Artificial general intelligence

Artificial general intelligence (AGI) is a hypothetical or developing AI system able to learn, reason, and perform a broad range of intellectual tasks across domains rather than being limited to specific tasks.

Artificial intelligence: The field and technology of creating machines that perform tasks associated with intelligence. AGI is a proposed broad capability within the wider field of artificial intelligence.
Narrow artificial intelligence: AI designed to perform a specific task or a limited set of related tasks. It is the task-bounded alternative to the broad competence sought in AGI.
Transfer learning: A machine-learning technique that applies knowledge gained on one task to another task. General intelligence requires useful knowledge to transfer beyond its original training tasks.
Alan Turing: A British mathematician and computer scientist who helped establish theoretical foundations of computation. His work framed questions about whether machines could exhibit intelligent behavior.
Human-level artificial intelligence: Artificial intelligence whose capabilities match people across a defined range of cognitive tasks. This common benchmark shapes debates about whether a system qualifies as general.
Artificial superintelligence: A hypothetical AI that substantially exceeds human cognitive abilities across nearly all domains. It describes a possible capability level beyond human-level general intelligence.
Reinforcement learning: A machine-learning approach in which an agent learns by taking actions and receiving rewards or penalties. It offers methods for learning behavior through interaction and feedback.
Dartmouth workshop: A 1956 summer research workshop that helped establish artificial intelligence as a field. Its proposal envisioned machines simulating aspects of learning and intelligence.
AI alignment: The study of making AI systems behave in accordance with human intentions and values. A capable general system may pursue goals in ways its designers did not intend.
Large language model: A neural network trained on large text datasets to predict and generate sequences of language. Language fluency alone does not establish broad, reliable competence across domains.

Linked from 17 pages