✦ScienexySCIENCE KNOWLEDGE LIBRARY
Science reference · SCI-CS-003-179

Artificial General Intelligence (AGI)

Artificial General Intelligence (AGI) is a proposed form of artificial intelligence with broad capabilities to learn, reason, and perform many different intellectual tasks.

Explanation

Today's AI systems can perform impressive tasks such as language generation, image recognition, coding assistance, translation, and scientific analysis. However, success across several tasks does not by itself establish general intelligence. Researchers debate what capabilities would be necessary for AGI, including flexible reasoning, long-term planning, robust learning, transfer of knowledge, adaptation to new environments, and the ability to operate reliably under unfamiliar conditions. Another challenge is evaluation: a system may perform well on a benchmark while still failing in situations that differ from its training experience. Discussions about AGI also involve questions about economic effects, governance, safety, social consequences, and the distribution of benefits. Because definitions and evaluation criteria remain disputed, claims that a particular system has achieved AGI should be treated carefully.

Example or application

A specialized chess program can defeat expert players but cannot automatically perform unrelated tasks such as conducting scientific research or managing a household. A hypothetical AGI system would be expected to adapt across a much wider range of tasks.

Text

Artificial General Intelligence refers to a hypothetical or debated level of AI capability in which a system can perform a wide range of cognitive tasks rather than being limited to a narrow set of specialized activities. Unlike narrow AI, which is designed for particular tasks, AGI would be expected to transfer knowledge between different domains, adapt to unfamiliar problems, and learn new skills with substantial flexibility. There is no universally accepted scientific definition or agreed-upon test for AGI.

Key Points

  • AGI refers to broad and flexible intellectual capabilities.

  • Narrow AI focuses on specific tasks or limited domains.

  • Knowledge transfer and adaptation are important proposed AGI capabilities.

  • There is no universally accepted definition or test for AGI.

  • Strong performance on multiple benchmarks does not prove general intelligence.

  • AGI raises technical, ethical, economic, and governance questions.

Applications

  • Research into general-purpose AI systems

  • Evaluation of broad reasoning and learning capabilities

  • Development of adaptable intelligent agents

  • Research into AI safety and governance

  • Study of the long-term social effects of advanced AI

Learning Outcomes

  • Define Artificial General Intelligence.

  • Distinguish AGI from narrow AI.

  • Identify proposed characteristics of general intelligence.

  • Explain why evaluating AGI is difficult.

Practice Questions

  • What is Artificial General Intelligence?

  • How does AGI differ from narrow AI?

  • Why is there no simple universal AGI test?

  • Why does success on many benchmarks not necessarily prove general intelligence?

Related Concepts

  • Narrow AI

  • Artificial Superintelligence

  • Generalization

  • Transfer Learning

  • AI Safety

Common Mistakes

  • Assuming that a powerful chatbot is automatically AGI.

  • Treating AGI as a precisely defined and universally agreed scientific category.

  • Confusing performance across several tasks with reliable general intelligence.

  • Assuming that AGI has already been conclusively achieved.

Sources and verification

Sources are shown as provided in the project data. Not every source has yet been independently verified. Please report errors or missing citations.

  • Legg, S. and Hutter, M., A Collection of Definitions of Intelligence, 2007.
  • Chollet, F., On the Measure of Intelligence, 2019.
  • Bubeck et al., Sparks of Artificial General Intelligence: Early Experiments with GPT-4, 2023.

Suggest a correction or source · More in Artificial Intelligence & Machine Learning