Books

RedLectern topic

AI foundations

AI capabilities and limits · generative and predictive approaches · basic distinctions

For: Newcomers

After this page you can: Explain what AI does and distinguish major approaches

Page reviewed

Part of: Artificial Intelligence

About AI foundations

AI is a broad term for computer systems that perform tasks associated with intelligence, such as recognizing patterns, learning, reasoning, or making decisions. Some systems rely on explicitly written rules; others learn patterns from data. A model is one computational component, while the larger system may also include its inputs, interface, and intended task. The label “AI” alone does not say how a system works or how reliable its output is. OpenStax

A useful first distinction is the kind of output a system is meant to produce. Predictive approaches estimate likely outcomes, such as a category or value. Generative approaches produce content, such as text, images, music, or video. These are distinctions in task and output, not a claim that the approaches never overlap. AI systems can also use explicit symbols and rules, learn patterns from examples, or combine methods. To understand a system, ask what task it performs, what information it uses, and how its result can be checked. MIT Media Lab

Start here

Learn the foundations

1.What AI meansAI covers computer functions that perform tasks often associated with human intelligence, including visual perception and decision making.

AI covers computer functions that perform tasks often associated with human intelligence, including visual perception and decision making. Narrow AI describes systems designed for specific tasks. Artificial general intelligence (AGI) is a proposed broader capability, but there is no universally accepted test for it. A system handling multiple tasks does not, by that fact alone, settle what AGI means. OpenStax Stanford HAI

An AI model is a computational representation that relates inputs to outputs. A chatbot is a conversational agent, while data science is a field concerned with collecting, managing, analyzing, and reporting data. These words refer to different parts of the picture: a technical model, an interaction format, and a field of work. OpenStax OpenStax OpenStax

Artificial intelligence

Computer systems performing tasks associated with perception, reasoning, learning, or action.

Narrow AI

AI designed or evaluated for a limited range of tasks.

Artificial general intelligence

Proposed AI intended to learn, reason, and apply knowledge across a wide range of tasks; there is no universally accepted test for it.

Artificial superintelligence

Hypothetical AI that surpasses human cognitive capabilities across most domains.

Generative AI

AI that produces new content such as text, images, audio, or code.

AI model

A computational representation used to make predictions, generate outputs, or choose actions.

Chatbot

Software that interacts with people through conversational messages.

Data science

Using statistics and computation to analyze data and communicate findings.

Try it

Classify each example as a field, an interaction format, or a computational component: data science, a chatbot, and an AI model.

Data science is a field of work. A chatbot is a conversational interaction format. An AI model is a computational representation used to produce predictions, outputs, or actions.

What Is AiAi Vs Data ScienceWhat Is Agi

2.Predicting and generatingPredictive AI analyzes data to estimate future trends, behaviors, or events.

Predictive AI analyzes data to estimate future trends, behaviors, or events. A predictive system might return a likely category or a forecasted value. Generative AI produces content, such as text, images, music, or video, by learning patterns from existing data. The most useful distinction for a newcomer is the intended output: estimating an outcome versus creating content. MIT Media Lab

AI methods also differ in how they represent and solve problems. Symbolic reasoning uses explicit rules and logic; machine learning uses data-driven methods. Hybrid approaches bring these methods together. For example, labeling a message as spam is a prediction of a category, while drafting a new greeting produces content. MIT Media Lab UK Parliament POST

Try it

For each task, identify whether the main output is predictive or generative: estimate tomorrow's rain probability; draft a new greeting; label a message as spam.

Tomorrow's rain probability is predictive. Drafting a greeting is generative. Labeling a message as spam is predictive because the result is a category.

Generative Vs Predictive Ai

3.Capabilities depend on the taskJudge an AI system against the task it was built or evaluated to do.

Judge an AI system against the task it was built or evaluated to do. For a predictive model, evaluation compares predictions with known outcomes. A model that fits its training examples may still perform worse on new data, so performance on training examples alone is not enough to establish how well it generalizes. Ask what inputs and examples were used, what outcome was measured, and whether the evaluation resembles the intended use. OpenStax

Generated text can sound plausible and still be inaccurate. Fluency alone does not establish that a response's claims are correct. For important claims, look for relevant evidence and check whether the system had suitable information for the question. A system's performance on a narrow task also does not, by itself, establish broad capability across unrelated tasks. UK Parliament POST Stanford HAI

Try it

A model matches every example in its training data but makes poor predictions on a new set of examples. What does this show, and what should you check next?

Matching training data does not establish that a model generalizes. Check its performance on separate data and whether that data reflects the intended use.

Explore subtopics

Natural language processing

Methods for analyzing and processing language beyond conversational LLMs.

0 subtopicsSymbolic and neuro-symbolic AI

Rules; search; planning; expert systems; combinations with learned models.

0 subtopics

Latest

No recent updates are available.

See the sources list

Books

Glossary

8 terms on this page.

Browse the AI glossary

Questions and comments