Subscribe
Share
Share
Embed
This podcast offers business solutions to help listeners develop and implement action plans for lean process improvement and implement continuous improvement projects, cost reductions, product quality enhancements, and process effectiveness improvement. Listeners come from many industries in both manufacturing and office applications.
Anas Saeed 0:00
There's a saying that garbage in, garbage out. Okay, AI will just amplify stuff. So that's why if a process is a mess, the AI will just amplify that mess. All the processes that were made, okay, that were currently before, they were made before the age of AI. So instead of bolting on AI onto that process to automate it, you have to basically redesign the process such that AI does the grunt work, and the humans are involved with the judgment, exceptions, relationship, and future.
Shayne Daughenbaugh 0:41
Hello, and welcome to the next episode of Lean Solutions podcast. I'm your host today, Shane Daffinbaugh. Guest today is Dr. Saeed, and he holds a PhD in artificial intelligence and has spent more than a decade in software development, delivering over 150 products. He is the founder of Alf Chi, where he helps small and mid-sized businesses apply AI to real operational bottlenecks, including lead responses, customer follow-up, and front desk workflows. His systems have generated millions in revenue for businesses using them, and I'm very excited to have Dr. Saeed come on. Dr. Saeed, welcome to the podcast.
Anas Saeed 1:19
Thanks a lot for having me, and thanks for that kind intro.
Shayne Daughenbaugh 1:22
I am very happy to have you here, and I'm very interested. When did you get a PhD in artificial intelligence? Like, how long has artificial intelligence been around that you could get that high of a degree?
Anas Saeed 1:35
So here's the thing: artificial intelligence. Okay, the way you define it, it's been so the first neural network created, I think, in the 60s. And before that, it was machine learning has been really common, but it basically like in the public sphere with ChatGPT, AI became extremely common knowledge. Before that, in the academia, like or in the world of academia, AlphaGo was the real game changer because that's all the game of Go, and it beat LaserDong, which was the best player at that time. And even before that, okay, machine learning had become really open, and the reason for that is because of GPUs and the internet. There was a lot of data, and there was a lot of ability to process that data. So those two key elements, neural nets were there before, but there was never the processing power or the amount of data to make it possible. And like even before, like in the 2000s or before that, there was a neural net that basically all the letters that you write on top of it the handwritten name and the address that used to be computerized using a neural net. So there were small typical applications, but it was extremely obscure. Again, no one knew it, especially for the public sphere. It came out in 2022 late with the launch of ChatGPT. Before that, like AlphaGo in 2016, and before that, neural nets existed for a while. It's just that they were not in the public sphere
Shayne Daughenbaugh 3:08
because because it seems it's it's all over now, and and and I am actually all for it because I mean there's there's a lot of of concerns and whatnot, but let's think about what this is doing to help us evolve, and I guess a revolution of how we do work, like AI and automation and those kind of things, are changing how we're doing work, right? Like, so, so let's talk about why is this so dangerous, and what should they do before they start automating things? Because I remember when I was working for the state, we got lots of of responses or or lots lots of engagements with with again our BTSD group that would get these requests for hey can you automate this process for me can you automate this and they pumped the brakes a lot of times and sent them our way and I'd love to hear from from you why why you think they decide to do that why did they slow things down? What were they afraid of?
Anas Saeed 4:03
So here's that saying that there's a saying that garbage in, garbage out.
Shayne Daughenbaugh 4:07
Oh come on! Okay,
Anas Saeed 4:09
AI would just amplify stuff. So that's why if a process is a mess, amplify the AI will just amplify that mess.
Shayne Daughenbaugh 4:17
Right. Okay.
Anas Saeed 4:17
And what has to happen? That's one aspect, and second aspect is that you basically have to like all the processes that were made, okay, that were currently before they were made before the age of AI. So instead of bolting on AI onto that process to automate it, you have to basically redesign the process such that the AI does the grunt work, and the humans are involved in the judgment, exceptions, relationship, and future planning. Now, if you take a air condition like a portable AC and put it in one room, it will cool down that room.
Shayne Daughenbaugh 4:49
Right. But if you
Anas Saeed 4:50
go to the restroom, if you go to the closet, if you go to some other room, it's not going to do it. Then you can't control the centrally. So that's why, like for long term, you basically have to rearchitect the building such as central air condition. So that's why the same thing is like with AI. Okay, it's like that analogy that I gave you. What you basically have to do is that you have to rearchitect that process such that it is built for the age of AI. So rather than like, so think of it this way: if you're just bolting on AI to a process, okay, that would be buying buying your employees ChatGPT or cloud licenses, but for that they could write pretty emails. But you can't have a process that automatically responds to emails, see what's important, see what's not important, and do it that way. So the main thing is that okay, instead of bolting on AI, you have to redesign it from scratch. You have to design it in such a way that humans are in charge of the judgment, relationships, exceptions, and future planning, and the AI does the grunt work, and also the fact that you have to see at the process that okay, is it a garbage process or is a great process? Is it a garbage process? Okay, AI is just going to be spinning its wheels.
Shayne Daughenbaugh 5:59
Yes, not
Anas Saeed 6:00
some of the key things to understand, and that's why okay, each of these automations has to be custom. You can't just take something off the shelf and do it because it does not. The tool does not understand your business context, what your applications are, what your exceptions are, what what can go wrong, what has gone wrong, what could go wrong. Okay, you have to have a human in the loop that could do that part. Okay, like a simple analogy I could give you is like with a car, there are three inputs: acceleration, braking, and steering. But there are millions of different. Like we would all, everyone knows that they're good drivers and they're bad
Shayne Daughenbaugh 6:37
drivers. Okay, with
Anas Saeed 6:38
Google Maps, every it tells you their directions, but still there are a lot of people who can't follow them. So the human in the loop, okay, that's the central aspect. And the biggest example of all I'll say is like, you could go to like any place has good restaurants, okay. You could go to any place and like get all the ingredients that that restaurant gets, but could you make the same food? No, I've tried, brother. I've tried. So that's why, like, you have to. That's why, if to automate a process, okay, there has to be someone who understands what it does, understands the limitations, understands the capabilities of AI, merges them together, and gets it done. That's the key aspect.
Shayne Daughenbaugh 7:14
Like, it sounds like this is a great opportunity to review and possibly renew or redesign the process because, like you said, I love how you just said it, especially with with legacy systems and just like a legacy way of thinking that this this process was designed before there is AI, so it's not gonna it's not gonna work terribly well together. So redesigning it for that, but I also loved how you mentioned that it is human centric. Like AI and automation needs to be human centric because you said it's the humans. Like the AI does the grunt work, but the humans are the ones that do the the discernment and the judgment and understand the exceptions and understand how things work. Like that that speaks to my heart in a way that I love. I love what you were saying there. So
Anas Saeed 8:02
here's what the way I'll explain it. Like, there's nothing artificial about artificial intelligence. The data is human. The algorithms are designed by humans, and the compute energy and everything is designed by humans.
Shayne Daughenbaugh 8:14
I mean, I knew this, but I never thought it that way.
Anas Saeed 8:18
That's one thing, and that's why, like, my work, okay, basically involves in helping companies become AI native, and when AI native companies is basically where the humans are involved in the judgment relationships, exceptions, but the AI takes care of the grunt work. You know, need to know what are the capabilities, what are the limitations, and how to best do it. That's the key difference. Like there used to be jobs where people used to walk around at night and light the all the street lamps used to work on natural gas, so they had to individually go and light them up. Okay, now we don't. We have electrical switches, but for the vast vast majority, okay, artificial intelligence is never gonna replace humans. It's never gonna replace jobs. It's gonna replace some small tasks, and that's why the key aspect is that artificial intelligence would greatly increase the productivity and efficiency of humans. Because if an artificial intelligence system can do this, okay, humans can use it and do something even bigger. Because, like, let me give you this simple example, okay? If you have to go from point A to point B, if you walk, you can only go so fast. You get a bicycle, you can go further and faster. You get a vehicle, you can go further and even faster. You get a jet, you can go further in faster. You get a rocket, you can go even more further and fast. But the rocket has to have someone in the driving seat telling it where to go. Right. Okay. Yeah. That's the thing that we will have to understand. So, like, especially for businesses, what will happen is that okay, you have these systems, okay, but someone has to always keep in control of it, okay. Like, if you look at the roadways today, what happens is that okay, a roadway isn't just you build it once you leave it. There's constant maintenance, upgradations, policing that's happening. The same thing is with systems. Things break down. Things have to be upgraded. Things have to be looked at. You have to see if it's being abused or not. So there's always going to be humans that are going to be involved in that loop. That's the key thing to understand. But what one thing that will happen is that things that were not possible before certainly become possible because they become feasible to do it, and only artificial intelligence allows it. That's what you have to understand.
Shayne Daughenbaugh 10:21
Nice. Now, now you've talked before about companies designing AI for for the happy path or for like the demo scenarios where when you know in the sandbox it looks like great, but the real operations are full of different exceptions, things that don't work, things that come from the side, you know, whatever it might be. How can leaders identify those exceptions before an AI solution goes live.
Anas Saeed 10:45
So one of the biggest things is that, like, here's where the design of the AI system comes in. Instead of someone in a glass office designing it, get someone from the trenches who's actually doing it every. Come on
Shayne Daughenbaugh 10:58
now, okay,
Anas Saeed 10:59
tell us. Because what happens is that okay, there's people sitting in glass offices. They have a notion or an idea of how it works, not the actuality of how it works. The perception in their mind of how it works or how it should work is way different from how it actually works. Okay, so that's why you basically for redesigning a process, the first thing is like encoding. It said that okay, you don't start coding; you start out on a piece of paper. The same thing is that with the designing a system, okay, you don't start out designing the system; you start out understanding what the system has to be designed for. So, for what you should do is that go talk to people who are actually doing it. Figure out okay what are the top 10 things that can go wrong. What's the weirdest thing that went wrong? What's the thing that okay you found that was out of the normal? Because only if you understand that you can understand what's going on. Like there's that book, The Art of War, and that there's that saying that if you know yourself and you know your enemy. You're going to be victorious in 100 battles. So you should know the capabilities of the AI system, and you know what the process works and how it's done and how it actually happens. Because one thing is that a lot of times people do it in a certain way that's comfortable for them. People might do it in a way that they find it much easier and comfortable for them. Okay, so the guidelines often exist just in paper, and that's why you have to dumb it down such that and you have in a list that some basic inter person who has no background knowledge can do it, and only then can it be implemented. That's one key thing that people miss.
Shayne Daughenbaugh 12:39
Because one of the things you started with was, hey, go and talk to the people who do the work and find out where does the system break down? Where does the weirdest exception that comes with this? You know, understanding from the ground floor, hey, that this, you know, I need to make it a system, an automation for this system. What does this system need? What are the needs of the system? What are the needs of the people that are going to be using the system? That's what I'm hearing you say. Is is trying to pull out some of those things. You know, I'm I'm assuming like looking back at some of the data. Where are some of the variations and what happened? What are the causes of those variations? Let's work those into how we're going to develop the system so that it can, I guess, overcome those. Is that is that we're looking to do? We're bringing all these other things, so it's not just a you know really simple, pretty little walkway, you know, example of like a demo. This is trying to see, hey, if if real life comes and punches the system in the face, how is it going to respond?
Anas Saeed 13:39
So one of the things is that this comes down to a design aspect. So what happens is that you have to basically do is that the first and foremost is that be aware of what could punch it in the face, and then you can design. Okay, do you want the AI system to have the capability to deal with it themselves, or you want a human to come in the loop,
Shayne Daughenbaugh 14:01
right?
Anas Saeed 14:02
Okay, that's the key. There's some exceptions that let's suppose it comes in so peculiar circumstances that you just say, okay, in this case, just call a human, and the human will take care. Okay, that's why, like I said initially, an AI native company is not one that just uses AI. It's one where the humans are in control of the judgment exceptions and relationships. So the first thing is that know what kind of things we can punch in in the face. What's not the happy path? Not the one that the people in the glass offices have in their mind, or the notion of it is in their mind. It's the one that's actuality being used. So you know know what it is, and then depending on the probability of it happening, or the percentage times it happens, or the numeric times it will happen. You either have to design the system or enhance the capability such that it could do it, or you can just tell it, okay, in this case you're not going to move forward. You're just going to call someone because that's why, like a lot of these AI systems, okay, they break down in actual in production because they are not. Designed for that, so once it messes up, people don't say that. Okay, the system wasn't designed correctly. People assume that AI can't do my work. They shut it off and go back to the old way. That's the key difference.
Shayne Daughenbaugh 15:11
Really, I didn't think about that. Yeah, because that that would bring you to just automatically make that decision. If you gave me something that didn't work for me, I wouldn't be looking at the system. I'd be just thinking, well, hey, I can't do this. It's nonsense. This is just you know the latest fad. Let's just push this aside, rather than hey, let's go back and look under the hood. What's going on with the system that didn't allow it to be able to move forward because of this, that, or the other? The data that you mentioned.
Anas Saeed 15:37
Yeah. So like in leadership, there's this concept that if someone's saying that they're not going to do this or they're not having the correct result, rather than taking on the face, you should say, okay, why is this person leading? Are there some inputs to this that's leading that person to this this result? So in the same thing, okay, instead of blaming that person on the surface level, think, okay, it might be these cases that he doesn't, that person does not have these inputs because of their. That's why they're not generating this result. Okay, it's the same thing. Okay, people sometimes tend to blame. Okay, AI doesn't work. Kick it out. Let's move on. It's too new a technology, and there's a lot of excuses. And there are some cases, and that's why like that humans, that person that actually implements it, it becomes critical because they should understand the limitations and the capabilities of artificial intelligence, and how to under implement it in their business. That's the key thing they understand. Because AI can do everything. AI does have limitations. It it needs guardrails. It needs context. It needs compute. It needs intelligence. Okay, the correct model, the correct tools and these sort of things. That's the key element to understand.
Shayne Daughenbaugh 16:44
Interesting. So, so I love how this conversation went from hey, let's automate everything to slow down. Let's you know automation is is a goal, but before you get there, let's look at how we can make this a more robust system that the AI can work in and work with. Now, my my final question for you is: as for for the listening audience and those that are watching, if if they have a process that they're considering, you know, automating with AI right now, like, hey, I know this is what we're going to do. We got some people that are working on this. What's the one thing they can do? Like, just just a step. They can let's give them one step. They can do this week that will help them determine is this process actually ready for for automation or does it need something more?
Anas Saeed 17:31
So the simple step that they could do is that write it down in a way that a six year old can understand. If you can write the process in a simple ways that a six-year-old can understand it and they can execute it, then it's ready. Because if that process is messy, it has a lot of ifs and buts and exceptions. Sure. And in this state, you do this and this and this. Okay. Yeah. All that juggling, that process is not ready for it. So write it down and have it documented and have all the data steps and everything that you could write it down and hand it over to a six-year-old so they can do it. So you know that there's a refinement technique to learn that okay, if you can't explain it to a six-year-old, you're not you don't understand it yourself. So simplify it down to how you could explain it to a six-year-old, and if you can do it, then you can implement.
Shayne Daughenbaugh 18:22
Would you suggest working with some of the legacy processes that I have with with you know businesses that have been around a long time? It's almost like they cobble systems together, systems and processes together. This one brings it over here. Would you suggest in this, if if there are a lot of exceptions, a lot of if ands, you know, and all of that that make the process really complex, and they can't explain it to a six-year-old, would it be beneficial to actually break that process up into smaller processes that you then can automate to get still the same result you want, but because it's so complicated and complex as a whole, that you know what? Let's take this part off. Let's take this part off. Would is is that something? Is that is that reasonable to think about?
Anas Saeed 19:09
So there are two aspects I'll say in this case. The first thing is that the system that you should design should work as a complement to the these legacies.
Shayne Daughenbaugh 19:18
Oh, okay.
Anas Saeed 19:19
Okay. Because here's the thing. In practice, okay, it sounds really good that let's suppose there's a business for 50 years, okay, and they have a lot of legacy systems that could have COBOL, that could have .NET, that could have Python, Java, and all these differences. Now, the what will happen is that if you basically like, hey, I'm gonna make it AI native and do this, there's gonna be something that's gonna break. Yeah. Okay. that's why what's essential is that in that case you should design a system such that it can work in conjunction with those systems. Number one. Yeah. Number two, what you should do is that you just don't create the system, deploy it. You have to continuously monitor and see how it works. The third thing I would say is that what you should do is that that's where the design element of the system comes in. AI is just the technology that you're going to use to implement. There, there could be other technologies, for example, IT ones where you basically have to code something to make it customized. That's the one thing. So what you do is that in this kind of system, you'd see what's the kind of system that could actually be implemented. Because if the whole thing can't be implemented, what you're suggesting is that see if a sub part of that could be implemented. You see the benefits, and then you have a much bigger deployment where you do the whole system. Okay, because the trust of the customer and the people who are using it is of the critical and the best, the most important thing.
Shayne Daughenbaugh 20:44
Sure. If you
Anas Saeed 20:44
design a system that doesn't work out and does a small flaw just for for whatever reason, people are going to lose trust, and once they lose trust, they're going to stop using it.
Shayne Daughenbaugh 20:52
Right.
Anas Saeed 20:53
That's why in that case, the first thing is that it should be work as a complement, even after deployment. It should be the fact that you continuously monitor to see if it's done correctly. Okay, that's why you should never use tools, and you should have. There's a difference between tools and systems. Okay. Okay. In this case, you should not have a tool; you should have a system. The third thing is that okay, you should see where I can be successful and use it just to that capability. Those would be the three critical levels,
Shayne Daughenbaugh 21:23
Doctor Sayed. This has been so much fun for my brain to to try to try to wrap around and engage with. If if other people, if some of our listeners or those that are watching, want to get a hold of you or understand a little bit more about your work, where could they reach out to?
Anas Saeed 21:40
So go to my LinkedIn or my website and basically reach out to me. You can just search my name, and I'm sure you'll be able to find
Shayne Daughenbaugh 21:48
it. Thank you so much for coming on on the episode with me, and and for you know talking this out with me. Those of you that are watching, we hope that you got something from this, at least that little nugget to move forward with. And we look forward to seeing you on the next episode. Thank you so much. Have a great day.