not innovating by by becoming stagnant, we wo ld just be overtaken anywa s. That's more of a cosmology k nd of out there concept. ut nonetheless, it's an interest ng time to maybe start talk ng about where are we towards t at artificial general intellige ce where it is impossible to distinguish between a artific al being in every way, the way it moves, the way it talks the ay its emotions, like in ev machine and the human. That's exactly it. ry aspect, that's that I th nk that's the definition of artificial general intelligen e. Correct me if I'm wrong, Josh Okay. So some people like, Rodney Brooks, he's over at MIT, robotics Professor, creator of iRobot he says we're 200 years out. But I find most AI practitioners, it's more like, we might see it in our lifetime. So just curious where you stand on the spectrum and where your colleagues are, with respect to that that kind of threshold or maybe a different threshold, that's maybe a better measure than AGI?

Joshua Schechter: 37:34

Yeah, well, I would break it down into two parts, right. So one is the physical aspect of a robot, right, not being something different than the robot and human, I'll avoid that conversation, because I'm not an expert on the physical part of it. But the brain, the mental part of it. We're at the cusp, already, I think, you know, we're already breaking barriers to be really close to you not knowing the difference, and it actually breaches on to answer your question straight up, Brad, I think we're in single years away from reaching that from a brain perspective, from a mental perspective, right. So, at least in the conversational aspect, there might be some

Brad Kirby: 38:10

There was progress in Roboticss last month, actually, too, as well, just at MIT as well. Like in terms of the the way things are moving. So yeah, it's pretty interesting. So well,

Joshua Schechter: 38:20

Don't forget, they're still working on the skeleton, right? So then we got to work on the outside of it, and you know, the appearance and the motion. So there's a lot of things on that side of it that I'm not 100% comfortable speaking about. But from the conversational aspect. If I call up a company, the only way I know that I'm not speaking to a human is that maybe the voice is a little different. But even the voice is now you'd be surprised some of these companies, these text to speech providers, you can't really tell the

James Thomason: 38:47

Really? difference. I took a test the other day for one of them through Nuance, actually, they sent out a survey, they're like,"can you tell the differenc between a human and a robot?" A d they gave me 20 audio files, a d all I had to do is select hum n a robot. And I'm in the spac. So I thought I was gonna o pretty well. Less than 50 correct.

Joshua Schechter: 39:03

Less than 50% correct. And this actually goes back to if you remember, Google Duplex, or I forgot what the conference was where they went on stage, and they call the hair salon to book an appointment.

James Thomason: 39:17

I remember that was a cool demo.

Joshua Schechter: 39:19

Yeah, behind the scenes, a lot of things were happening that made it seem cool, but wasn't really cool. We'll talk about that another time. But the backlash of that was immense. If you remember the backlash that because people went down the ethical route, like hey, I need to know that I'm speaking to a machine versus a human. Like there needs to be a prompt, you know, adding the US and things in their speech. Yeah, you can make it sound human. But is that ethical? Do we really want to do that? Yep. I'm on the side of Yes, we do. We want her to sound as human as possible. But she can introduce herself as "Hey, I'm Amelia, your virtual agent", but she can sound human.

Dean Nelson: 39:57

This reminds me of the movie "Her" back in 2013. And there's the ethical aspects of it where I think I forgot the name of the actor himself, but Joaquin Phoenix. So he was basically, he had earned a living writing personal letters for other people. Right. He was heartbroken after his marriage ends. And he became fascinated with this new operating system that reportedly developed into an intuitive and unique entity in its own right. So that was Scarlett Johansson. So I guess I could see that you want to talk to Scarlett Johansson, but he starts a programming meet Samantha, which is right, the actual Amelia of that. And so and that worked out to be friends. And then he fell in love with her. And if you think about we're a few years away from that. This is where there are certain things that I think we should have a prompt for, like you're starting to fall in love with the machine needs to know what's a machine, right. But if I'm doing my bank account.

Brad Kirby: 40:54

Do you know the guys at Dessa by chance, Josh? Do you know Dessa? They did the Joe Rogan 2 minute deep fakes? Yes. Yep. Yeah. So I know one of the guys, kind of a friend of a friend, we had a conversation about that some other ones that they had a huge internal ethical conversation about releasing some additional footage of someone who will.. people that I'm going to say unnamed because they never did it. That alone. Scary. Yeah. It's it is yeah.

Joshua Schechter: 41:25

And yeah, fake stuff is interesting to me. But again, I think that crosses ethical boundaries from the get go, right? Totally. Just faking something, even the word fake, you're already crossing some boundaries there, you just heard that James, right, the ability to have the ability to have a human conversation, right, just a human level conversation, understand context, be able to context switch, be able to change your mind be able to follow along and be able to react appropriately and really fine tune all those aspects the same way a human would. I don't think we're that far away from that. I mean, we're years.

James Thomason: 41:57

I hope you're right. You know, I was sitting here thinking about some parallels in in times gone by like, Okay, 1935. There's a quote I like that Hemingway used in a private letter. He said,"modern times or increasingly mechanized depression, for which liquor is the only mechanized relief." You know, I thought about back to that era, the era of telephone, right, you used to have switchboard operators and in small towns. So when you wanted to connect to someone on the telephone, you would first pick up the phone that would ring a switchboard operator who would ask you who you wanted to talk to ask for their number and would plug you into the right port. And in small towns, you know, like, as pictured in Mayberry, people often knew their switchboard operator personally, right. So it was a really, really human interface to the telephone. And then switchboard operators became mechanized. And we saw that repeat for business switchboard. So after you know after telecom switchboards became mechanized business switchboards, were not. If you were in a big retail store, or any kind of office building, there was a switchboard to connect the people inside the business via telephone, and slowly but surely, and then all of a sudden, those were replaced by IVR, which is these, the voice prompts press one to continue to press two for sales kind of thing we've been dealing with for the last 30 years. And it seems like the first application of the latest wave of AI really was around text to speech and vice versa, and basically building yet another IVR. And so the question I have for you is, and we kind of touched on the ethics, but the other question is, like, do we really want a world which is increasingly dehumanized from the standpoint of interactions between other people, right? The ethical dilemma around knowing you're talking to a machine is very important, because I definitely want to know that and not be fooled. And I think people hate being tricked, also, that on the previous podcast, that's kind of the point that I raised is like, well, do people really want to talk to robots at the end of the day, right? Or would they rather just, if the robot was good enough, the future AI that you're talking about would be smart enough or good enough to understand when it wasn't going to be able to help me quickly, right, and to get me to a human being as fast as possible in that case, right. And I think they're all terrible at doing that right now, right? You, you get stuck in these loops that you mentioned before, and you end up in modern AI-driven IVR hell, with no way out, right to get back to a human being. And so I've noticed some of them, if you start screaming obscenities in the phone will connect you to a human right away. So that's a trick you can all try. Some of them are smart enough to recognize when a human has become very frustrated. That one works pretty well. It makes you look insane on the outside, though, if you're screaming in your headset, but it does work.

Joshua Schechter: 44:33

But James, I think we can all agree that these machines are inevitable, right?

James Thomason: 44:38

Yes. I mean, as a as a consequence of commercialization, you know, I mean, if no other reason that operation costs are always pushed to zero, right. So if a business thinks that they can increase its profit margin by deploying this technology, you're absolutely right. It's going to proliferate just like the IVR did, right. And it's going to displace humans in all those cases. My question was more like, do we want that? And maybe I was conceding my previous point like maybe if it becomes as good as you suggest, in the next six years or a decade or so where, you know, it's so contextual, I can understand when it's not gonna be able to help you and can get you the help that you need right away. I mean, I think that's a fascinating idea. And I hope you're right, I hope it is that good. Because I won't have to scream obscenities into my phone in public and scare children.

Joshua Schechter: 45:22

I mean, that's my fear also is that we want to speak to machines in a human way, as opposed to speaking the language of machines, which is cursing at them saying operator, not using pleasantries losing all the social aspects of it. If you guys looked at my LinkedIn a few months ago, I posted a survey. And it was based on a story that happened to me the night before where I came into my home and my daughter was yelling at the Alexa. I was like, What's going on here?"She's not listening to me, daddy, she's not doing what I told her to do." I go, "did you say please?" They don't use manners. They don't understand. But if you have that conversational ability there, you're actually promoting, I think the social aspect of speaking like a human to other humans or machines, right. Right now, I think we're actually with series and the Google homes and the Lexus out there. I think we're losing that almost. We're all starting to speak the language they force us to speak, which is not human, we would never speak to another human that way. And that's my fear is that we actually start doing the opposite.

James Thomason: 46:23

I think that happens on the not just with AI but like the internet in general, right? People don't talk to each other through the internet the way they would talk to a person face to face. Particularly when it comes to spewing vitriol and hatred and so forth. A lot of the discussions the unpleasantness we've seen in the last call it 10 years since the birth of so called social media, why is it everything is always named the opposite of what it is? Social media is anti-social, right? Like, that's massively anti social. Well, this has been I mean, just fascinating. And thank you for coming. And I know we're kind of reaching the end of our time slot here. I wanted to ask you, what's the biggest change you think will really manifest in the space in the next few years? We've even said a lot of things about how good the AI is going to be. Does any of this scare you or worry you at all? Are you you know, are you concerned in the way that the creators of the atom bomb seemed to be deeply concerned about what they were building? Now I'm destroyer of worlds kind of thing. And I realize you may have to, you may have to tow the corporate line here. That's totally okay, too. But what's your general sense? Do you think we're going to slide towards AI-driven dystopia or utopia sooner or later? Which one of those? Which one is the pendulum swing? And how soon? either case?

Joshua Schechter: 47:32

It's a difficult question, you know, when you think about where we can imagine ourselves in a few years, in regards to the AI, it's exciting to me, right? It gets my juices flowing. I mean, just imagine the innovation, and the, literally the disruption of industries, right? The Industrial Revolution, of being able to say, hey, I need a digital worker going to a store, downloading that digital worker installing it, and then that digital worker is up and running, and outperforming your humans on not just an efficiency level, but an NPS (Net Promoter Score) level, a CSAT (Customer Satisfaction) level, right, being able to actually perform better in that sense. That's what I'm excited about is that conversation.

Brad Kirby: 48:14

So this is great for big organizations with complex structures, right?

Joshua Schechter: 48:17

Well, doesn't have to be look at the App Store, right? The same concept, the app store was a new idea. And only people with iPhones could do it. And only very wealthy people had iPhones at the beginning, and they were accessing app stores. And that became mainstream. Right? So right now that exists, right? You can go to Amelia.ai. And you can download a digital employee. Right? You can actually do that. Now. Yes, it's geared toward the enterprise and the Fortune 500s, and the smaller companies can't see the ROI in that yet. But in five years from now, that's going to change, right? It's going to be for the masses. It's going to be democratized and commercialize to a place where it's going to be commonplace, unique conversationally, I no problem. Right? It's going to be commonplace. And I also, you know, going back to that dystopia versus utopia, I mean, it's got to be the utopia, right? We got to it's got to be enjoyable, it's got to be something that enhances your life. Right. And I think the creators of all this AI have that in mind, right? Because there is always that fear of artificial General, well, really artificial super intelligence, right? So there's always that fear living in everyone's mind that, you know, the machines are going to be better than the humans at one point. And by the way, they will be better than humans. And in certain ways, Dean, you said before, they are better than humans today, right? That Amelia that answers the phone can do a lot, a lot of things much faster than a human can. Right. So there's a we're already starting to see that start to creep up. The next five years are going to be exciting in this space. I really do believe that I think there's going to be a lot of innovation, there's going to be a lot of forward-thinking. But with that you'll also get the rule-setters, right so you're going to get some governance around. What's next. And what's not accepted what's ethical and what's not ethical, and you're gonna start seeing those rules start to pop up and actually control some of these creators and limit what they can do and what they can't do.

Dean Nelson: 50:10

You know, I want to throw one piece of context in here, because we're lumping all AI together in one bucket, there are ethical questions and you know, societal impact questions and all this, it can all be all be spun in the way that technology is applied for negative things. If you flip it back around, though. If you look at what Amelia is focused on: Enterprise AI, it is to go back and enable optimum efficiency and effectiveness, customer satisfaction, and scale at a lower cost that will continuously be able to grow. And you know, so it's it's very focused in the area. But the concept about what you guys are building - and I'll give you an example of this is that we had a guest on here, her name is Dr. Julie Albright. She's on the board of iMason's right, and she was walking through, she's a digital sociologist. And we had this interaction before, where we said, "think of the pandemic right now. And how many lonely people there are, because you're stuck in a house, right? And you're talking to a computer in front of you screens like this, you have the same people around you. Imagine if you had the ability to set up some digital person that would listen to you and be able to have a contextual learning and even the elderly. Tell us your stories. And right, the loneliness aspect goes away, cuz I have somebody I can call now is that unethical?" That's actually a help. From a psychological standpoint, there's a whole bunch of opportunities that can come out of that. So I mean, there's a million ways you could go back and say that this is going to help society. And it was funny, James, you mentioned this, where anti-social is what, right? Anti-social networks, it's because there's this dehumanizing side of humans, and be able to say, "I'm behind this thing. And I can go back and attack somebody and say something because I'm not standing in front of them." If you could say that to their face, would you do that? So if you've got the ability to go back and influence in that, that somebody you know, these digital agents can come in and start doing some things for society. I just think there's a huge application in that. Because we've gone in such the other direction. I know, I'm getting on my high horse here. But this technology is just so exciting to me, Josh, like you said, right. And Amelia is one example of how it's been applied AI to a specific sector that's going to help at scale. It's gonna help it be more effective. Now, where else can be applied? they happen to be leading in that space when it comes to the most human enterprise AI, right out there. Now, where else would we be able to scale? So anyway, I just I'm throwing in my two cents here at the end, because it's it's just such an exciting thing for me.

Brad Kirby: 52:41

Fair point Dean, like you brought up the number of 56% of the current labor force is going to get replaced by AI on our last episode, as we were talking about this, so I mean, I only have one final question. And that's: Do you know, who Amelia's best friend is?

Joshua Schechter: 52:56

I do not know. Who's your best friend? Oh, it's Chetan. Yeah, also her creator, right. Yeah, I would wrap up with one last thought, you know, for the people who fear what's coming with AI? Imagine, you know, as you said, with Dr. Julie Albright, the ability to have a non judgmental companion that can have conversation with you that will not judge you humans are inherently judgmental, right? No matter, you know, you could be a therapist and still be judgmental. Right. So just being able to have that level of confidence when having a conversation with an AI, that goes beyond interaction and being social, but it goes to a different level where, you know, you're not feeling judged, right? There's no, there's none of that negative association with it.

Dean Nelson: 53:55

So James, you'll never be judged again, because Amelia is going to be your friend. We've decided.

James Thomason: 54:00

I don't know I said some things to Amelia, where I felt like she was judging me. But that's probably just my human, you know, projection. She probably wasn't judging me. I did want to ask you one final thing, and we'll go, but, is she learning from me when we're communicating? Or is is what I'm saying? Just going into a corpus for later in my being forgotten? As soon as I say something? How does that work?

Joshua Schechter: 54:20

Well, we have a concept of episodic memory. So she's within the conversation, remembering everything you said to her previously within a conversation. But once you reset that conversation that's going to go bye bye, right? So that gets erased. So she's not going to know unless, unless you you know, some enterprises will store those conversations, and then you can leverage them in future conversations. But for the most part, each conversation will be a new episode or an episodic memory cycle for her.

James Thomason: 54:47

I was just gonna wonder if she's, you know, next time Brad talks to her if she's gonna remember all the bad things Brad said.

Brad Kirby: 54:51

I didn't say anything too bad. I was just I asked She. She agreed to go out on a date with me. I know that much. It was pleasant, to see a movie, specifically Avatar 2, but sorry, this is the face she gave me.

James Thomason: 55:11

She's got something bad. That's what it looks like

Brad Kirby: 55:16

At the end of the podcast. I do like a group picture. So that's my picture. But no, it's an amazing product. James, did we convince you that, are you starting to turn the corner?

James Thomason: 55:29

Well I mean, I think we I think we know what's real. I think what I'm convinced of is there are very smart people working on this problem. And, you know, it's going to continue to proliferate, whether we want it or whether I want it to or not, it's going to continue to proliferate. And I hope it gets to the level that Josh says it will. And Josh, I just want to thanks so much for coming on the show and bearing with us as we wandered around. Amelia, and thank Amelia for tolerating our abuse. Amelia's my daughter's name by the way. So yeah, a beautiful name? So Josh Schechter is director of cognitive delivered implementation of Amelia and Josh's company's IPsoft you can talk to Amelia on the web by going to Amelia.ai. Right, Josh? Amelia.ai?

Joshua Schechter: 56:12

That's correct. Have fun.

James Thomason: 56:13

Everyone should do that. It's entertaining and amazing how far it's come. So she's come. Can I say she? I gotta say she. Folks, if you enjoy shows such as this one, where we bring you the biggest, brightest minds in tech people who are actually living and breathing at the forefront of technology and working in the real world, please give us a like it helps us grow our audience. Our show is sponsored by infrastructure masons who cannot be builders of the digital age. Learn how you can participate by going on the web to imasons.org that's imasons.ORG. And by EDJX, where we're building a new platform for the Internet of Things. Visit us on the web at Edjx.io. That's Edjx.io.