OpenAI has recently introduced a five-tier system designed to map its progress toward the elusive goal of developing artificial general intelligence (AGI). This new framework, disclosed during an all-hands meeting and reported by Bloomberg, aims to help employees and the public better understand OpenAI’s advancements. However, it’s important to note that this classification system describes technologies that do not yet exist, leading some to see it more as a marketing strategy to attract investors.
AGI, a hypothetical form of artificial intelligence capable of performing any intellectual task that a human can do, is at the center of OpenAI’s vision. This technology continues to generate significant interest and discussion, despite its potentially disruptive impact on society. OpenAI CEO Sam Altman has repeatedly expressed confidence that AGI could become a reality within this decade. Much of Altman’s public communication revolves around preparing for the societal changes AGI might bring, making a structured approach to communicate AI milestones both internally and to the public quite logical.
OpenAI’s five levels span from current capabilities to highly advanced, hypothetical systems. Presently, the company’s technology, exemplified by the GPT-4 model that powers ChatGPT, is classified at Level 1. This level encompasses AI systems capable of conversational interactions. OpenAI suggests it is on the brink of achieving Level 2, known as “Reasoners,” which would signify AI capable of solving problems at a human level without needing external tools.
The five stages of artificial intelligence, according to OpenAI, are:
- Chatbots: AI with conversational language capabilities.
- Reasoners: AI with human-level problem-solving abilities.
- Agents: AI systems that can take actions independently.
- Innovators: AI that can aid in invention.
- Organizations: AI capable of performing the work of an entire organization.
During the recent meeting, OpenAI leaders showcased a research project utilizing their GPT-4 model, which they believe is nearing Level 2. This would represent AI achieving reasoning abilities comparable to a human with a doctorate.
The upper levels of OpenAI’s classification propose even more advanced, hypothetical AI capabilities. Level 3 “Agents” would autonomously handle tasks for extended periods, while Level 4 “Innovators” would generate novel inventions. Ultimately, Level 5 envisions AI managing entire organizations.
It’s worth noting that this classification system is still under development, with OpenAI planning to refine it based on feedback from employees, investors, and board members.
The concept of ranking AI capabilities is not unique to OpenAI. Similar frameworks have been proposed, such as Google DeepMind’s five-level assessment system and Anthropic’s “AI Safety Levels,” which focus on the safety and risks associated with AI. However, these systems often raise questions about the feasibility of quantifying AI progress and what truly constitutes an advancement.
Despite the excitement, there remains no consensus within the AI research community on how to measure progress toward AGI or even if AGI is a well-defined, achievable goal. As such, OpenAI’s five-tier system might be best viewed as a visionary tool designed to communicate its aspirational targets and attract investment, rather than a precise measure of technical progress.
Post Picture: DALL-E3
