Fundamentals of AI

In this blog we will see about

1. Machine Learning

2. Deep learning

3. Generative AI

AI is the broadest term. It makes machines behave like humans (thinking, decision making). This includes everything from a simple ‘if-then’ program to complex systems that can drive cars, Chatbots, Recommendation systems.

Machine Learning

This is a subset of AI. Instead of being programmed with specific rules, machine uses algorithms to find patterns in data and make decisions. It learns from experience.

Example: A spam filter. You don't program it with every ‘spam’ word. Instead, you show it thousands of spam emails, and it learns to recognize the patterns of a scam on its own.

How ML Works?

Uses data which can be labeled or unlabeled, structured or unstructured. Then train the data with ML algorithms. Once the model is trained it will be used in real time.

1. Data collection

• Labeled: In image classification, each image labeled like, if the image is vehicle(car), animal (cat).

• Unlabeled: A collection of images without labels.

• Structured: Formatted data in the form of rows and columns, databases.

• Unstructured: Text, images, audio and video.

2. Training the Model

• Data is fed into an ML algorithm.

• The algorithm learns patterns from the.

• Example words like ‘win money’, ‘free offer’ is likely spam. Normal conversation means not spam.

3. Model Creation

• After training, a model is created.

• This model represents learned patterns.

• The model now understands patterns like: Suspicious words, Email structure, Frequency of certain phrases.

4. Prediction

• The model is used on new, unseen data.

• It predicts outputs based on what it learned.

• Example: A new email arrives: ‘Congratulations! You won a money’. Model predicts it as a Spam.

Deep Learning:

A subset of Machine Learning that uses Neural Networks. It requires large amount of data and high computing power to solve very complex problems, like recognizing a face or translating a language.

Example: A photo app that can identify ‘cat’ in your pictures. It looks at millions of pixels to understand it is a cat.

How Deep Learning Works?

Example: an image of a cat

• The image is just numbers (pixels).

• These numbers go into the network.

• Neural Network Layers process it. First Layer detects basic things like edges, lines Middle Layers detects shapes like ears, eyes. Final Layer Combines everything like ‘This is a CAT’.

• Prediction is made, Output: ‘Cat’ (or probability like 95% cat).

• Error is calculated:  Actual image is dog. Wrong prediction.

• Learning: The network adjusts itself. Changes internal weights and tries to reduce error next time.

Generative AI (GenAI)

A subset of Deep Learning. Traditional AI is about classifying or predicting (e.g., Is this a cat?), Generative AI is about creating something new (e.g., Draw a cat wearing a tuxedo).

Example: A chatbot. It does not just find a pre written answer. It generates a new response based on the patterns it learned during training.

How GenAI Works?

Generative AI uses already trained big models (foundation models). It works on models from deep learning so we don’t need to build a model from scratch.

Traditional ML: Collect data for each task, then we need to label the data and train a separate model. Example, One model for spam detection, another for translation and another for chatbot.

Generative AI and FMs: Use one big pretrained model & just give instructions (prompt). So, the model can write text, summarize, answers questions and generate images.