Navigating the Future of AI in Computing: A Conversation with Massoud Alibakhsh, CEO of Omadeus

0:28Welcome to the AI Chat Podcast. Currently, Masood is working on revolutionizing human and machine communication through his latest venture, which is Amadeus. Thank you so much for coming on the show. Thank you for having me, Yadin. Thanks for that intro. Talk to us a little bit about what it is that Amadeus does and why you believe perhaps today with AI is a shift that is going to help propel that forward. Yes, this is something that I feel like I discovered it back in, I guess, 2017. I'd been thinking about it and working on it then as I was explaining and building these tools and trying to optimize communication between all the people within the company.

1:16I discovered that there is a way by studying actually software in itself. Software itself is a communication tool, typical software, any software that you buy from Salesforce or IBM. And most of the communication is done within every company through these software packages that take data in a structured fashion and communicate it from one person to another. And that's really the majority of the automation we use in all these companies, just like we did. Like we took Jira and Asana and Trello and Slack, and these are all software. But although Slack actually is a different kind of software, and then the reason why Slack or email found its way into the corporations was really that the lack of or inability of traditional software that communicates information in a structured fashion.

2:10And that leaves a lot of holes because humans communicate with natural language and natural language can cover a lot of subtleties in communication. So that email and Slack and these tools found their way in the corporate world to really fill the gap or the inability of traditional automation in these organizations that fell short in communicating the complete picture of information. So they came in to complement those tools. But they really created their own problems. They were promising panaceas and they ended up being Pandora's boxes. And studying all of that kind of led me towards this model that I discovered.

2:59And based on that, we invented lots of interesting technologies. And I essentially discovered that there is a way of totally optimizing communication amongst people. But you can't start with the communication itself. You can't start with the message or the messenger. You have to actually start with the workflow and look at what's going on on the workflow. So that's all, you know, I wrote a paper on that in 2022 and it was published by IEEE. And I've got a little video we put out that kind of covers all that. And I'll make that available for you if you want to share with the audience. I think it's, I recommend everyone to watch it because it gives a good insight into why email doesn't work.

3:43Why do we have so many problems with email? And why doesn't it scare? Or even what they call the social media network for enterprise, Slacks and Yammer and Teams, and why don't they work? I mean, they solve some of the email problems, but why are they still not perfect? Why do they fall apart? And that's because they are trying to map a very complex, multidimensional model we call workflow. and workflow is really an abstraction of assembly line. And we inherited that, we abstracted that since the early 1900s. And this is how we, as large groups of people, can get together 500, 5 ,000 or 50 and organize ourselves and build products or deliver services.

4:34And that's the magic of how humans are connected together. And that's where the secret is. secret lies in making that the pivot of communication. Yeah. And let me take a stab at explaining this concept. I'll drop a link to that video that you mentioned in the description for this for people. But let me take a stab at explaining that concept. You let me know where I fall short. And then let's talk a little bit about how you are using this to solve problems with your company, with Amadeus. And how it all ties to AI. Because that's really the whole, Yeah. So essentially, you know, everyone knows emails have a big problem when you have, you know, a thousand people in your organization and you send an email out to everyone.

5:14There's a ton of people that it's the email's not relevant. You've been in organizations like this. Everyone's been in organizations like this, right? You get an email about a specific project. You're not really relevant. You're not a decision maker. The information is irrelevant to you, but you just get a ton of like this kind of noise in your inbox. So we upgraded a little bit to the concept of Slack, where essentially you have channels or Microsoft Teams. You have different channels. You've got the marketing channel. You have the HR channel. You have a bunch of different channels, and perhaps you're relevant to that organization, so you're in the channel.

5:42Again, the same problem is you scale up to a bigger organization. A lot of people can be messaging in that channel, and there's a lot of noise that may not be relevant to you. Email doesn't scale because the pivot of that communication model is the individual. That means the individual is responsible for analyzing the information and figuring out who to route it. Now, if you start scaling the number of these people from 3 to 5 to 10, 50, now each individual needs to have a very complicated routing map in their table, in their head, a routing table in their head, and be able to analyze the information and figure out who to route it to.

6:18That's really too much task to ask. It goes beyond 4 or 5. Now, you know, you got it. That's easy to see. And in the social media model, which is like Slack and Yammer and some of these tools, that pivot shifts from the individual to the subject. Instead of creating, me sending the information in an email to you, we'll just create a channel called Paint. We'll create a channel called Nailing the Chair. So I put information. I can now analyze that information as a human. And instead of trying to figure out who to send it to, I'll look at the content of that information and say, well, that's basically paint content.

7:00So the assumption is that the mapping is easy. So that's the abstract model. But the problem is that you're building chairs, lots of chairs. You're making thousands of chairs. It's moving on the assembly line, right? you may be interested in a salesperson may be interested in a particular chair that's a red chair he doesn't care about any of these chairs but he has to subscribe to that channel that says paint now he's going to get a lot of bombarding messages for better access yeah so that's for that example i can kind of the the show why why these things fail now not everybody even it's like in these models who creates these channels how do you name them like it just becomes channels galore anybody who's ever used these systems they know what happens as you add more individuals and more subjects.

7:45And that's because this stuff is really not thought through or analyzed in an abstract form. These tools found their way in an evolutionary process into these corporate settings because Facebook was extremely successful and people asked, hey, one individual can communicate to thousands of people in a nice way and we can't do this with email. Why can't we have a Facebook thing for our company? And that actually drove a lot of companies with opportunities to try to build these tools. And Yammer was one of the first successful ones, and then Slack became successful. And you got a lot of tools that are similar to this.

8:22But they ended up not really solving the corporate problem. And in fact, they really exacerbated it in many interesting ways because now you have corporate data sitting in all those databases that are being input by the automation and software that is the main nervous system of the corporation. And then you have information sitting in emails that is really the natural language, and some of that information is even duplicate. That's because you're extracting things from the database reports. You're attaching them. And then you have some of that information sitting in silos of Slacks and Yammer and Microsoft.

8:59So the corporate data is scattered all over the place. So that's the base of the problem. that for me as an engineer, I get very excited about because we want to solve these problems. And this problem is prevalent all throughout the corporate world, all around the world. And as we know now, fast forward, we get to where data is oil. Why is data oil and or data is gold? Why? Because that's what you feed to AI. Okay, so in order to create some sort of meaningful intelligence or output from or work from an artificial intelligence engine, you need to feed the data. But your data is scattered all over the place.

9:44And this is what the success of ChatGPT that demonstrated the power of large language models. Now, a lot of corporate CEOs are asking the following questions. How do we get AI into our company and how do we actually benefit from this and how do we use it to create efficiencies? This is exactly where we are right now. We're in this space. Yeah. And what I see on a daily basis, and it's almost becoming repetitive, and I'm sure most of your audiences would have similar experiences, that we're beginning to see repetition of the same stories because AI is being used as a tool in a sense. and there are problems that are basically what I call low-hanging fruit.

10:37And you could easily see the application of AI, whether it's a large language model where you take a manual and feed the AI the manual or the frequently asked questions and then you use it as a better chat bot to answer questions of your client. Or you summarize a legal document or you fake somebody's voice or all the wonderful things that you can do with MidJourney or DALI. So there are a myriad of these tools that are producing a lot of wonderful, wonderful results and also help with productivity or improving productivity, but they're all long-hanging fruit and they're all superficial. Because this way, from our perspective, that we're using AI as a tool where we think AI itself is a platform, similar to the fact that graphical user interfaces back in the 1980s were a platform, and a lot of software had to be rewritten, or every software had to be rewritten.

11:38The software you use today is based on graphical user interface. It's based on that computational model. And that was a major revolution that happened in the 80s and the 90s, and a lot of software companies had to change their model of writing their software that's continuing to this day. And the same, a different revolution happened in the 2000s. And that was the transference from local area network computation to what we call cloud today, which is the internet. It's a distributive model. It's a different kind of architecture. It's a completely different way of writing software. And the way we organize as engineers is different.

12:15We have DevOps, we have back-end people, we have front-end people. It's a completely reorganization of human expertise and the way we actually approach building the system. Now, what we see is that we are entering a new revolution, and that is the artificial intelligence revolution. And we need a new, brand-new architecture, brand-new way of breaking those business software into a new model that can take full advantage of the artificial intelligence. And we've invented one, and we call that object messaging and intelligent objects. And we feel that the major revolution and the major impact that the artificial intelligence, namely large language models are going to have on human society, is through this re-architecting all of our software for this new platform.

13:12And in that world, our software. And I'm going to explain why that is and what is our software, what's the problem with the software that we're using today, whether that's your accounting system or your electronic medical record system, if you're a physician, or if you project management systems, if you're an engineer or a project manager, any software that we use. It's outdated, outmoded. It needs to be broken up, thrown out, rewritten all over again for this new platform. So what would you say are the major shifts as far as this platform goes? Shifting from AI as a product to a platform, what do you think the biggest changes are going to have to be?

13:57Okay, so I'll describe the story. See, when the computers showed up in the work environment, the corporate world, business, they showed up and they said, well, we're going to bring some cool automation to your place. And let's say you were a big manager in a bank somewhere or a manufacturing facility or whatever. And I'd ask you, Yadin, what's your existing process? And where do you keep your information? And you would kind of describe your process. This is what we do. I fill out this form and Susie gives it to Jane and Jane gives it to David. They fill out this section and it goes to the floor.

14:34So I would look at that and I'd take all the forms as a software engineer. and I would document that whole process flow and your physical process flow and I'd draw a flow diagram, a flow chart, and then I would extract a typical data flow diagram or a control flow diagram. These are all diagrams that actually I would be using to automate your process, your physical process, and take all your forms essentially and put them on the screen on a computer. That's the model for business process automation, basically, in a very simplified way. Essentially, I take your physical forms and stick them on a screen, and then I take the data, which is structured data, and then that data is very specific.

15:24It's name, Joe, size, 15, date, 11, whatever. It's like you can't put something else in, and that's what's referred to a structured data. And that would get pushed into a database and then it's a lot faster. It's automated. It standardizes things. You don't have to fill things out on the form of the typewriter. And the access is easier. Somebody doesn't need to put all these forms together and create a report for the high manager. You do this magical query and whatever information you want, you kind of make real-time reports on somebody's screen. And this is what we've been living with since then.

16:01That's the model. Even through graphical user interface revolution, this model didn't change. Even though we changed the way humans interacted with the computer, before they would just interact with text and just read questions that will answer the questions. But now there are objects on the screen that you interact with. Click on, it opens up, it tells you something. And you move things, you drag something, you drop it somewhere else. So that graphical user interface facilitated and made human and machine interaction easier, much, much easier, and made it less intimidating. And that was really coupled with the introduction of personal computers, they were called at the time, PCs.

16:46I don't know if we still refer to them during the 80s. And so computers ended up being everywhere, but that model stayed the same. Take the forms, stick them on the screen. Even during the transition to the internet, that model didn't change. The same forms on the screens put them now on the screens that are pushed through computers that you don't see somewhere in the cloud, right? Now, AI is here, and we've been waiting for this. We've been waiting for this because based on the model that we created, we've been waiting for a reliable way to be able to process natural language 100 % reliably. The problem has been so far that computers really didn't understand natural language.

17:33And it did not in a reliable way. If anything, ChatGPT can demonstrate is that now we can actually understand. We can build machines that understand humans 100%. Chai GBC, for example, is not all reliable when it's generating answers. And a lot of the Google people coined a phrase called hallucination. So, it can hallucinate. But even when it hallucinates, it still understands you perfectly. It makes stuff up, but that means that it just understood you. So we take that part of that technology that understands language, and we actually have designed a new approach of creating software. It's like beyond the whole taking all the forms and then sticking them on the screen, we're saying, no, look, this is an intelligent machine.

18:31We can actually really remove the burden, a lot of the burden that we put on humans in terms of determining who to send information to, when to send it to, where to store information, and which creating file folders, put things in this file folder, put it in that file, organizing the information. I mean, why do humans have to do this? And it's like we've got a computer here. This computer, you can program it to play chess, for God's sake. You can't program it to file your data for you where it's supposed to be. But see, that's a good point. Yeah, I really just think, why should I open up my, turn on my phone, which is a computer, and look at there with 50 ,000 apps, try to figure out which one to press?

19:20I mean, why can't I just talk to the darn thing and say, I'm hungry? Right. So just say, what do you feel like? I mean, did you ever see there was a movie actually called, I think it was called Her, that was not too long ago where a new operating system comes in. The only interface was the natural language, actually. It's a famous movie at the time. It was popular. But that's where we need to go to. We need to go where humans interact with the machine using natural language. and graphical user interfaces are also good because there are a lot of times you just point at something instead of using 10 words you can just point at something right so the combination of the screen and natural language could probably make the ideal user interface but you need a different kind of architecture to build this software and the way that architecture the way we describe it is instead of just taking dumb forms and sticking on the screen and then connecting them these forms together with automation and putting the data in the database, you need to go one step further as a software engineer.

20:32Actually, a few steps further. You need to, in that environment after you talk to Yedin and he describes to you the process, is you need to go further and start doing the semantic analysis and identify the real and virtual objects in that environment. And then you need to find a way of representing those objects with programming and make those objects intelligent. There's a way of creating data around that object in traditional forms to give it, for example, I could kind of represent an x-ray. Let's say an electromagnetic orb. I can make an x-ray and represent that as an object. I can tell it has attributes that you're an x-ray.

21:15The technician who took your picture was Masood on this date. So these are all structured data. But what we can do is actually this object. Now we can give it a large language model. We can insert a large language model such that it can actually communicate with humans. That x-ray object can communicate with humans using natural language. And then we add other pieces to this object. We give it ways, like, imagine like diary books. We'll give it a black diary such that it can record the communication between the cells, the x-ray, and the x-ray is responsible for recording that communication between the doctor, the primary doctor, and the patient.

21:54that it will give it a green book to record all the conversations between the primary doctor and specialists and any other or nurses. So basically, we make the object intelligent and self-aware, and we give it the ability, just like a human, to process natural language. So it can not only have dialogues with humans and understand what's happening, but it can also record information. For example, if the doctor needs to have a conference call with the patient, the object itself can hold that conversation and sit there and be present and watch that conversation and record the video and store the video within itself.

22:36No longer anybody has to say, well, I'm going to put this video file in this folder and I'm going to remember that it belongs to patient Masood, right? The object itself knows that it's an X-ray. It's present. And it's actually sitting there listening. And it can sit and listen to the entire conversation between two specialists, the doctor and the specialist. And they're talking about whether this is cancerous, it's not cancerous. And the object, just like a human, can remember and understand. And in fact, it can even do better than the normal traditional LLMs because it doesn't have to search the entire space.

23:12It knows that it's an X-ray. So it can focus, actually. The structured data that defines that object allows it to focus and have constraints about the way it processes that natural language. At the end of the call, it may be that you and I are the doctors and we talk about, okay, now we say, how about golf this weekend? See, the object knows, okay, this is not related to me because now they're talking about golf. So this object becomes super intelligent about itself as to what happens to it and who the stakeholders are and how to keep track of those stakeholders when critical information happens to itself.

23:51It knows how to navigate the workflow. It knows how long it has to stay in each stage. And it knows all the different classes of stakeholders, what kind of information to share with whom. So if you think of your system as a collection of all these intelligent objects, now, what does that remind you of? There are systems like that on planet Earth. They're biological systems, actually. A lot of biological systems are built in such a way where we have, for example, in my body, in the back of my brain, There's not a little file folder that keeps track of cell number 57 in my liver. That information is self-contained within the cell of my liver, and the liver cell knows exactly what it is.

24:44It's a fairly complex machine, and it knows its function. It has memory even. It knows how to communicate with its neighbors. It's intelligent. It's self-aware. And the collection of those cells create the liver, and the liver itself is an intelligent. super component that's connected to the rest of the system through veins and arteries and nervous systems. So we're actually adopting this model and taking it to silicon and saying, you can actually represent a business or manufacturing process and go beyond just the workflow process and the basic automation and identify these virtual or real objects and represent them in your program.

25:29And add LLM to them so they become intelligent, so they can process all kinds of information and interact with humans and also each other. Because if they understand LLM, these objects can talk to each other in natural language as well. They can query each other because they all have memories. They can say, hey, the x-ray could query the blood test object or the specific object inside that blood test. The x-ray could interact with the HDL or your blood count, for example. I'm not a doctor. I don't know if that would make sense, but you can imagine relevant objects that are basically as a software engineer, what you're doing, you're creating this framework.

26:10All you have to do is create this framework and then you create objects and you give them life And these objects can create super objects and the super objects can be connected systemically to each other. And you could even design a super object that is a brain that's sitting there, similar to our brain, that the brain is sitting there coordinating the efforts and communications at a higher level between the liver and the kidneys. And so it's a lot more complex than that as an analogy. The biology of humans is a lot more complex. As a matter of fact, it's a source of the inspiration. But if you and what we've done is actually we created this model, which we call object messaging and intelligent objects.

26:59And this is a brand new model. And based on this model, what we did is we spent quite a number of years, a few years trying to build a platform, actually, and framework. But this was too complex. And a lot of engineering problems had not been solved. You know, this was brand new territory. So we decided after almost two years that we just pivot and build an application using this model so we can actually demonstrate to the world what this thing looks like, how it feels. to interact with a system like this. So we had a problem back in my last company that I was just referring to, trying to integrate Jira with Trello and Slack and Google Docs and all these tools and writing software.

27:48So we decided let's solve that problem using this model. So as a result, we've created project management, collaboration, communication, documentation, powered by AI in one package using this model, such that from the moment you log into the system, at every point you're interacting with the AI. And it's implicit, it's sitting there. You can call it and you can interact with it. But essentially, the system is a collection of all these objects that handle your project management. And these objects are intelligent. They know what they are. They know, for example, if they're a feature or they're a bug.

28:26They know if something happens to them. They know who to go to, who to complain to. If they sit too long in a particular stage, they'll look at their clog like they're expired. They run around and tell their stakeholders, hey, you know, I was supposed to be here 30 minutes, but I'm here now for 35 minutes. And it gives a chance to interact with the appropriate stakeholders. It's a fascinating experience. It's really a taste of the next thing to come. So we've kind of demonstrated to ourselves how to build something intelligent on the machine. I feel like this is an entire paradigm shift for software and for project management and for so many things.

29:09This is such an interesting concept. Talk to me about what does this actually look like in practice? So you're talking about, right, you have these different elements. For some reason, when I'm thinking about this, all I can think about is all of the folders inside of an organization, all the different departments that organize their projects by folders or by different things. So all of this has intelligence to all of them. What is the integration? Let's say I'm a regular corporation that uses Microsoft Teams to organize and pretty much the whole Microsoft suite to do everything. Is this an integration I can pull into my current system?

29:48Do we have to recreate new systems for something like this? What's this going to look like for your average enterprise? I think I'm pretty convinced that we're going to have to go through this tectonic shift the same way that we did during the graphical user interface revolution, where we ended up having to rewrite all this software, and as we did during the LAN to the internet, where we ended up having to rewrite all of our software. Because the current model, the way, for example, Jira is written, the way Slack is just a tool, it's basically really just trying to solve the problem of natural language and converging it with structured data.