Artificial Intelligence

ELI5 (Explain Like I’m 5)

Imagine you add a shirt to your Amazon cart but don’t buy it. Later that day, you see an ad for that exact shirt on YouTube or your TV. That’s not a coincidence; it’s AI in action.

AI systems noticed your action and compared it with patterns from millions of other shoppers to predict you were still interested. 

But it goes further: AI doesn't just show you the shirt you liked; it guesses you might also like a specific bag or pair of shoes, even if you never looked at them, simply because it knows what people like you usually buy.

Next, think about Google Maps changing your route because it expects traffic ahead, or a shopping app rearranging products based on what you’ve looked at before.

In all these cases, no one is manually watching or deciding for you. 

Artificial Intelligence learns patterns from large amounts of data and uses them to make predictions about what is likely to happen next.

How AI Works

At a high level, Artificial Intelligence works by learning patterns from data and using those patterns to make decisions or predictions.

Instead of being programmed with fixed (if…else) rules for every situation, an AI system is trained on many examples. These examples could be past transactions, images, sensor readings, text, or user behavior. From this data, the system learns what usually happens and what doesn’t.

Once trained, AI systems follow a simple loop:

1. Take input.
2. Apply learned patterns to that input.
3. Produce an output (a prediction, classification, decision, or recommendation).
4. Use a feedback mechanism to check if the output was correct.
5. Improve the next output based on the feedback.

Most AI systems do not understand context or intent the way humans do. They work by measuring similarity, probability, and patterns based on past data. If the data changes or is incomplete, their outputs can vary and also be entirely wrong (termed a hallucination).

Classes of AI Systems

AI systems are best understood by the kind of output they produce, rather than by how they are built internally. Using this lens, most real-world AI applications fall into the below broad classes.

1. Predictive & Decision AI

These systems analyze historical and real-time data to predict outcomes or make decisions.
They are part of everyday experiences that people don’t consciously label as AI, such as:

• Traffic and route prediction
• Ad targeting and ranking
• Fraud and risk detection, spam filtering
• Recommendations (products, videos)
• Demand forecasting 

The output is typically a score, label, ranking, or decision, not new content.

2. Generative AI

Generative AI systems create new content rather than just analyzing existing data.

They are used for:

• Text generation (chat assistants, summaries)
• Image generation
• Video generation
• Music and audio generation
• Code generation

These systems learn patterns from large datasets and generate outputs that did not previously exist. Tools like ChatGPT, Sora, and Dall-E fall into this category. 

3. Perception AI

Perception AI focuses on interpreting the real world, especially unstructured data.

Common applications include:

• Image and object recognition
• Speech recognition
• Face and fingerprint recognition (on your mobile phones too!)
• Handwriting recognition

The goal is to convert raw inputs like images, audio, or video into structured data.

4. Control & Optimization AI

These systems take actions or optimize outcomes in dynamic environments.

Examples include:

• Robotics
• Autonomous vehicles
• Energy and traffic control systems
• Supply chain optimization

Here, AI doesn’t just predict; it actively influences real-world behavior.

5. Discovery & Insight AI

Discovery-focused AI systems are used to identify patterns or insights humans may not explicitly look for.

They are common in:

• Scientific research and drug discovery
• Anomaly detection
• Customer segmentation
• Pattern mining and simulations

The output is often an insight, hypothesis, or previously unseen relationships.

Most modern AI systems fall into one or more of the above classes.

Generative AI is highly visible today, thanks to the rise of tools like ChatGPT and Gemini. But it represents one part of a much broader AI that has been quietly shaping products and decisions for years.