Understanding Generative AI

What Is Generative AI?

Generative AI is a category of artificial intelligence that learns patterns from training data and uses those patterns to generate new content.

For example, a generative AI system can write an email from a short instruction, create an image from a text description, or generate Python code from a programming requirement.

Traditional predictive AI might estimate whether a transaction is fraudulent. Generative AI might draft an explanation of why a transaction was flagged.

Types of Generative AI

• Text generation: Creates articles, emails, summaries, and conversational responses.
• Image generation: Produces images from text prompts or other images.
• Audio generation: Creates speech, music, or sound effects.
• Video generation: Produces or modifies video content.
• Code generation: Generates, explains, or modifies programming code.

How Does Generative AI Work?

Generative models learn statistical patterns and relationships from training data. They use those learned patterns to produce new outputs based on an input or prompt.

Different architectures support different tasks:

• Transformers: Widely used in modern language models.
• Diffusion models: Commonly used for image generation and other generative tasks.
• Generative Adversarial Networks (GANs): Use competing generator and discriminator networks to learn data distributions.
• Variational Autoencoders (VAEs): Learn compressed latent representations that can be used to generate data.

Generative AI does not simply retrieve a perfect answer from its training data. Its outputs can be inaccurate, incomplete, biased, or inconsistent.