AI Foundations Course
Lesson 3 of 9
Estimated learning time: 6–8 minutes
What You’ll Learn
By the end of this lesson, you will understand:
- What traditional software does
- How machine-learning-based AI differs
- How rules and data play different roles
- Why AI is useful for complex or difficult-to-define tasks
- Why traditional software is still extremely important
- Why modern applications often combine both approaches
Introduction: Are AI and Software Different?
When people hear the word AI, they sometimes imagine that AI is completely separate from ordinary software.
It isn’t.
AI systems are built using software, algorithms, data, and computing infrastructure. The important difference is often how the system obtains the logic it uses to produce results.
In conventional software, developers generally write explicit instructions and rules.
In machine learning, developers design a learning system and provide data so that the model can learn statistical patterns that can later be used for predictions or other outputs.
A simple comparison is:
Traditional software:
Input + Rules → Output
Machine-learning system:
Data + Training → Model → New Input → Output
Let’s make this easy to understand.
1. How Traditional Software Works
Imagine you are building a simple calculator.
You don’t need to show the computer thousands of examples of addition.
You simply give it instructions for how addition works.
For example:
Input: 10 + 5
Rule: Add the two numbers.
Output: 15
The developer has explicitly defined the logic.
Traditional software can be extremely powerful because computers are very good at following precise instructions.
Other examples include:
- Calculators
- Payroll systems
- Accounting software
- Inventory systems
- Website forms
- Many business applications
- Operating-system functions
When the rules are clear, conventional programming can be an excellent solution.
- What Happens When the Rules Become Difficult?
Now imagine you want to write a program that can recognize whether a photograph contains a cat.
You might try to create rules:
If it has two eyes, pointy ears, fur, whiskers and a tail, call it a cat.
But immediately you encounter problems.
Some cats have different-shaped ears.
Some photographs are dark.
The animal might be partly hidden.
A dog might also have fur, ears and a tail.
A person might be wearing a costume.
Suddenly, writing a rule for every possible situation becomes extremely difficult.
This is where machine learning can become useful.
- How Machine Learning Approaches the Problem
Instead of manually writing every possible rule, we can provide a machine-learning system with many examples.
For example:
Cat image → Cat
Another cat image → Cat
Dog image → Dog
Another dog image → Dog
During training, the model learns statistical patterns from the examples.
NIST describes machine learning as systems that adapt and learn from data, with the goal of improving accuracy.
After training, the model can receive a new image and produce a prediction.
The developer hasn’t manually written a separate rule for every possible cat photograph.
4. The Key Difference
This is the most important idea in this lesson.
Traditional software
The developer specifies the rules.
Data + Programmed Rules → Output
Machine learning
The developer creates the learning system and provides training data.
Training Data + Learning Algorithm → Trained Model
Then:
New Input + Trained Model → Prediction/Output
This is why machine learning can be particularly useful for problems where it is difficult to write precise rules for every situation.
5. A Simple Email Example
Let’s imagine a spam filter.
Traditional rule-based approach
A developer might create rules such as:
If an email contains certain words, mark it as spam.
Another rule might say:
If the sender appears on a blocked list, mark the message as spam.
This can work well for clearly defined situations.
But spammers can change their wording.
They can use different spellings.
They can change addresses.
They can create new patterns.
Machine-learning approach
A machine-learning spam filter can be trained using many examples of:
Spam emails
and
Legitimate emails
The model can learn patterns that help it classify new messages.
This doesn’t mean it will always be correct. It can still make mistakes.
That’s an important lesson we learned in Lesson 2.
6. Traditional Software Isn’t “Old” or “Bad”
This is something beginners sometimes misunderstand.
The comparison isn’t:
Old software = bad
AI = good
That’s not true.
Traditional software is still the right choice for many tasks.
If you need a system to calculate a salary according to clearly defined rules, traditional programming may be more appropriate than machine learning.
If you need to calculate taxes according to a specific formula, explicit rules are often desirable.
If you need a system to follow a fixed sequence of operations, conventional software can be reliable and easier to test.
The best technology depends on the problem.
7. AI Isn’t Always the Better Choice
More AI doesn’t automatically mean a better system.
Machine-learning systems can require:
- Large amounts of relevant data
- Computing resources
- Careful evaluation
- Ongoing monitoring
- More complex development and maintenance
NIST notes that machine-learning systems can be versatile but may require substantial data and computational resources.
So before adding AI to a project, an important question is:
Do we actually need AI for this problem?
Sometimes the simplest solution is the best solution.
8. Modern Systems Often Combine Both
Here’s something very important:
It isn’t always AI vs traditional software.
In many modern applications, the two approaches work together.
For example, an application might use conventional software to:
- Manage user accounts
- Store information
- Handle payments
- Control access
- Display webpages
At the same time, it might use an AI model to:
- Understand natural language
- Recommend content
- Detect patterns
- Generate text
- Analyze images
So a modern product can contain both traditional programming and AI components.
This is often a practical way to build useful systems.
9. A Simple Side-by-Side Comparison
Traditional Software
- Developers write explicit rules
- Works well with clearly defined logic
- Behavior is often predictable
- Rules are changed by editing software
- Usually easier to explain for simple tasks
Machine-Learning-Based AI
- Models learn patterns from data
- Useful for complex pattern-based tasks
- Outputs may be probabilistic
- Behavior may change through retraining
- Some models can be difficult to interpret
This is a simplified comparison. Real-world systems can combine both approaches.
10. The Best Question Isn’t “Which Is Better?”
Instead of asking:
Is AI better than traditional software?
Ask:
Which approach is appropriate for this particular problem?
If the problem has clear, precise rules, conventional software may be ideal.
If the problem involves complex patterns that are difficult to describe with explicit rules, machine learning may be useful.
And sometimes the answer is:
Use both.
That mindset will help you make better decisions as you learn more about AI.
Try It Yourself
Think about these three tasks:
Task 1
Calculate the total price of five products.
Would you use traditional software or machine learning?
Probably traditional software.
Task 2
Determine whether a photograph contains a particular object.
Could machine learning be useful?
Yes.
Task 3
Build an online shopping application that recommends products.
Could both approaches be useful?
Absolutely.
The website, accounts, payments and database might rely heavily on conventional software, while the recommendation system could use machine learning.
Remember This
The simplest way to remember today’s lesson is:
Traditional software
Humans define the rules → Computer follows them
Machine learning
Humans provide data and design the learning system → Model learns patterns → Model produces outputs
And remember:
AI doesn’t replace traditional software. AI is built with software, and many useful systems combine conventional programming with AI.
Key Takeaway
Artificial Intelligence and traditional software are not enemies or competing technologies.
Traditional software is excellent when we can clearly define the rules.
Machine-learning-based AI becomes especially useful when we need systems to learn patterns from data and handle problems that are difficult to describe with fixed rules.
The real skill is knowing which approach—or combination of approaches—is appropriate for the problem.
← Previous Lesson
Lesson 2 — How Does AI Work?
Next Lesson →