Artificial intelligence, machine learning, and deep learning are closely related ideas in computer science, but they are not the same thing. Artificial intelligence, or AI, is the broad goal of making computers perform tasks that seem to require human intelligence. Machine learning, or ML, is one way to achieve AI by letting systems learn patterns from data instead of following only fixed rules.
Deep learning is a more specialized form of machine learning that uses layered neural networks to model complex patterns.
The relationship is often shown as nested sets: deep learning inside machine learning, and machine learning inside artificial intelligence. Traditional AI can include rule based systems, search algorithms, and logic, even when no learning happens. Machine learning usually depends on training data and improves performance by adjusting model parameters.
Deep learning is especially useful for images, speech, and language because many network layers can automatically learn useful features from raw input.
Understanding What Is AI?
A rule based program works by following instructions written by people. A simple spam filter might block a message when it contains certain words or comes from a known suspicious address. This can be clear and reliable when the situation is limited.
It becomes difficult when there are too many possible cases. A programmer cannot realistically write a separate rule for every handwriting style, road scene, accent, or medical image.
Machine learning changes the job. Instead of listing every clue, people provide examples and a learning method adjusts internal settings until its predictions become less wrong.
Training is not the same as using a finished model. During training, a model sees many examples, compares its output with the known answer, and measures the error. It then changes its parameters by a small amount.
This process repeats many times. In supervised learning, photos of animals may carry labels such as cat or dog. The labels give the model feedback.
After training, testing uses new examples that were kept separate. A model that performs well only on its training examples may have memorised them.
This problem is called overfitting. Good evaluation checks whether the model can generalise to unfamiliar data.
Deep neural networks learn through layers of connected units. Early layers can detect simple parts of an input. For an image, these may include edges, colours, and small shapes.
Later layers combine those parts into more meaningful patterns, such as eyes, wheels, or faces. The final layer produces a prediction. This layered approach is useful because people do not need to hand code every feature.
It has limits. The model depends strongly on the examples it receives.
If a face recognition system is trained mostly on one group of people, it may work less accurately for others. Biased data can create unfair results even when the program appears technical and neutral.
Students meet these ideas in recommendation feeds, keyboard word suggestions, translation tools, voice assistants, fraud checks, and image search. Each application needs different choices about data, speed, accuracy, privacy, and human oversight. A recommendation system may be acceptable when it makes an occasional poor suggestion.
A system used in healthcare or driving needs much stronger testing because mistakes can harm people. When studying this topic, focus on the flow from input data to training to prediction to evaluation. Notice the difference between a pattern that is useful and a decision that should be trusted.
AI systems can process large amounts of information, but they do not understand the world in the same way humans do. People remain responsible for setting goals, checking outcomes, and deciding when a system should not be used.
Key Facts
- AI is the broad field of building systems that perform intelligent tasks such as reasoning, planning, perception, and decision making.
- ML is a subset of AI where a model learns from data instead of being programmed with every rule explicitly.
- Deep learning is a subset of ML that uses multilayer neural networks, often written as y = f(x; theta).
- A common ML goal is to minimize prediction error, for example Loss = (1/n) * sum((y - y_hat)^2).
- In supervised learning, the model maps inputs to labeled outputs, written as y_hat = f(x).
- More data and larger models often help deep learning, but they also increase computation time and energy use.
Vocabulary
- Artificial Intelligence
- The broad area of computer science focused on creating systems that can perform tasks associated with human intelligence.
- Machine Learning
- A branch of AI in which computers learn patterns from data to make predictions or decisions.
- Deep Learning
- A type of machine learning that uses neural networks with many layers to learn complex patterns.
- Training Data
- The examples a model uses to learn relationships between inputs and outputs.
- Neural Network
- A computational model made of connected units that transform input data through layers to produce an output.
Common Mistakes to Avoid
- Saying AI, ML, and deep learning all mean the same thing, which is wrong because they describe different levels of a hierarchy with different methods and scope.
- Assuming every AI system learns from data, which is wrong because some AI systems use fixed rules, logic, or search without machine learning.
- Thinking deep learning always performs best, which is wrong because it often needs large datasets, high computing power, and may be unnecessary for simple problems.
- Believing more training data automatically guarantees a fair and accurate model, which is wrong because biased or low quality data can produce biased or unreliable results.
Practice Questions
- 1 A school uses a rule based program that follows fixed if then statements to schedule classrooms. Is this AI, ML, deep learning, or more than one of these? Explain your choice.
- 2 A model is tested on 4 examples with true values y = [3, 5, 2, 6] and predictions y_hat = [2, 5, 4, 7]. Calculate the mean squared error using Loss = (1/n) * sum((y - y_hat)^2).
- 3 A company wants to identify objects in millions of photos. Explain why deep learning might be preferred over a simple rule based system, and give one tradeoff.