The Road to Artificial General Intelligence (AGI)

The recent leadership crisis at OpenAI has sparked rumors that the organization has achieved Artificial General Intelligence (AGI) with ChatGPT. However, the research community has yet to reach a consensus on what exactly qualifies as AGI.

In this article, we explore a fascinating pre-print from Google DeepMind, which introduces a taxonomy of “levels of AGI” to help us better understand the progress being made towards AGI.

Defining AGI Levels:
The taxonomy proposed by Google DeepMind categorizes AGI into six levels, ranging from Level 0 (no AI) to Level 5 (superhuman). These levels are determined based on an AI’s capabilities compared to those of human adults. To differentiate between narrow AI and general AI, the authors explain that narrow AI focuses on specific tasks, while general AI exhibits broader capabilities.

Level 0: No AI:
At this level, we have task-specific computer programs without any AI. These programs can be likened to compliers, which carry out specific functions but lack the ability to learn or adapt.

Level 1: Emerging:
This level represents AI that is at least equal to or slightly better than an unskilled humanin performing tasks. Narrow AI at this stage consists of simple rule-based systems, while general AI examples include ChatGPT, LLama 2, and Bard.

Level 2: Competent:
At Level 2, AI reaches a level of competence that surpasses the 50th percentile of skilled adults. Narrow AI examples include smart speakers and language models designed for specific tasks like essay writing — remains to be achieved.

Level 3: Expert:
Here, AI surpasses at least the 90th percentile of skilled adults in terms of proficiency. Narrow AI examples at this stage include spelling and grammar checkers such as Grammarly, as well as generative image models like Midjourney — remains to be achieved.

Level 4: Virtuoso:
AI at this level outshines at least the 99th percentile of skilled adults. Narrow AI examples include AlphaGo, the AI system that achieved remarkable success in the game of Go — remains to be achieved.

Level 5: Superhuman:
At Level 5, AI surpasses the capabilities of every human in different tasks. Narrow AI examples include Alpha Fold, which can predict a protein’s 3D structure from its amino acid sequence — remains to be achieved.

The Road ahead

The taxonomy of “levels of AGI” proposed by Google DeepMind provides a helpful framework for understanding the progression towards AGI. It allows us to categorize AI systems based on their capabilities compared to human adults. While ChatGPT and other advanced AI models show promising signs of AGI-like traits, true general AI, which surpasses human performance across various domains, is yet to be achieved.

The path to AGI is a complex and evolving journey, with multiple milestones to be reached along the way. As we continue to advance in AI research and development, it’s important to maintain a balanced perspective and acknowledge both the progress made and the challenges that lie ahead. AGI holds significant potential for transformative impact, and its realization will undoubtedly reshape various industries and aspects of our lives.

We are intrigued and optimistic about the possibilities it brings, but cautious about the challenges that need to be addressed before we can fully embrace its potential. AGI represents an exciting frontier in technology, and its emergence will shape the future in unimaginable ways.

“Let’s continue to monitor the progress, engage in responsible AI development, and navigate the road to AGI with a blend of enthusiasm and thoughtful consideration.”