The capability ladder is a way to visualize and structure the development of artificial intelligence. At the lower rungs of the ladder, AI systems are designed to perform narrow, well-defined tasks such as image recognition, language translation, or playing specific games. As one moves up the ladder, these systems gain the ability to handle more complex and less structured problems, ultimately aiming for artificial general intelligence (AGI) that can understand, learn, and apply knowledge across a wide range of tasks, similar to human cognitive abilities. The framework helps researchers and developers to systematically build and evaluate AI progress, setting clear benchmarks for capabilities at each level.
The concept of a capability ladder gained traction as AI research evolved from narrow AI in the mid-20th century to more ambitious goals. The term itself became more prominent in the 2000s as discussions about AGI and the scaling of AI capabilities became central to the field.
Significant figures in the development of the capability ladder concept include early AI pioneers like Alan Turing and John McCarthy, who laid the groundwork for AI as a field, as well as more recent contributors such as Ray Kurzweil and Nick Bostrom, who have extensively discussed the progression towards AGI and the scaling of AI capabilities.