Welcome back. I'm Kence Anderson, let's talk about the value of the problems that you're going to be solving with autonomous AI. Here's how they engagement with autonomous AI usually starts. Either someone has a problem that they can't solve, and they hear that AI might be able to help, or someone hears about AI and wonders whether they have problems that might match up to new capabilities. How do you determine when to use autonomous AI? Should you replace all automation with autonomous AI? I get that question all the time. The answer is no. Some processes operate really well with their existing control systems. No need to fix what ain't broke, so to speak. Other processes are difficult to manage, but autonomous AI is not going to be able to do any better. We can generate great results with AI for still other processes, but it just might not be worth it. The return on investment doesn't justify the cost to design, build, and implement the AI. Machine teaching, is about pairing the right capabilities to high-value decisions that will radically improve systems and processes that you're working on. Whether you're trying to optimize an industrial process, make a dent in climate change, or whether you're attacking some other societal problem through better decision-making, machine teaching puts the right capability in the right place. Let me show you how to evaluate systems and processes to determine whether autonomous AI is a good investment. You don't need to replace every control system, optimization algorithm and expert system with an AI brain. You don't need autonomous AI for every application. Autonomous AI as a strategic investment that is best suited to high-value problems. How do you define high-value? It boils down to whether the improvement you can make with AI really matters or not. Almost every brain I've ever designed is for a system or process where a 1 percent improvement in the key performance indicators leads to over 1 million US dollars in savings or increased revenue for one facility. That matters. Key performance indicators are the goals that drive the process. Let me give you an example. I consulted a mining company about a process where a 1 percent increase in throughput, that's the amount of rocks crushed by a rock crusher every hour, was worth 15 million dollars per year at one mine site. The AI that we designed and built generated a 7 percent improvement over the existing control system and simulation. That's a lot of value. Financial return on investment is the most obvious way to identify a high-value problem, but it's not the only thing that counts. There's different kinds of value, monetary value, social change value, justice value are just a few examples. I worked with a bank to design an AI that control the robotic arm to lift bags of coins onto counting tables. Here, the value is not simply monetary, the problem we were trying to solve was that the bank personnel sometimes was getting hurt, lifting the heavy bags of coins onto the counting tables. The value that AI generated by controlling the robotic arm was an injury prevention and improvements to the health of the bank workers. See how we can generate value beyond monetary measures with autonomous AI. Here's another example. Microsoft built and deployed in an autonomous AI to its Redmond West Campus that controls cooling systems, and saves 10 to 15 percent energy over existing methods. Yes. That's monetary value, but it generates even more value towards Microsoft's climate goals. 50 percent of all energy in buildings, that's all buildings everywhere, not just at Microsoft is used by heating, ventilation, and air conditioning systems. If we reduce building energy by 15 percent, that's tremendous progress toward slowing climate change. Selecting a high-value problem is one key to identifying a good use case for autonomous AI. There's two other factors that I want to cover too, though. How limited is current automation and how much can autonomous AI help improve the situation? Let's say that you have high-value problem. Maybe the value is monetary, or maybe it's more about reducing emissions or helping people. Great. The next thing to consider is how good are the decisions that are already being made? If existing methods have already improved the process as much as they can, then we move on to the final consideration. The last thing to consider is, can the problem benefit from the unique capabilities of autonomous AI? Here's how you know, does the problem still rely on human decision-making? That's a sure sign that autonomous AI can improve the situation. In fact, that's one of my favorite questions to ask while discussing use cases. What are the humans doing here? Why can't the automated system make the decisions on its own? Often there's an automated system that works really well at times, but humans have to step in, when the automated system makes bad decisions. If the existing automation can't do much better than it already does, and if evidence suggests that AI can do better, that's how you harvest all that value you identified in the first step. Great use cases for autonomous AI generate high value and processes where human-like decision-making is required and traditional automation sometimes fails. You'll learn about the limitations of traditional automation in analyzing the problem, and you'll learn about the unique capabilities of AI, what we call the superpowers of autonomous AI in learning the solution. For now, what you need to know, where these three conditions intersect is where you find the best use cases for autonomous AI, and machine teaching helps you unlock the value. The last thing to consider is the amount of effort it will take to build an autonomous AI system. We'll discuss that in detail in Courses 2 and 3.
