Comparing levels of AGI
General Artificial Intelligence (AGI) is an advanced form of AI that aims to simulate human intelligence across various activities and contexts. Unlike narrower forms of AI, which are specialized in specific tasks, AGI aspires to possess reasoning, planning, learning, perception, and communication abilities at a level equal to or sometimes surpassing human capabilities.
What makes AGI particularly interesting is its ability to solve problems and make decisions in a holistic manner, which can significantly transform various fields of life, from medicine and economics to engineering. Therefore, AGI is not just a technological issue. It is also an ethical and philosophical topic that requires reflection on how these technologies will be integrated into society.
This complicates discussions about the future of artificial intelligence, yet it is worth becoming familiar with these concepts.
An OpenAI Roadmap to AGI
A recent article on Bloomberg illustrates how OpenAI and Google understand and approach the concept of AGI.
As of now, OpenAI is at Level 1 of its five-tier system. This level encompasses AI systems that can interact with humans using conversational language, exemplified by ChatGPT. These AI systems have shown remarkable capabilities in understanding and generating human-like text, making them invaluable in various applications from customer support to content creation.
OpenAI's new classification system serves as a roadmap to AGI, outlining the gradual enhancement of AI capabilities:
Level 1 (Current): AI that can interact in conversational language.
Level 2 ("Reasoners"): AI that can solve problems like a human with a doctorate, without tools.
Level 3 ("Agents"): AI capable of performing tasks over several days on a user’s behalf.
Level 4 ("Innovators"): AI that can innovate and create new concepts.
Level 5 ("Organizations"): AI that can manage the functions of an entire organization.
A critical transition in OpenAI’s framework is moving from Level 2 to Level 3, which will involve developing “Agents.” These AI systems will be capable of autonomously performing tasks over extended periods. This leap will require significant advancements in AI's ability to plan, execute, and adapt to new information, embodying a more autonomous and dynamic form of intelligence.
Google DeepMind’s AGI classification framework
In a November 2023 paper, several researchers at Google DeepMind proposed a framework of five ascending levels of AI, including tiers such as “expert” and “superhuman.”
Similar to OpenAI Google's framework consists of 5 levels - with level 0 being no AGI. The rankings resemble a system often referred to in the automotive industry to assess the degree of automation for self-driving cars and define each level based on the percentile rank of skilled adults that the AI matches or exceeds in performance.
source: Position: Levels of AGI for Operationalizing Progress on the Path to AGI (arxiv.org)
This framework emphasizes both the technical capabilities and ethical considerations, ensuring that progress towards AGI is responsible and beneficial to society.
Comparing those two frameworks
OpenAI and Google DeepMind both aim to advance AI, but their approaches are quite different:
OpenAI’s Linear Framework
OpenAI follows a linear framework, focusing on a step-by-step progression toward Artificial General Intelligence (AGI). Each milestone builds directly on the previous one, gradually enhancing the AI’s capabilities. This structured, sequential process aims to achieve AGI by steadily advancing through defined stages.
Google DeepMind’s Matrixed Framework
In contrast, Google DeepMind employs a matrixed framework that evaluates AI development along multiple dimensions. Rather than just a linear progression, this approach considers how well AI capabilities generalize across various tasks and environments. The matrixed framework provides a more comprehensive assessment of AI progress, ensuring advancements are versatile and broadly applicable, not just incremental.
Future Goals: Advancing Toward AGI
Although the goal (AGI) is shared, understanding fundamental concepts and the paths to achieving the goals can differ.
OpenAI proposes a linear development path, with distinct stages that systematically increase AI's capabilities. On the other hand, Google DeepMind adopts a multidimensional approach, allowing for diverse and comprehensive analysis of AI progress.
As both organizations continue to innovate and refine their approaches, their collective efforts will undoubtedly contribute significantly to the eventual realization of AGI.
Enjoyed this article? Don't miss out! Join our LinkedIn community and subscribe to our newsletter to stay updated on the latest business and innovation news. Be part of the conversation and stay ahead of the curve.
Transform your business with our expert
