Artificial intelligence

Definition

A set of reusable methods and system capabilities for learning, reasoning, perception, and goal-directed action through computational systems.

Scope

This entry treats artificial intelligence as an umbrella category of reusable computational methods and system capabilities, covering symbolic reasoning, statistical learning, neural networks, and agents. The 1955 Dartmouth proposal and the 1956 project are historical reference points for the name and the organization of research; earlier work, such as neuron models, predates them.

Multiple approaches

Different methods can coexist and be combined. Machine learning is an important approach but does not encompass all of artificial intelligence. This entry does not equate performance on a specific task with general intelligence.

Principles and impact

Problems solved

Enabled computational systems to undertake cognitive and decision-making tasks that are difficult to specify through fixed rules for every case.

Principles

Uses computable representations, inference, learning, and action processes to work with goals and environments.

Improvements

Not recorded

Significance

Provided a common research and engineering scope for multiple approaches to intelligent computing.

History and applications

People

John McCarthy、Marvin Minsky、Nathaniel Rochester、Claude Shannon

Organizations

Not recorded

Events

Not recorded

Applications

Reasoning、Learning、Perception、Decision support

Limitations

Capabilities depend on the task, data, and environment; a result on one benchmark cannot guarantee reliability across settings.

Upstream and downstream relations

Upstream

←Bayesian networksusesBayesian networks are used for probabilistic knowledge representation and inference under uncertainty in AI, and can also be used in statistical modeling more broadly.

Sources: Fusion, Propagation, and Structuring in Belief Networks — Commentary on a Classic Paper Hosted by the Original Author

←Stored-program computerenabled byProgrammable computers provide a general-purpose platform for implementing AI programs; the Dartmouth proposal included automatic computers among the requirements for the research. This relation does not claim that all neural or intelligent systems can only be implemented in this hardware form.

Sources: A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence

Downstream

→Language-model agentshas specialized formLanguage-model-driven agents are a class of goal-directed AI systems; the ReAct approach recorded here does not represent the entire history of agents.

Sources: ReAct: Synergizing Reasoning and Acting in Language Models What is Artificial Intelligence? — Basic Questions

→Machine learninghas specialized formWithin the engineering scope of AI covered by this family, machine learning improves performance through data and experience; AI also includes approaches that do not center on learning.

Sources: Some Studies in Machine Learning Using the Game of Checkers What is Artificial Intelligence? — Basic Questions

→Symbolic artificial intelligencehas specialized formSymbolic AI is a class of approaches within AI that solve problems through explicit representations, rules, and search.

Sources: Computer Science as Empirical Inquiry: Symbols and Search What is Artificial Intelligence? — Basic Questions

Definition

Scope

Multiple approaches

Principles and impact

Problems solved

Principles

Improvements

Significance

History and applications

People

Organizations

Events

Applications

Limitations

Upstream and downstream relations

Upstream

Downstream

Explore this technology's connections