Do you know the difference between the AI that writes your emails and the AI that forecasts your sales? If you’re not sure, you’re not alone. Three weeks ago, neither was I.
I’ve used AI since before ChatGPT made it a household name. I remember using Grammarly back then to proofread my writing. Now AI helps my team draft guest replies, marketing content, reports and presentations, saving us hours every week.
So when I enrolled in Strategic Leadership with AI at Singapore Management University, I expected a refresher. Maybe a few new tricks here and there.
Instead, I got a reality check: I knew how to use AI, but I didn’t know what it actually is. That gap matters, because it shapes every decision we make about where to invest, what to trust and what to keep human.
Here’s what I learned, in plain terms. I hope it helps my fellow leaders.
AI is not one thing
My biggest assumption was simple: AI is AI. I thought the tools all did more or less the same job.
In our first session, our program leader, Atul Bengeri, used the story of the blind men and the elephant. One touches the trunk and calls it a snake. Another touches a leg and calls it a tree. Each is right about his part and wrong about the whole. Most of us experience AI the same way, through the one tool we use most.
In reality, AI is a family of different technologies, each built to solve a different kind of problem. And they all run on one thing: data. As Atul put it, without data there is no AI, only automation.
The main types of AI, in plain terms
AI grew in waves over 70 years. Each wave was shaped by the computing power and data available at the time, and each fixed a weakness of the one before. The earlier types didn’t disappear. Most are still at work in businesses today.
| Type | What it does | Where it helps a business | Its limit |
|---|---|---|---|
| Rule-based AI | Follows fixed if-then rules written by people | Approval workflows, auto-replies, standard checks | Breaks when something falls outside the rules |
| Expert systems | Captures a specialist’s knowledge as rules | Diagnosis, product configuration, compliance checks | Slow and costly to build, and can’t learn |
| Supervised learning | Learns from past examples with known answers to predict outcomes | Revenue and demand forecasts, likelihood to buy or cancel, fraud detection | Only as good as the historical data |
| Unsupervised learning | Finds hidden groups and patterns without labels | Customer segments, unusual transactions | Finds patterns, but people must judge what they mean |
| Reinforcement learning | Learns through trial, feedback and reward across a series of decisions | Pricing or inventory decisions that play out over time | Hard and costly to train |
| Generative AI | Creates new text, images or code from patterns in vast data | Drafting, summarizing, research, customer replies | Sounds confident even when wrong |
Underneath the last few sit neural networks and deep learning: layered systems, now with billions of connections, that made today’s generative AI possible.
One story from the module stayed with me. In 1964, MIT built a chatbot called ELIZA. Type “I am feeling sad” and it replied “Tell me more about feeling sad.” People thought it was remarkably intelligent. It was only matching patterns. Today’s tools are far more advanced, but the lesson holds: generative AI predicts the most likely next word. It doesn’t make decisions.
How this helps you make better decisions
Once you can tell the types apart, AI strategy gets much clearer. These are the lessons I’m taking into my own businesses.
Start with the problem, not the tool. An MIT Sloan article in our readings found that many companies chase AI as the next flashy project instead of starting from a clear business problem. Decide first what you need: to follow a process consistently, predict an outcome, understand your customers, or create content. Then choose the kind of AI that fits.
Don’t expect a chatbot to forecast. Claude and ChatGPT are generative AI. They’re excellent for drafting, summarizing and brainstorming. They aren’t built to forecast your bookings or flag fraud. That’s machine learning, trained on your own data.
Separate tools from solutions. The same article distinguishes AI tools that save individuals a few minutes from AI solutions built into processes that move revenue or costs. Most of us are still using tools. The bigger strategic value is in solutions.
Get your data in order. Every type of AI beyond simple rules depends on data. If your data is scattered, outdated or incomplete, your AI will be too.
Keep people on the decisions. AI can inform a decision. It shouldn’t own it. When my team first trusted AI answers, some started copying replies without checking them. Now verification is mandatory. In luxury hospitality, AI can draft a reply in seconds, but it can’t read what a guest doesn’t say.
Before your next AI investment, ask
- What problem are we solving, and how will we measure the result?
- Which kind of AI fits it: rules, prediction, pattern-finding or generation?
- What data does it need, and who keeps that data accurate?
- Which decisions stay with people, and who checks the output?
Staying a student
Being an early adopter made me comfortable, and comfort is where blind spots grow. If you’re a leader who already uses AI every day, you may know less about it than you think. I did.
Understanding the basics won’t turn you into a data scientist. It will make you a sharper buyer, a better judge of risk and a clearer decision-maker.
Three weeks down. Still learning. Keep learning.