My guest is Magnus Revang, the chief product officer of Openstream.ai. Their engine, which they call Eva, allows users to interact with technology naturally, using plain language but avoiding AI issues like back-end complexity or hallucinations. We’re going to talk about true conversational AI, human collaboration, agents, and how you build a solution with data that’s never been written down. All, on this edition of PeopleTech.
Transcript
Mark Feffer
Welcome to PeopleTech, the podcast of WorkforceAI.News. I'm Mark Feffer. My guest today is Magnus Revang, the chief product officer of Openstring.AI. Their engine, which they call Eva, allows users to interact with technology naturally using plain language, but avoiding AI issues like back end, complexity of the informations we're going to talk about. True conversation, the AI human collaboration agents and how you build the solution with data. That's never been written down all on this edition of people tech.
Hey, Magnus, welcome. You folks are very active in building conversational AI, and I wondered if we could talk about that a bit. Can can you tell me what sort of led you into wanting to get into conversational AI and what you think it's potential is?
Magnus Revang
Of course. Yeah. So conversational AI is the ability for a machine or an AI to hold the conversation with a human and natural language conversation. UM, and really, it's the most natural way to communicate, right? If you look at the science fiction movies going back, you know, back to the 1950s that people imagine sort of the future operating system, the future way to communicate with machines, they they've always ended up. Pointing out that it's going to be a conversation. A natural English conversation.
Speaker
Machine.
Magnus Revang
And conversational AI has always been a part of computing. Uh, way back. But really in it was, I think in around 2014, 2015 uh, where this concept of chat bots started getting a rise. In the market, what people didn't tell you at that time was that conversational AI was. Very little AI at the time, so basically it was the ability to take what somebody said and do two things with it. Basically it it tried to to find out by saying this what is the intent. Of the user, what's the intention and what keywords were mention? That uh fills out the the the slots of the request. So basically that just mapped to a part uh to a decision tree that were were part of of a dialogue tree, so to speak. Right. And from there it. It was very little AI. And and that's how conversationally I started out. Now my interests have always been unscripted. Conversational AI, conversational AI that can deal with things where where the the scripts it becomes too. Complex for a. Script and when when Bert first came in 2018 and later on the GT versions and you know 2022. With shaft, GPT and instruction following L alms, things started to become very. Now one thing that you have to think about in conversational AI is that if you're doing something for consumers, you you can. You don't need to have full control, right? Just like metal. Then you don't have full control what it's going to say, but if you're an enterprise where you're doing things like. Opening account cancelling credit cards you know, and doing, doing customer service things you can't really live with something that. Is unscripted and low control. You need to have unscripted and high control right and that is where you know, I play around with conversational AI where you have full control over what the machine can say, but also the ability of the machine to deal with things that. A scripted approach. Cat, uh. So what would would kind of like a thing that scripted approaches can't do is is simple things where there is some sort of ambiguity. For example, if I go a simple thing of what's your name? Right. Well, for a lot of women, for example, they don't necessarily remember. If. Ohh, did I get this account before I was married or after? Uh, is it easy? Am I registered with a maiden name or my real name? Right. And being able to to say both. Is is actually hard when it's a scripted approach because the scripted approach. Assumes that you're going down to 1 and you're not splitting and doing both the two in parallel. To check for example. So that's a simple example of of where you can trip up any, you know scripted chat bot. Is by just giving two names or or maybe I use my middle name three names right and you start to add add to it, another is in scheduling right where where? The natural way to to schedule is that I give you a set of constraints. So if you ask me then can we do? The podcast next week, right? I will go. Hey any? Monday or Tuesday after 3 o'clock. Right. And and and suddenly I have constraints on it and you go maybe. Yeah, 6:00 on Tuesday. Is that OK? And that's a little bit too late, right? Maybe maybe 5. So so it becomes a negotiation between us in a way and being able to deal with that in scripted. Is also, you know, almost impossible because the number of branches you would need to the script would be too many to create anything, so you'd need to add alternate approaches to it. Our lamps can come in. And do some of it, but with RLMS you don't have the full control. You have don't have the guardrails to say don't go outside of these things. So often you need a more symbolic AI system to determine what to say. And then you can use LLNS to. Determine how to say it right and and you can combine the technologies in in various ways and that's how you you can you can get to this this form of conversational AI with high control and with with the kind of the the strengths of LLM. These systems and having unscripted approaches. And, but conversational AI is kind of expanding as well. So so for us, I mean we do things like avatars and emotion and personality detection, being able to render and personality and being able to render. Motion into both in paraphrasing of what's being said, but also in tone of voice if it's going to be rendered for a text to speech right. So all of those things come in as well and and it's becoming quite complex.
Mark Feffer
How far along are we with this from the from the end users point of view? I you know. A lot of people say, well, you know, I I was expecting Star Trek, you know, I'm. I'm. I'm waiting for something that's that natural and that that.
Speaker
Good.
Mark Feffer
And what you're describing is very complex. So where where are we? I mean where where? Is the reality. Leaving us right now, where do you think it could go?
Magnus Revang
So the reality is that if you. If you dial back the controller bit and live with uncertainty. Of things going off the rails and stuff like that, you can do a lot more than if you need full control over the tools being used and and that nobody's cancelling a credit card off of tricking your agent or whatever. So so that that's kind of the. The the the challenge right in all of this is that consumer technology, you see all these demos that. Are. Super impressive, right? But then you find out that well, there is limitations in how you can. How you can force it to follow business process, how to follow compliance directives, how to follow business policies and stuff like that? If you if you start to tweak the technology to be able to do that as well, suddenly it gets a lot less impressive because it's basically. Storm on just audio from a lot of of conversations that it's mimicking, right? But but the training data you have no way to go through that to see if if there's the policy and and stuff like that. But so this is kind of the balance that we see that enterprises can't necessarily. Adopt the technology. That you know, demo's nice in the voice setting, there's also, you know, if you look at a lot of the voice demos out there, they kind of trick you as well because if you have a conversation with somebody that is flirty. And pays attention to your laughs. You know, stuff like that. And it's very casual, you know. A lot of people. See through the faults of it, they will accept some inconsistencies and stuff like that because it's it's so nice to have this kind of conversation where I'm I'm the center of attention and stuff like that. But if you're doing customer service you you don't want your. Assistant to go flirting with the customers and, you know, doing stuff like that there, there's certain professionality that needs to be there so. For for most brands, I should say there are some brands where where you you you could get away with it, but uh and and suddenly you you you might not be as tolerant for what? Uh support assistant and add to that you're actually mad at the brand because the credit card didn't work or or something like that. So you're starting from a. And and if somebody comes along and starts to flirt with you and laugh and stuff like that, then I'm actually having a problem. You know, that that might not come be as good as well, so, so. So when you adopt this to the enterprise, there's so many more factors to consider, so many more. Control mechanisms that you need that a lot of the the same technologies and can't actually be brought into that domain easily.
Mark Feffer
You know, I always think that conversation AI is kind of like the Holy Grail of user experience and in the user interface. And it it makes me wonder, you know, why aren't there more people out there really focusing on on developing this? And I I wonder, are there a lot of people out there doing it? They're just doing it quietly. It seems like sort of a natural.
Magnus Revang
It it it is that the problem again? Is this balance between control and and non control cause cause a lot of you know if you look at the field of conversational design. Which is a a field that that was emerged as a way to make these interfaces. These natural language interfaces, and what they concentrated on initially, and it's it's slowly moving away from that. But initially what's the script dialogue, is it basically? Great scripted dialogue, right? That was seen as as compelling and stuff like that. And and really that's not where you want to go, you want to create systems. Where you're able to have any conversation within that conversational system, right? And that's much harder because there is no standardization. There is no standard model of how to do that. We have a lot of of technology there, but. It's it's, you know, it's not based on on standards. Now. I would say that conversational AI. Is kind of just the beginning as well, and I think this is important to to to say because conversational AI is just the. The human and natural approach to what you can say is unplanned or unscripted interaction. So sometimes you want to bring in the abilities that. A phone or or a computer hub into the conversation. In the form of multimodal interaction. So if you're going to, you know, point out where did the accident happen? If you're talking to an insurance company, do you want them to kind of say in natural language and address, you know, most are driving on the road, right. So they don't know the address where the accident happened. So or just show a map. And have them just point at the map, right this in. This area circle. Like and and and that becomes sort of the input, or if your cable company right and somebody calls in and you know what? What kind of setup box do you have? Or just take a picture of your setup box. I take a picture of the setup box, I can see what type it is and what lights are. On the front of it, so suddenly I have two pieces of information, right. And I I can actually do more than what the human could. In in in that in that conversation. So I think it's important to think about it more as this the the ability to go back and forth and the ability to go down many paths and have you know a long conversation or a short conversation depending on the skill level or the the complexity of the problem that's being solved. UM, and I think that is more important than most important aspect of conversational AI. Uh, so. So I don't think it stops at just being natural language. It's it can go further than that.
Mark Feffer
So how did how? Does this align with the work of agents or or how do they they go together because agents are really the big thing right now that everyone's.
Speaker
Talking about.
Magnus Revang
Yeah, and and if you ask 10 people about the definition of agent, you will get 12 answers, right? So so one thing is that you have agents that in is single single agents and and single agents is just an encapsulation of an AI system. Ways to communicate and. Ran the route outputs right, so the you know simple little land could be an agent, right? And and a virtual assistant could be an agent or a virtual assistant with an avatar and the ability to detect emotions and all stuff like that. That could be an agent. Is that? Encapsulation of capabilities around the AI and and the knowledge base and everything. So so agent is is sort of like a software construct in a way. Now the interesting thing is when you start to talk about multi agents. Systems as well, where you have multiple agents that are communicating between each other. In the system and sometimes also communicating with humans. So human. Operator or multiple humans and their. A lot of AI agents. Is together solving a problem that might require the agents to communicate between each other? And share information. They can be from different vendors. They can be from different system.
Speaker
Yes.
Magnus Revang
And and stuff like that. And I think that, you know, agents we have had for a long time, uh multi agent systems as we see them now. You know, although there is multi agent protocols back from the 1990s, uh, the the the current ones just rely on. Natural language as the as the communication between the the the agents in the multi agent system and in modern kind of large language models it makes it possible to have that communication and it's very, very interesting. What what what can be done.
Mark Feffer
Because it it sounds challenging because you also have to have this conversation with something or somebody that's going to tell all these multiple agents what needs to be done. And I I can't imagine that's easy.
Magnus Revang
You know multi Agent systems are. Hard to build. They are also you have to do these three things you wanted to solve, right? First, you want it to solve by building to communicate between agents without having a predefined API or protocol to communicate, so they can use natural language between and. They can figure out what is being said and how this applies to me, right? So so so you can go away. Protocols and stuff like that. But you also wanted to kind of solve the the discoverability. So if I put 5 agents together or I put 10 together in a multi agent systems, I would. I'd want to do the minimum amount of work as a developer or or business user. When I invite the new agent then. In telling all the other agents what is capable of and and how to use it right and the third is of course the the ability to to kind of self assess. This is the promise of multi agent systems is that they self discover they self assemble and and and do things like that now. The reality is that in the agent frameworks that are out there today, very little self discovery and self assembly is taking place that is actually done by the developer. So they are doing the discovery and they are doing and the setup graph of agents or of workflow or something. Like that to make them work together, you know? But. But the promise is much grander. Right? And and what? What current systems look like is more like microservices. Frameworks with AI. It's not very different but. But I do think that with. If you give users the ability to. Instruct their own agents. Create their own specialist agents and then instruct groups of agents what to do and guide them and stuff like that. That that multi agent systems can can deliver you know in in, in a couple of years on that self assembly and self discovery. Problem as well and that's where it becomes really, really interesting because. The capabilities of swarms of dumb AI services. There's probably going to be much higher than the capability of a single. Smart AI system so.
Mark Feffer
Let me shift gears for them because there's something else that you guys are up to that really fascinates me, which is you just talk of unstructured data. Everyone talks about that as a challenge, but but you're actually putting together knowledge bases based on unwritten materials. So you haven't even gotten to what structure? Can you talk? About that, what do you folks up to?
Magnus Revang
So so that that, that's cool that you mentioned. So of course we have a knowledge ingestion system that takes all kinds of data from Excel, PDFs and stuff like that like like that and brings that into and create knowledge graphs and stuff like that from it. But the interesting thing, when you work with professionals, right? Highly paid. Experts is that they have a lot of of knowledge that is only in their head and they're not writing it down. It's just how it always been and this is how we do it or you know, I always do this. That and and when we have a conversationally our front end and we have multiple agents behind, uh, what we do is that we capture knowledge from the interaction between the expert user. And agents. So if somebody for example is, is, is is doing. Uh, writing a report on investments or or whatever an investment report and and they go and and you you can say things like well whenever you encounter this code or that code, they're effectively the same. Category. Well, you know that's a that's a, that's a thing that the system didn't know, but it's being told as it's working with the expert user. So it stores it for. Subsequent. Work with on on that domain and and capturing that expert knowledge is you know there's so many neurons to it. You have to figure out, oh, is this for this particular case or is this something that applies to all cases, right? What's the scope of it? Is is it? A guidance, is it a correction? Is it a feedback on what I need to do? Is it? Is it additional knowledge? Is it an additional instruction? There? There's all kinds of like facets around just just what what something is that that that somebody says? But the the the idea, right? And this is what a lot of AI systems, which is strange, a lot of AI systems out there. Don't learn. They've been trained, but they don't learn right and and to me that's a very, very strange thing, because why would you give up? The ability to continuously learn. Now you sometimes you would want there to be supervised. Sort of checks and balances on what is being learned, but having an AI system that is just trained one time and doesn't learn after that is is is just, you know, it could be anything, right? It it that's that's not the the ability of AI to learn. It's what you really want and you want that to happen over time especially. When you work with expert users because you want to capture that, that knowledge and the more you can capture the more value the system gets. So it's it's it's about capturing more and more overtime.
Mark Feffer
How do you capture it?
Magnus Revang
Yeah. It is a different ways to do it, but but basically you're categorizing what kind of feedback you're the the user is giving right? That's the first thing and you're determining in from that feedback if there is. If this is something that is. You know, outside of the scope of this particular task. So is this something that applies to other tasks as well? And if it is, you run that into a system that captures that that instruction and you paraphrase it in a way that's generic. And you in most cases we can do it unsupervised, but in most cases you will have a supervised. Of these are the most common feedbacks given by the expert users, right and and and an admin goes in or a super user goes in and goes oh this. Yeah, I would want to add that and I want to add that and I want to add that. And when those are added, those are added to the system's interpretation. Of of what to do so it is added to the instructions to the system and if it's LLM you're added to the the prompt and then you optimize the prompt and stuff like that or if it's not an LLM you have to maybe change the weights and do some reinforcement training or things like that, but it's it's possible to do in in a lot of different AI systems. To actually capture that and and make it part of later runs so, so, but you know it's it's it's interesting because. For the Super users that are doing this in a supervised fashion as well, they might go and say hey, Oh yeah, I'm taking, I'm taking this for granted, right? Or isn't just an assumption we always make. I didn't know that that could be that that was had to be explicitly said. And that makes training of.
Speaker
And.
Magnus Revang
Expert users and stuff like that also easier because it might make them more aware of that unwritten knowledge. So it actually also affects the the expert users and and how good the humans can be. In addition to the machines and I think for a lot of the. The the tasks that we use AI for your. You know you're not replacing. Expert users or or highly paid expert users in things like investment banking and insurance underwriting and stuff like that anytime soon. You're helping them be faster. More precise. More effective so they can do the things that humans are really good at, which is kind of the the intuition part. They can spend more time on that versus collecting information, researching information, comparing different, you know, things to each other and stuff like that. But machines are. Really good, right?
Mark Feffer
Mavis, thank you so much. It was it was great to talk with you. I hope we can do it again and again. Yeah. Thank you.
Magnus Revang
Absolutely. Thanks for thanks for having me. I'm, you know, I'll I'll gladly be on again.
Mark Feffer
My guest today has been Magnus Revang, the chief product officer of Openstream.AI And this has been PeopleTech, the podcast of WorkforceAI.News. We're part of the Work Defined podcast network, find them at www.wrkdefined.com and to keep up with technology and subscribe to workforce AI today. Where the most trusted source of news in the HCM tech industry find us at www.workforceai.news.
I'm Mark Feffer.
Image: iStock
Share this post