Day 3 - Problems Where ML Fits vs. Problems Where It Doesn't

Introduction

Machine learning is powerful, but it is not a universal tool suited to every problem. Today we will learn the criteria for distinguishing problems that suit machine learning well from those that do not, and compare how it differs from the traditional programming approach.

This judgment is very important in practice, because it is common to "adopt machine learning where it wasn't needed, only to increase costs and get worse results." The exam also frequently asks questions like "Is machine learning appropriate for this scenario?"

Traditional Programming vs. Machine Learning

First, let's sort out the fundamental difference between the two approaches.

In traditional programming, "a human writes the rules and the computer executes them as-is." In contrast, the key difference with machine learning is that "given data and answer examples, the computer produces the rules (a model)."

💡 Related theory: Traditional programming is data + rules → results, while machine learning is data + results → rules. It's easy to remember if you think of the inputs and outputs as swapped.

Problems Where Machine Learning Fits

The more of the following conditions are met, the better a choice machine learning is.

1. The rules are too complex or too hard for a human to write out one by one
   • Example: It is practically impossible for a human to write code rules for recognizing a cat in a photo.
2. Patterns exist but are hard to express explicitly
   • Example: The characteristics of spam email keep changing and are subtle, making them hard to pin down as fixed rules.
3. Sufficient quantity and quality of data is available
   • You need plenty of historical data (examples) to train on.
4. Some margin of error is acceptable
   • Machine learning is not 100% accurate. It gives probabilistically good answers.
5. The rules must change as circumstances change
   • The model can be retrained on new data to adapt to change.

Suitable Examples

• Image/speech/natural language recognition
• Recommendation systems (products, movies)
• Spam and fraud detection
• Demand and price forecasting
• Customer churn prediction

Problems Where Machine Learning Doesn't Fit

Conversely, in the following cases machine learning is overkill or inappropriate.

1. The rules are clear and simple
   • Example: "VAT is 10% of the amount" is just multiplication. No machine learning needed.
2. There is almost no data
   • Without enough examples to learn from, machine learning cannot work properly.
3. 100% accuracy and complete explainability are mandatory
   • When legal or safety requirements allow zero error and every decision must have a clearly explained basis, machine learning — probabilistic and sometimes a "black box" — can be risky.
4. Simple calculation or lookup is sufficient
   • Database lookups, fixed formula calculations, and the like are more accurate and cheaper with the traditional approach.

Unsuitable Examples

💡 Related theory: Avoid the misconception that "machine learning is always smarter." Applying machine learning to problems with clear rules can increase costs and actually reduce accuracy. Choosing the simplest method that is sufficient is good design.

Decision Flowchart

Here is a quick order of checks for deciding whether to adopt machine learning.

Can a human write the rules easily and clearly?
   └ Yes → Use traditional programming (machine learning unnecessary)
   └ No ↓
Is there enough data to train on?
   └ No → Acquire data first (machine learning is difficult for now)
   └ Yes ↓
Is some margin of error acceptable?
   └ No → Review carefully (machine learning may be risky)
   └ Yes → Machine learning is likely a good fit

Striking a Realistic Balance

In practice, the two approaches are often used together. For example, clear rules are handled with traditional code, and only the complex judgments are delegated to machine learning. What matters is the perspective of "picking the right tool for the problem," not "only one of the two is correct."

Today's Summary

• Traditional programming is data + rules → results; machine learning is data + results → rules.
• Machine learning fits when the rules are complex, data is plentiful, and some margin of error is acceptable.
• It can be a poor fit when the rules are clear, data is scarce, or 100% accuracy is mandatory.
• Choosing the simplest method that is sufficient is good design.

📝 Practice Questions

Click a choice to reveal the answer and explanation.

Question 1

For which of the following problems is traditional programming more appropriate than machine learning?

• ARecognizing cats in millions of photos
• BDetecting ever-changing spam email
• CCalculating tax by multiplying an input amount by a fixed 10% rate
• DRecommending products based on a customer's purchase history

Question 2

Which statement most accurately describes the input/output relationship of traditional programming and machine learning?

• ABoth take data and rules as input and output results
• BThe traditional approach produces results from data + rules, while machine learning produces rules from data + results
• CMachine learning takes rules as input and generates data
• DThe traditional approach produces results without any data

Question 3

In a certain business task, there are only a handful of historical records available for training, and the results must be legally 100% accurate and fully explainable. Which judgment is most appropriate for this situation?

• AAdopt machine learning immediately, since it is always more accurate even with little data
• BMachine learning may be unsuitable due to the lack of data and the requirement for complete accuracy and explainability
• CUsing reinforcement learning solves it perfectly even without data
• DThere is no problem since machine learning always guarantees explainability

Question 4

Which of the following is hard to consider a characteristic of problems suited to adopting machine learning?

• AThe patterns are so complex that humans struggle to write explicit rules
• BSufficient quantity and quality of training data is available
• CSome prediction error is acceptable
• DThe rules are very simple and unchanging, and a simple calculation suffices