Classification of AI Applications

As we embark on our journey through Edge AI Engineering with the Arduino UNO Q, it’s essential to understand the fundamental classification of AI applications that structure this book. Our exploration is divided into two parts, each representing a different paradigm for implementing artificial intelligence.

Fixed Function AI vs. Generative AI

AI applications can be broadly categorized into two approaches that represent different capabilities, interaction models, and implementation strategies:

Fixed Function AI (Reactive)

Fixed Function AI, or Reactive AI, operates by analyzing specific inputs according to predetermined patterns and rules and then producing consistent outputs for given scenarios. These systems:

  • Respond to specific triggers: They activate only when presented with particular inputs.
  • Follow defined patterns: Their behavior is predictable and consistent.
  • Excel at structured tasks: They perform exceptionally well at classification, detection, and pattern recognition
  • Operate within boundaries: Their capabilities are limited to their specific programming.

In the second part of this book, we review fixed-function AI through computer vision applications:

  • Image classification for identifying objects in images
  • Object detection for locating and labeling multiple objects

These applications demonstrate how edge devices can deliver reliable, efficient AI in constrained environments, focusing on specific, well-defined tasks. On the UNO Q, they also introduce a toolchain we do not use elsewhere: App Lab, Arduino’s application framework, and the arduino-app-cli command line, which package a trained model, a web interface, and an MCU sketch into a single deployable application.

Generative AI (Proactive)

Generative AI, also known as Proactive AI, is the core of this book and represents a fundamental shift in capability. These systems can:

  • Create new content: They generate novel text, images, or solutions.
  • Understand context: They interpret and respond to nuanced situations.
  • Engage in dialogue: They maintain contextual awareness across interactions.
  • Adapt to novel scenarios: They apply knowledge to situations beyond their explicit training.

The first part of this book explores Generative AI at the edge:

  • Small Language Models that bring conversational AI to edge devices
  • Vision-Language Models that combine visual and textual understanding
  • Physical computing integration that connects AI to the real world
  • Agentic AI, where the model is given a set of tools it can decide to call on its own

This progression from Fixed Function to Generative AI mirrors the evolution of artificial intelligence itself—from specialized systems designed for specific tasks to more flexible, creative systems capable of addressing a broader range of challenges.

Summary Table

AI Type Core Focus Example Applications Typical Use Cases
Fixed Function (Reactive) Data analysis, assessment, automation Fraud detection, spam filters, image recognition Banking, healthcare diagnostics, security systems
Generative (Proactive) Content creation, anticipation, dialogue ChatGPT, DALL·E, predictive maintenance, smart assistants Content creation, customer support, design, automation

Conclusion

  • Fixed Function (Reactive) AI is ideal for applications requiring efficiency, predictability, and low resource use, where tasks are well-defined and do not require creative output or adaptation.
  • Generative (Proactive) AI is suited for scenarios demanding creativity, personalization, and anticipatory actions. It enables richer, more human-like interactions and innovative solutions across industries.

The Edge AI Advantage

Both Fixed Function and Generative AI gain unique benefits when deployed at the edge:

  • Reduced latency: Processing happens locally, eliminating network delays
  • Enhanced privacy: Sensitive data remains on-device
  • Operational reliability: Systems function regardless of network connectivity
  • Resource efficiency: Optimized models utilize limited hardware effectively

By understanding these fundamental classifications, you’ll realize how different AI approaches serve distinct purposes and how each can be effectively implemented on edge devices like the UNO Q.

As we progress through this book, this classification framework will help contextualize each project and technique, connecting individual implementations to broader AI concepts and applications.

How This Book Fits the Sequence

This book is the generative-AI volume of a two-book set. Its companion, EdgeML Made Easy, covers fixed-function AI on the Raspberry Pi — image classification, object detection, and object counting on the Pi Zero 2W, followed by a second part on the Pi 5.

In IESTI05 the two are used in sequence, and that sequence runs opposite to the internal order of this book:

  1. Fixed Function first, on the Raspberry Pi Zero 2W, from the companion book. Starting here is deliberate: reactive models are easier to reason about, the training loop is visible end to end, and the hardware is inexpensive enough for everyone to have one.
  2. Generative AI second, starting on the students’ own laptops. This book opens with SLMs at the Edge, which runs a small language model on a PC — so the concepts of quantization, GGUF, and memory budgeting land before any board is involved, and nobody is blocked waiting on hardware. It then moves to the Arduino UNO Q for the rest of Part 1. By this point the vocabulary of inference, quantization, and latency is already familiar, so the discussion can move directly to what changes when the model generates rather than classifies.
  3. Part 2 of this book then revisits classification and detection on the UNO Q. The tasks are already familiar; what is new is the App Lab / arduino-app-cli deployment path and driving MCU actuation from inference results over Bridge.
  4. The Raspberry Pi 5 material from the companion book remains available as optional reading and as a platform for group projects.

If you are reading this book on its own rather than as part of the course, none of this matters — every chapter here is self-contained, and Part 2 introduces the computer vision concepts from the beginning.

A chapter on Retrieval-Augmented Generation (RAG) at the edge is planned for a future revision. Until then, the Advancing Edge AI chapter of the companion book covers the topic on the Raspberry Pi, and the approach transfers to the UNO Q with the llama-server setup from Part 1.