Chalk X

Summary:

• Using the Cynefin sensing model, we can better understand how to use AI in different situations while recognizing its limitations and strengths in each. 
• For simple problems, AI generation is more expensive and prone to inconsistency when automation could be used instead. 
• For complicated problems, The AI user’s understanding of the situation is critical to effectively using AI. 
• For complex problems, AI cannot build a good product or product feature, but it can help us in ways to discover them. 
• Knowing when and how to leverage AI successfully can maximize its usefulness and minimize inefficiency and risk.  

Cynefin: A Sensing Model

What kind of problems does AI solve well? To answer this question I first want to bring back up the Cynefin model.

This is a sensing framework in which situations can be analyzed, and we can respond accordingly. For a short recap of my article, Complexity is Key, here is how to use the sensing model diagrammed above. 

• ? – We walk into our kitchen in the morning and we are hungry. The red question mark is where we start within the model, we may not know what situation we are in. We always start in the unknown.
• A simple situation is that we walk into the kitchen hungry and decide to make toast, but the toaster isn’t working. We see that the plug is not in an outlet, we respond by plugging it in. 
• A complicated situation is that we walk into the kitchen hungry and decide to make toast. However, the plug is in the outlet. We check to see if anything is stuck. We change outlets. We grab a voltmeter and determine where the electricity isn’t getting to in the toaster. We choose the broken part and replace it. 
• A complex situation is that we walk into the kitchen hungry and with guests at home. What may their food allergies be? What are they hungry for? We ask a series of questions, along with what we may have available and the ability to cook in our kitchen, and we come up with a plan of what to make. 
• A chaotic situation is that we walk into the kitchen hungry, and the kitchen is on fire. We quickly grab a glass of water to throw on the fire. Sadly, this was an electrical fire, and we were electrocuted. Slipping on the now-wet floor, we fall and try to grab ahold of the counter. Missing the counter, we accidentally grab the wet, burning toaster and pull it down on top of us as we land unconscious on the kitchen floor. 

Note that in each of these situations, we started in the same “problem space:” hungry and going to the kitchen. We need to respond very differently based on the environment and actual situation. 

Perhaps the last one is because I watched too much of Tom Hank’s 1980 classic movie, “The Money Pit.” In this movie, Tom Hanks and Shelly Long buy a home only to find a series of problems with it. If you haven’t seen it, here is a scene from it to get a sense of the movie.

It is also a great example of chaotic problem-solving. Mostly because Walter Fielding, Tom Hank’s character, has bought a home that has way more wrong with it than he thought. He thinks everything is a simple fix and only deals with surface issues. As a result, his situation moves into the chaotic space but he is acting like it is in the simple space. This is a good example of how to use the Cynefin model. Knowing what kind of situation you are in helps you address problems more effectively. 

How to Best use AI in Various Situations

Let’s examine each space and discuss its pros and cons as they may apply to the situation above and how that extrapolates into business scenarios.  

1. In the simple space we ask AI, “How to fix a broken toaster simply.” The first thing AI tells us is:
   •  Check the Outlet: Make sure the toaster is plugged in securely, and the outlet is functioning. Try plugging in a different appliance to see if the outlet is working.
   • Pros:
     • It gave us several simple solutions, including the one that worked.
     • The interface was simple.
   • Cons:
     • In a work environment, these simple solutions can sometimes be called the best-known methods. This could have been a standard checklist. Do we need to spend time and money using AI, whereas a simple checklist would have allowed us to solve the problem? 
     • AI does not give consistent results, which is part of its strengths but also a weakness. Consider if you used AI to come up with how to put a car together each time. In addition to being expensive, it would be inconsistent. It creates variable quality and outcomes, making repairing cars much harder. As a result, AI helps create a checklist if you don’t know it, but it shouldn’t be used every time the situation arises.

AI can help with generation, but once the checklist/recipe is known, automation is preferred due to cost and quality.

2. The complicated space. Here we tell AI. “None of those suggestions work. As a master electrician, tell me how to diagnose and fix my issue.”
   • What AI tells us? Please only try these steps if you’re comfortable working with small appliances and have unplugged the toaster to avoid electric shock. Test Continuity of the Power Cord: Using a multimeter, check for continuity across the two prongs of the unplugged toaster plug. Set your multimeter to the continuity or resistance (ohms) setting. If there’s no continuity, the cord might be broken, and you’ll need to replace it.… 
   • Pros
     • It has a simple interface and is easy to use. 
     • It gave me a lot of things to try that I may not have known before. 
     • It does warn me I may hurt myself if I try some of these. 
   • Cons
     • The warning about hurting yourself may not come with corporate usage. It definitely won’t come with software usage, which will generate code or outcomes that could put your company at risk. See my article on High-Volume Waste. Quickly, our corporate outcomes can become Tom Hanks in the Money Pit. This means we think we are in a complicated space, but without our knowledge, we are moving into the chaotic space as we build the software/device/product equivalent of a money pit. 
     • All the cons from the simple space apply here as well. 
     • Reading something and knowing how to use it are two different things. AI presumes I know how to use a multimeter, and not every reader will. Misusing it may yield inconsistent results. AI is not part of the full sensing situation.

When using AI for complicated problems, an expert that knows what “good” looks like needs to check it to prevent waste, danger, and chaos.

3. In the complex space here, we ask AI, “Once I fix my toaster, what do my friends want to eat?”
   • What AI tells us: Once that toaster’s up and running, here are some classic breakfast ideas your friends might enjoy: Avocado Toast with Poached Eggs – A trendy, filling option that’s easy to customize with toppings like chili flakes, feta, or smoked salmon…. What kinds of flavors do your friends usually go for? 
   • Pros:
     • It gave me a series of very plausible things to make with toast. If I ask for a follow-up recipe, it can quickly provide it. In fact, I can ask for ten different recipes for Avocado Toast with Poached Eggs, and I will get ten varieties of them. 
     • It does ask what my friends like. If I follow up with, “What questions should I ask my friends before making something?” It will ask the questions I would think to ask and even some I may not have thought of, like what time constraints my friends may have. 
   • Cons:
     • AI has no idea what I have in my kitchen or what I can make. I can add those things in, and it will help, but when it says Poach an Egg. I may or may not know how to make a Poached Egg consistently. Writing effective security software, waterproofing a boat, or any other complex task does take skill to follow. Skill that cannot be gained by reading AI outputs.
     • The one that should jump out is that AI has no idea about my friends and cannot answer for them. This is true of customers as well. AI doesn’t know what our customers will want or pay for. 

AI doesn’t know our capabilities or what our customers will use, and it should not be used in isolation to create features or product ideas.

4. In the chaotic space here, we ask AI, “My kitchen is on fire. What should I do”
   • What AI tells us: Stay Calm and Act Quickly. Try to stay calm. Acting quickly and decisively is key to keeping everyone safe… 
   • Pros
     • It lists out the correct courses of action in a fire, including not using water on grease or electric fires. 
   • Cons
     • If you are trying to read an AI response with your kitchen on fire, you are more doomed than Tom Hanks in The Money Pit. 

Do not try to use AI in a chaotic situation; it may be accurate but unhelpful.

Final Thoughts

One of the keys that stands out is that AI doesn’t know its situation. You have to tell it. AI will lead you down the wrong path if you guess wrongly or presume an orderly system. It doesn’t know the actual situation, environment, resources, people, and problems surrounding you at any moment. 

AI can be quite helpful right out of the box when used for simple questions, finding best known methods, brainstorming, or helping you find blind spots. 

Different AI large language models have different strengths. Still, when using AI in a complicated domain, you need to provide as much information about the problem as possible. Ask AI to ask you more questions before responding. This may help get a better reply, but an expert should check the result before using it blindly. 

Complicated problems are a spectrum, and for recurring situations with a more complicated background, unique environment, or where your company has a competitive advantage, I recommend building a RAG (Retrieval-Augmented Generation) model. A RAG model can use specific kinds of data to let you add your particular expertise to the already powerful large language models. For example, if your company specializes in a specific type of coding language where you’ve built expertise over the years, moving that data into a company-specific model can secure proprietary knowledge and uniquely accelerate your developer’s productivity. These models and implementations are not free and unsuitable for every situation.