The term “artificial general intelligence,” also known as AGI, is commonly misused within the AI industry. Google DeepMind aims to clarify and strengthen the concept.
AGI refers to the capability of machines to encompass a broad range of human knowledge, distinguishing it from specialized computer programs that excel in specific tasks such as stock selection or language translation. While some view achieving machine flexibility akin to human cognition as a significant milestone, the definition of AGI often varies, with some viewing it as a threshold where machines align with human intelligence.
Google DeepMind researchers advocate for a more precise understanding of AGI, suggesting that rather than a singular goal, it should be viewed as a spectrum with current chatbots representing an initial stage. They emphasize the importance of defining key attributes like achievement, generality, and autonomy in AI systems.
Drawing inspiration from intelligent driving’s six levels of autonomy, the researchers propose a structured approach to categorize AI algorithms based on their performance and versatility. They argue that AGI models should not only exhibit universality but also achieve specific task objectives, surpassing human capabilities in a defined context.
The team introduces the “Levels of AGI” framework, categorizing algorithms as “emerging,” “competent,” “expert,” and “virtuoso” based on their performance relative to human abilities. This framework aids in distinguishing between specialized and general AI, facilitating a clearer understanding of AI capabilities.
While some may debate the classification of AI models within this framework, the DeepMind team’s initiative encourages a more nuanced consideration of AGI progression. By delineating between performance and generality, researchers like Julian Togelius commend the effort for shedding light on the AGI discourse and promoting thoughtful discussions on the subject.
