Algorithmic AI Systems: an overview…

All right, this video is going to be about how you actually build AI systems that are super useful and are probably going to be uh how the future of you know the the death of the white collar worker actually happens and it's going to be through stuff like this.

So, if you remember a while back, I made a video on how to make AI systems feel like AGI.

And the whole point of that video was you had to have a a single voice agent that speaks to the uh the human and then behind that disguise behind that one voice of that agent, there are many different channels and routes of of you can think like a a submployee architecture structure for for that agent to actually go and operate.

So, it's it's not one AI agent that has access to maybe five different tools.

Instead, it's one AI agent that has access to 15 different AI agents who are all specialized to do certain tasks.

Um, and now we're going to go even a bit further down and kind of discuss, well, what happen?

How do we actually make those 15 other AI agents accurately make decisions and analyze different stuff in the business so that that one executive voice agent that the human is talking to is super useful and it comes through this.

So recently I I built a system where uh from just spoken conversation it would output a fully 180 between 150 180 line item task schedule for a new residential construction build.

And essentially what that system did was the the LLMs that I used in there, instead of actually having them create the tasks for the schedule, I instead had the LLM parse the uh the voice inputs from the contractor and turn those into variables to plug into an algorithm or an equation that then made the decision on the business's behalf.

So I'm kind of I'm starting to shift my my way of of building these systems from good prompting to good algorithm designing uh around the around the problems of the business themselves.

So um right now the the point in the AI revolution of where we are is people like me are building these systems that are coming for office workers jobs.

If if you do any type of intelligent uh reasoning or decision- making, uh these are the systems that we're first going to build to to kind of in a way um analyze all of the inputs and already have algorithms to to essentially give you a like a list of five to 10 what if scenarios that are accurately scored with with probabilities.

So this example I have here is the actual the game engine that I'm building for Opulence, our software as a service. And you'll see here each of these six different categories in purple are different aspects of a general contractor's mental burden that they have to think through.

So scheduling and sequencing in a in a traditional GAN chart or Microsoft Excel, you have a resource and crew management, weather and environmental impact and issues, materials and procurement, financial financial bidding uh and intelligence, and then site safety and qual and and quality metrics.

So, there's going to be more stuff than just these ones I've listed, but these are kind of the way that I could group out the six different mental thinking burden that that a good contractor goes through.

And now in black in all of these, these are different algorithms or equations that I'm essentially going to build and design to to hand over to an AI agent or an AI cluster um to use as tools.

So all of these you can really look as as they're going to be tools for AI agents because I don't want I don't I don't trust at least where the the AIs are right now and usually maybe in the future after it already gets you know general intelligence.

I want their their recommendations for business decisions that need to be made.

They need to come from math. You know math is a way of modeling reality.

Math is a way of modeling reality and as we humans do, we like to predict things.

So AI is an extension of our of our thinking power and right now it's it's too stupid and to to just based on the already ingrained math it has going on its head.

Um it's too stupid to to make reliable decisions in my humble opinion.

So instead, I'd rather have I'd rather I'd rather utilize the true power of this technology, which is uh understanding and and making meaning from messy unatategorized uh voice input or or words or any type of data and then translating that to being the Rosetta Stone for variables to plug into each one of these algorithms. And then based on the outputs of those algorithms, re offering up or reshowing that now you know calculated information back to the human in the loop so the human can make the decision.

Um so you can imagine let's go for an example. So let's say the weather and environmental impact uh part of the game engine weather delay probability.

So this algorithm would be okay looking at our schedule what tasks are done this week the weather um you know is it raining this week is it whatever what is the per what is the probability that we're going to get a weather delay and uh which tasks are on more of a a more of a high lookout than others.

So the AI would would run that either automatically or on the request from the user speaking through it via voice and you know the agent would then go and activate this this part of its brain or this part of its cluster.

It would then run this. It would put in uh all of the data input inputs that it needs and then this would output something like I don't know 0.81.

So like be 81% chance. And then we can basically say if if anything's higher than 77% chance then you need to indicate to the um to the contractor which tasks are going to be affected like the probability of that based on based on the weather. And then what gets really cool is okay so now that we have this information like oh my gosh my concrete pore is going to be affected by this weather.

Okay. So that means as a contractor tra traditionally you would just have to kind of sit there and you know look at your Excel and and kind of exist in your own thoughts and and think of all the possibilities. But now you can have a conversation with your schedule with your buddy and you can basically go and send this over to the um the scheduling and sequencing sequencing engine.

Oh actually no you would actually send it to here. I forgot I I wrote an algorithm in this bucket. So then you would send the agent would then send it and say uh okay hey weather task rescheduler.

Um how do I reschedu my my schedule and not screw up everything? You know I don't want to push out the the critical deadline.

I don't want to I want to have minimal damage to my schedule and not not affect any of the subcontractors or or the other work. How can I do that?

And this again this is another algorithm.

It's another equation where I have to figure out the inputs of it and basically the AI will recommend three to four to five different options that the contractor can do. Um, and then the the the contractor just engages in this dialogue where it's like, you know, I really like it that idea, but you know, I'm thinking based on my experience that maybe we should go this direction.

So maybe he can modify these certain plans that are given. It can just say, I like that one.

Go commit it to the schedule.

And if he says go commit it to the schedule, then he can go and send it over yonder to the schedule build. Oh, not the schedule builder to the um to over here like over here.

This would be the part of the brain that actually commits changes to the schedule.

So, um, yeah.

So, your goal as an AI engineer, if you're building, if you're building a a thinking executive agent buddy for some type of niche or business, your job is to become an algorithm builder. Um, part of your human intelligence is best utilized with taking reality, taking all of the potential variables that could affect a decision or an outcome in a business and translating that into an equation or an algorithm that based on the output of that equation would signal what the best decision is to do or or or maybe not the best but what is the better alternatives.

Um, and this is the core of what I call algorithmic AI building.

Um, or algorithmic agent building, where the agents are not the ones making the decision.

The agents are the one essentially doing the research and using their algorithms as tools to go and then present all that all of their research and their their good hard work to um the CEO, which now is all of us humans.

Um, and it's very it's very much the same thing that happens in the working world, right?

like you go send your market analysis people or you go get consulting and and they do a lot of this work and those consultants they don't have, you know, necessarily equations in their head, but they're doing they're doing physical mental labor in the world that can be modeled by some type of math equation.

um you just need to know the inputs and you need to know what they're doing in their head like what are they actually thinking about reasoning about and then you need to turn that into a algorithm and AI is really good is really good with helping with that.

So if you say you know I'm not good with math um I don't want to hear it because I'm not either but I still make it work.

Yeah, that's kind of going to be all.

I'm going to start tackling these one by one.

Um, we already have a lot of the schedule builders built, but um, there's a lot more to build.

There's like, you know, 40 different algorithms here I need to build.

So, I'm going to be in my lab building.

Uh, yeah, comment if you have any questions or thoughts. Um, and yeah, keep keep on keeping on.