AI vs. Machine Learning: Understanding the Core Difference

Learning Objectives

• Define artificial intelligence (AI) and machine learning (ML) in plain language.
• Identify how ML fits inside the broader AI landscape.
• Compare goals, techniques, and applications of AI vs. ML.
• Walk through hands-on Python examples that train a simple ML model.
• Recognize real-world scenarios where AI doesn’t require ML—and vice-versa.
• Understand how a modern SQL editor like Galaxy can support AI/ML-driven analytics.

1. Foundations: What Is Artificial Intelligence?

Artificial intelligence is a broad computer-science discipline focused on building systems that perform tasks normally requiring human intelligence—think reasoning, perception, planning, or language understanding. The term dates back to the 1956 Dartmouth Conference, long before today’s deep-learning boom.

Classic AI research produced:

• Expert Systems (e.g., MYCIN medical diagnosis in the 1970s)
• Rule-Based Chatbots (ELIZA, 1966)
• Search & Planning Algorithms (A* search, game-tree minimax)
• Constraint Solvers for scheduling and logistics

Notice these approaches rely on human-coded rules rather than data-driven learning.

1.1 Key Characteristics of AI

• Goal-oriented (e.g., winning chess, diagnosing illness)
• May rely on symbolic logic or rules
• Does not always require statistical learning
• Focuses on intelligent behavior, not the method itself

2. Foundations: What Is Machine Learning?

Machine learning is a subset of AI that automatically discovers patterns in data and uses them to make predictions or decisions, without explicit rules programmed by humans. Arthur Samuel famously defined ML in 1959 as the field that gives “computers the ability to learn without being explicitly programmed.”

2.1 Types of Machine Learning

• Supervised Learning – Learn from labeled examples (e.g., spam vs. ham emails).
• Unsupervised Learning – Find hidden structure in unlabeled data (e.g., customer clustering).
• Reinforcement Learning – Learn via feedback signals after taking actions (e.g., AlphaGo).

2.2 ML Workflow at a Glance

1. Collect & clean data (often stored in SQL databases).
2. Split into training & test sets.
3. Choose an algorithm (e.g., decision tree).
4. Train the model.
5. Evaluate performance.
6. Deploy & monitor.

3. AI vs. ML: Side-by-Side Comparison

AspectArtificial IntelligenceMachine LearningScopeUmbrella discipline for intelligent systemsNarrow field focused on data-driven learningCore Question“Can we make machines act intelligently?”“Can machines learn patterns from data?”Primary InputsRules, logic, knowledge graphsDatasets (features & labels)Example TechniquesLogic programming, search, constraint solvingNeural networks, regression, clusteringOutputsPlans, actions, reasoning chainsPredictions, classifications, embeddings

4. Hands-On Example: Spam Detection in 30 Lines of Python

Below we train a basic Naïve Bayes classifier—illustrating ML in action.

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import pandas as pd

# 1. Load a tiny dataset
sms = pd.read_csv("https://raw.githubusercontent.com/justmarkham/pycon-2016-tutorial/master/data/sms.tsv", sep='\t', names=['label', 'text'])

# 2. Split
X_train, X_test, y_train, y_test = train_test_split(sms['text'], sms['label'], test_size=0.2, random_state=42)

# 3. Build pipeline: vectorizer + classifier
model = make_pipeline(CountVectorizer(), MultinomialNB())

# 4. Train
model.fit(X_train, y_train)

# 5. Evaluate
preds = model.predict(X_test)
print("Accuracy:", accuracy_score(y_test, preds))

What just happened?

1. No explicit IF/ELSE spam rules were written.
2. The model learned which words correlate with spam.

This is quintessential machine learning, and therefore a subset of AI.

5. When AI ≠ ML

Consider a logistics planner that uses A* search to find the shortest route. It’s intelligent in the sense of planning, yet relies entirely on graph search—not on learning from data. Similarly, a deterministic rules engine for loan approvals might be called “AI” in marketing copy, but contains zero ML.

Real-World Non-ML AI Examples

• Game-tree solvers in classical board games
• Deterministic chatbot scripts (IVR phone trees)
• Constraint-based employee scheduling systems

6. When ML Doesn’t Feel Like AI

On the flip side, data scientists often view a simple linear regression as just statistics. Yet by definition it’s a supervised learning algorithm and thus part of ML—and, by extension, AI.

7. Galaxy in the AI/ML Workflow

While Galaxy is primarily a SQL editor, it bridges the data pipeline crucial for ML:

• Data Extraction: Write complex SQL to assemble training datasets. Galaxy’s AI Copilot helps generate performant queries fast.
• Version & Reuse: Endorse and share the SQL that sources your model inputs, ensuring reproducibility.
• Governed Access: Collaborate with ML engineers and analysts without emailing CSVs—run queries directly in Galaxy.
• Exploratory Analysis: Use result previews and (upcoming) visualizations to inspect feature distributions before exporting to Python notebooks for training.

In short, Galaxy sits upstream of ML, guaranteeing trustworthy data flows into your models.

8. Common Misconceptions & How to Avoid Them

“If it uses ML, it must be AI—and vice-versa.”

False. ML is a technique inside the AI toolbox. Not all AI uses ML, and not all ML systems exhibit broader intelligence.

“ML replaces the need for subject-matter expertise.”

Wrong. Feature engineering, data interpretation, and ethical considerations demand human insight.

“Bigger models automatically mean better AI.”

Larger models can overfit or be inefficient. Quality data and clear objectives matter more.

9. Practice Challenges

1. Classification: Using a public Titanic dataset, write SQL in Galaxy to select passengers’ features, export to CSV, and train a logistic regression in Python. Aim for >78% accuracy.
2. Rule vs. ML: Implement a rule-based spam filter (e.g., block any email with “FREE” and “WIN”). Compare its false-positive rate to the ML model above.
3. Explain in 60 Seconds: Record yourself explaining AI vs. ML to a non-technical friend. If they can summarize back accurately, you nailed it.

Key Takeaways

• AI is the broad quest for machine intelligence; ML is one way to achieve it through data-driven learning.
• Not all AI uses ML (rule systems), and some ML feels like plain statistics.
• Understanding the distinction helps you choose the right technique for a problem.
• Tools like Galaxy ensure the data feeding your ML models is reliable, versioned, and shareable.

Next Steps

• Dive deeper into supervised vs. unsupervised learning algorithms.
• Explore ethics in AI—bias, fairness, and transparency.
• Set up Galaxy, connect your database, and start versioning the SQL pipelines that fuel your ML projects.