What Is Artificial Intelligence? Unpacking Narrow AI, General AI, ML, and Deep Learning

Everywhere you look today, software is described as “AI-powered.” From recommendation engines that pick your next movie and fraud filters that protect your credit card, to large language models that write code and summarize research papers, Artificial Intelligence (AI) has moved from academic research labs into the core of modern software engineering.

Yet in everyday conversations, terms like Artificial Intelligence, Machine Learning (ML), Deep Learning (DL), and Generative AI are often used interchangeably. In this opening chapter of the AI Fundamentals course series on aimlrevi/sited.com, we unpack what intelligence means in machines, distinguish Narrow AI from General AI, map out the concentric hierarchy of AI subfields, and walk through a hands-on Python comparison.

1. What Does “Intelligence” Mean in a Machine?

At its core, Artificial Intelligence is the branch of computer science dedicated to building systems capable of performing tasks that typically require human intelligence—such as perceiving patterns, understanding language, reasoning under uncertainty, planning ahead, and learning from experience.

Computer scientists Stuart Russell and Peter Norvig organize definitions of AI along two dimensions—human-centered vs. rational, and thinking vs. acting:

The Turing Test vs. Rational Agents

In 1950, Alan Turing proposed the Imitation Game (now known as the Turing Test) in his landmark paper Computing Machinery and Intelligence. Rather than debating the philosophical question “Can machines think?”, Turing proposed an operational test: if a human interrogator conversing via text cannot reliably tell whether they are talking to a human or a machine, the machine demonstrates intelligent behavior.

While conversational fluency is a major milestone, modern AI engineering evaluates systems by rational utility—how accurately, safely, and efficiently a system perceives its environment and takes actions to achieve a well-defined objective.

2. The Capability Spectrum: Narrow AI (ANI), General AI (AGI), and Superintelligence (ASI)

1. Artificial Narrow Intelligence (ANI / Weak AI): AI designed and trained to perform a single task or a tightly bounded set of tasks. Every AI system deployed in the world today is Narrow AI—whether it is AlphaFold predicting 3D protein structures, an autonomous driving perception system, a credit-scoring model, or a conversational LLM.
2. Artificial General Intelligence (AGI / Strong AI): A hypothetical system possessing broad, adaptable cognitive abilities across virtually any intellectual domain, capable of transferring knowledge seamlessly across unfamiliar tasks.
3. Artificial Superintelligence (ASI): A theoretical stage in which machine intelligence surpasses the best human minds across every field, including scientific discovery and creative problem-solving.

3. The Concentric Hierarchy: AI vs. Machine Learning vs. Deep Learning vs. Generative AI

You can visualize the relationship between AI’s major subfields as nested concentric circles, where each inner layer is a specialized subset of the outer layer:

• Artificial Intelligence (The Outer Circle): Encompasses all approaches to machine intelligence, including non-learning classical methods (such as tree search like A* pathfinding, Minimax in chess engines, and rule-based expert systems) as well as statistical learning methods.
• Machine Learning (The Second Circle): Instead of a human programmer writing explicit if-else rules for every edge case, Machine Learning algorithms examine historical data and automatically fit mathematical parameters to map inputs (X) to predictions (y). Classical ML includes Linear Regression, Logistic Regression, Decision Trees, Random Forests, Support Vector Machines (SVMs), and K-Means Clustering.
• Deep Learning (The Third Circle): A specialized subset of Machine Learning inspired by biological neurons. Deep Learning passes data through multi-layered Deep Neural Networks to automatically learn hierarchical features—from raw pixels to edges, shapes, and whole objects.
• Generative AI (The Inner Circle): While discriminative deep learning models classify or predict a label (e.g., “Is this email spam or not?”), Generative AI models learn the underlying probability distribution of a dataset to generate brand-new text, source code, images, and audio using architectures like Transformers and Diffusion Models.

4. A Brief History of AI: From Dartmouth to Foundation Models

• 1950–1956 (The Birth of AI): Alan Turing publishes Computing Machinery and Intelligence (1950), and John McCarthy coins the term “Artificial Intelligence” at the 1956 Dartmouth Workshop.
• 1957–1974 (Early Optimism & First AI Winter): Frank Rosenblatt builds the Perceptron (1957); Minsky and Papert highlight linear separability limits (XOR) in 1969, triggering the first AI Winter.
• 1980s (Expert Systems & Backpropagation): Enterprise adoption of rule-based Expert Systems grows, while Hinton, Rumelhart, and Williams popularize Backpropagation (1986) for multi-layer neural networks.
• 1990s–2000s (Statistical ML Triumphs): Support Vector Machines, Random Forests, and Gradient Boosting become industry standards, and IBM’s Deep Blue defeats world chess champion Garry Kasparov (1997).
• 2012–Present (Deep Learning & Transformers): Powered by GPUs, AlexNet wins the 2012 ImageNet competition, and the 2017 Transformer architecture (“Attention Is All You Need”) launches the era of Large Language Models.

import numpy as np
from sklearn.tree import DecisionTreeClassifier

# 1. Classical Rule-Based Approach: Human writes explicit thresholds
def rule_based_alert(cpu_pct: float, mem_pct: float) -> int:
    if cpu_pct > 85.0 and mem_pct > 75.0:
        return 1  # Urgent
    return 0      # Routine

# 2. Machine Learning Approach: Algorithm learns thresholds from historical logs
# Historical data: [cpu_pct, mem_pct]
X_train = np.array([
    [30.0, 40.0],
    [55.0, 50.0],
    [88.0, 82.0],
    [92.0, 79.0],
    [45.0, 95.0],  # High memory leak even at moderate CPU!
    [48.0, 93.0],
])
# Historical outcomes: 0 = Routine, 1 = Urgent incident
y_train = np.array([0, 0, 1, 1, 1, 1])

model = DecisionTreeClassifier(max_depth=2, random_state=42)
model.fit(X_train, y_train)

# Test on a new server state: 46% CPU, 94% Memory
new_server = [[46.0, 94.0]]
print("Rule-Based Prediction:", rule_based_alert(46.0, 94.0))  # Outputs 0 (Missed!)
print("ML Model Prediction:  ", model.predict(new_server)[0])  # Outputs 1 (Caught!)