Practical No. 3

Generative Artificial Intelligence (Generative AI)

Introduction

Generative Artificial Intelligence (AI) refers to a class of advanced machine learning systems that create new, original content—such as text, images, audio, or structured data—based on patterns learned from vast datasets. Unlike traditional AI that classifies or predicts based on input, generative AI can produce new artifacts that resemble human-created content. Its emergence has revolutionized creative industries, scientific research, and everyday technologies by automating content creation and enabling novel applications that were once the domain of human creativity alone.

Core Principles of Generative AI

At the heart of generative AI are learning models that learn the distribution of training data to mimic and produce similar outputs. These models are built on neural networks—complex mathematical systems inspired by the human brain.

How Models Learn and Generate

During training, a generative model ingests massive datasets (such as millions of text sentences or images) and identifies deep statistical patterns within the examples. Once trained, the model uses these learned patterns to produce novel outputs that are plausible but not direct replicas of training data.

Latent Representations and Output

Most generative models operate through a latent space—an abstract representation of input data compressed into numerical vectors. The model maps this latent space back into the output domain (e.g., text or images).

Types of Generative AI Models

Different architectures power various generative AI systems. Each has unique mechanisms and typical applications.

Figures 1-4. Examples illustrating text-to-image workflows, generative model structures, and architectural ideas behind latent representations and generative processes.

1. Generative Adversarial Networks (GANs)

GANs consist of two neural networks—the generator and the discriminator—training in competition. The generator tries to create realistic output, while the discriminator learns to differentiate real from generated examples. Through this adversarial process, GANs produce highly realistic outputs, especially images.

2. Variational Autoencoders (VAEs)

VAEs compress data into a latent space and reconstruct it. They create smooth and varied generative outputs useful for realistic but diverse content.

3. Diffusion Models

These models add noise to data and learn to reverse it to generate new, high-quality images. Models like DALL-E, Imagen, and Stable Diffusion use this technique to turn text prompts into visual content.

4. Transformers and Large Language Models (LLMs)

Architecture such as GPT (Generative Pre-trained Transformer) models excels at text generation, enabling human-like writing by predicting sequences of words.

Illustrative Examples of Generative AI

Generative AI spans numerous real-world systems that produce diverse types of content:

Text Generation

ChatGPT is a prominent example of a language model that creates coherent essays, dialogues, and responses given user prompts.

Image Synthesis

Stable Diffusion takes written descriptions to generate visual art, landscapes, and creative images.

Art and Culture

The painting Edmond de Belamy—created via a GAN—sold at auction, illustrating how generative AI intersects with fine art.

Other Creative Outputs

Other specialized generative AI can compose music, design optimized 3D models, or generate synthetic data for training purposes.

Applications Across Domains

Generative AI has widespread influence:

Healthcare

It supports personalized treatment planning, synthetic medical imaging data, and advanced diagnostics assistance.

Entertainment and Media

Automated script writing, music generation, and digital art creation are some applications transforming creative workflows.

Business and Design

From generative design in engineering to automated marketing content, businesses use generative AI to reduce creative time and cost.

Data Augmentation and Simulation

Generative models can create training data in scenarios where real data is limited, aiding robust machine learning systems.

Challenges and Ethical Considerations

Despite its potential, generative AI raises concerns about copyright, bias, and misuse.

Copyright and Data Rights

Models trained on internet data may reproduce patterns from copyrighted works, leading to disputes about ownership of outputs.

Security Risks

“Data poisoning”—maliciously corrupting training data—can compromise models and output integrity.

Therefore, ethical AI frameworks and legal guidelines are becoming important aspects of responsible generative AI deployment.

Conclusion

Generative AI represents a major milestone in artificial intelligence, enabling machines to create rather than just analyze. These systems harness advanced architectures like GANs, diffusion models, and transformers to produce content across modalities—text, images, audio, and beyond—transforming industries from art and design to healthcare and business innovation. As this technology evolves, balancing creativity with ethical responsibility remains essential.

References

OpenAI. (2026). Generative models. https://openai.com/index/generative-models/

Simform. (2021). What is generative AI: Examples, use cases, and models. https://www.simform.com/blog/what-is-generative-ai/

LeewayHertz. (2023). Generative AI models: What it is, types, workflow and applications. https://www.leewayhertz.com/generative-ai-models/

AltexSoft. (2023). Generative AI models explained. https://www.altexsoft.com/blog/generative-ai/

Wikipedia contributors. (2026). Stable Diffusion. https://en.wikipedia.org/wiki/Stable_Diffusion

Wikipedia contributors. (2024). Edmond de Belamy. https://en.wikipedia.org/wiki/Edmond_de_Belamy

NVIDIA. (2026). Generative AI overview. https://www.nvidia.com/en-eu/glossary/generative-ai/

Blockchain Council. (2025). Generative AI examples. https://www.blockchain-council.org/ai/generative-ai-examples/