What is Artificial Intelligence (AI)?

Key Takeaways​

AI isn't magic; it's high-level computer science built to mimic human problem-solving. At its core, modern AI relies on using training datasets to generate models (smart code with statistical capabilities).

Instead of hardcoding every rule, data scientists train models to plot, categorize, and uncover patterns in massive datasets. When you feed new, unseen data into a trained model, it leverages those learned patterns to make predictions or decisions.

Main Discussion​

The Core Mechanics of Model Training & Inference​

Modern AI separates the system lifecycle into two distinct phases: Training (backend preparation) and Inference (user execution).

   +-----------------------------------------------------------+
   |                     TRAINING PHASE                        |
   |                                                           |
   |  +------------------+     +----------------------------+  |
   |  |  Training Data   | --> | Data Scientist / Algorithm |  |
   |  | (Labeled/Unseen) |     |     (Statistical Math)   |  |
   |  +------------------+     +----------------------------+  |
   |                                         |                 |
   |                                         v                 |
   |                               +-------------------+       |
   |                               |  Trained Model    |       |
   |                               +-------------------+       |
   +-----------------------------------------|-----------------+
                                             |
                                             v
   +-----------------------------------------------------------+
   |                     INFERENCE PHASE                       |
   |                                                           |
   |  +------------------+     +----------------------------+  |
   |  |  New User Data   | --> |       Trained Model        |  |
   |  | (Unseen Sample)  |     |   (Pattern Matching)       |  |
   |  +------------------+     +----------------------------+  |
   |                                         |                 |
   |                                         v                 |
   |                               +-------------------+       |
   |                               | Prediction Output |       |
   |                               +-------------------+       |
   +-----------------------------------------------------------+

1. Training Phase: A large dataset (e.g., millions of fruit images) is processed by statistical code algorithms. The algorithm learns boundary features across multiple dimensional axes (like mapping color, shape, or weight).
2. Inference Phase: The user provides brand new data. The model maps this input against its learned mathematical boundary to return a precise classification.

Mathematical Representation of Machine Intelligence​

AI taxonomy forms a nested hierarchy where each inner subfield leverages more specialized techniques:

$\text{AI} \supset \text{Machine Learning (ML)} \supset \text{Deep Learning (DL)} \supset \text{Generative AI (GenAI)}$

+-------------------------------------------------------------------+
| Artificial Intelligence (AI)                                      |
|  +-------------------------------------------------------------+  |
|  | Machine Learning (ML)                                       |  |
|  |  +-------------------------------------------------------+  |  |
|  |  | Deep Learning (DL)                                    |  |  |
|  |  |  +-------------------------------------------------+  |  |  |
|  |  |  | Generative AI (GenAI)                           |  |  |  |
|  |  |  |  (LLMs, Diffusion Models, Image Gen)            |  |  |  |
|  |  |  +-------------------------------------------------+  |  |  |
|  |  +-------------------------------------------------------+  |  |
|  +-------------------------------------------------------------+  |
+-------------------------------------------------------------------+

• Artificial Intelligence (AI): Overarching discipline focusing on machines performing tasks requiring human-like intelligence.
• Machine Learning (ML): Subset of AI using statistical algorithms to learn patterns from data without explicit programming.
• Deep Learning (DL): Subset of ML powered by multi-layered Artificial Neural Networks capable of processing complex, unstructured data.
• Generative AI (GenAI): Specialized subset of DL focused on creating new content (text, images, synthetic audio) using foundational models.

Chronological Milestones in AI Evolution​

• 1950s (Foundational Concepts): Alan Turing introduced the Turing Test (evaluating if a machine can conversationally pass as human). John McCarthy coined the term "Artificial Intelligence".
• 1970s (Rule-Based Systems): Development of Expert Systems (e.g., MYCIN for bacterial infection diagnosis) driven by strict conditional rules (IF-THEN logic).
• 1990s (Statistical ML Era): Algorithmic advances and computing power enabled data mining. In 1997, IBM's Deep Blue defeated world chess champion Garry Kasparov via mass move-computation.
• 2010s (Deep Learning Breakthroughs): Neural networks unlocked new capabilities. In 2016, Google DeepMind's AlphaGo defeated Go champion Lee Sedol using advanced Deep Learning, solving a game with possibilities far exceeding chess search trees.
• Present Day (Ubiquitous GenAI): Pervasive adoption across code assistance, automated reasoning, real-time voice synthesis, and multi-modal analysis.

Applied AI Architecture: Intelligent Document Processing (IDP)​

A key modern enterprise pattern is Intelligent Document Processing (IDP), which combines multiple AI sub-domains to convert unstructured files into structured database records:

[ Invoice PDF / Image ]
          |
          +---> Computer Vision (CV) -----> Locates text regions & layout
          |
          +---> Deep Learning (DL) --------> Extracts visual features
          |
          +---> Natural Language Processing (NLP) --> Parses entity meaning (Seller, Buyer, Total)
          |
          v
[ Structured JSON Data Output ] ---> [ Enterprise Database ]

Exam Guide​

Exam Tips​

• Distinguish the Taxonomy: Expect questions testing your ability to categorize a workload. GenAI creates new data, Deep Learning utilizes neural networks, Machine Learning relies on data-driven statistical training, and AI is the umbrella term.
• IDP Domain Overlap: Understand that enterprise solutions like IDP aren't a single model—they orchestrate Computer Vision (visual identification), NLP (context understanding), and Deep Learning (feature extraction).
• Training vs. Inference: Know the structural difference between generating a model using training data versus running predictions on new data during production inference.

Practice Test​

Question 1​

A financial company wants to automate its invoice processing pipeline. The system must intake PDF document scans, locate text fields, interpret line-item meanings, and convert the output into tabular data. Which combination of AI disciplines powers this solution?

• A. Deep Learning and Rule-Based Expert Systems only
• B. Computer Vision, Deep Learning, and Natural Language Processing
• C. Generative AI and Statistical Anomaly Detection only
• D. Reinforcement Learning and Computer Vision

Question 2​

An engineer needs to select the correct technology domain for a solution that generates novel synthetic images based on text prompts. Where does this capability fall within the AI hierarchy?

• A. Machine Learning, but outside of Deep Learning
• B. Artificial Intelligence, but outside of Machine Learning
• C. Generative AI, which is a subset of Deep Learning
• D. Rule-Based Expert Systems, which sit inside Deep Learning