What Is Artificial Intelligence? (From “Computerizing” Intelligence to Game-Playing Machines)

Artificial Intelligence (AI) is the field of building systems that perform tasks we typically associate with human intelligence—such as perception, reasoning, learning, and language understanding. Classic framing treats AI as building intelligent agents that “act” to achieve goals under uncertainty, often by learning from data.2

A useful way to organize “what is AI?” is around four complementary perspectives:
(i) engineering intelligence into computers,
(ii) programming intelligence,
(iii) [making machines intelligent]{def="Learning/optimizing so behavior improves over time}, and
(iv) learning through game play.

Below, each perspective is developed into a concrete learning pathway with examples, process steps, and self-check questions grounded in how AI is defined and studied across major references.4

1) Putting your intelligence into computer (knowledge representation + algorithms)

This view starts from the idea: if we can describe intelligent behavior as steps, then we can encode those steps and the knowledge they use into a computer.

Two foundational ideas show up repeatedly in AI history and definitions:

• knowledge representation: encoding facts/rules as data structures.
• search: exploring possible actions/paths when the correct move is not immediately obvious.

Early AI systems often relied on explicit rules and representations (for example, logical rules or hand-coded heuristics). Even today, many AI systems combine learned components with explicit reasoning or structured knowledge.2

Key learning targets in this perspective:

• Translate a human strategy into an explicit decision procedure.
• Identify the “state” (what you know), the “actions” (what you can do), and the “goal” (what counts as success).
• Use an algorithmic framework (often search + evaluation) to select actions.2

2) Programming with your own intelligence (rule-based / heuristic AI)

In this perspective, you (the developer) supply the intelligence—typically through:

• heuristics
• expert rules
• planning

This is common in systems where:

• the task is well understood,
• high reliability is required,
• and data for learning is limited.

However, the approach can be brittle: rules that work in one situation may fail in slightly different situations. This is one reason modern AI strongly emphasizes learning and adaptation.2

3) Making a machine intelligent (learning, adaptation, and optimization)

Here the goal is not only to encode steps, but to enable systems to improve performance with machine learning.

A standard relationship is:

• AI is the umbrella field for intelligent behavior,
• machine learning is a major method inside AI where models learn from data rather than relying exclusively on manually written rules.2

This perspective connects directly to how “intelligence” is operationalized:

• the system is rewarded/punished based on outcomes,
• it adjusts internal parameters to increase future success.

In controlled settings, this becomes: define a performance objective, then optimize using data and feedback.2

4) Playing a game (controlled intelligence: search + policy + evaluation)

Games provide an unusually clear environment for intelligence:

• rules are explicit,
• states are well-defined,
• and success is measurable (win/loss, score).

Game-playing AI often combines:

• adversarial search
• evaluation functions
• policies

Historically, game play was a proving ground because it requires selecting actions under uncertainty and maximizing long-term outcomes—not just immediate reactions.2