OpenAI Shares Five-Step Plan Toward “Human-Level” AI Computing

OpenAI has introduced a five-tier system to gauge its progress in developing artificial intelligence capable of surpassing human performance. This initiative aims to enhance public understanding of the startup’s approach to AI safety and future developments.

During an all-hands meeting on Tuesday, OpenAI, known for creating ChatGPT and leading in the AI advancement race, unveiled this new classification to its employees. An OpenAI spokesperson confirmed that these tiers, intended to be shared with investors and the broader public, span from current AI capable of conversational interactions (Level 1) to AI capable of managing organizational tasks (Level 5).

According to the spokesperson, OpenAI executives informed employees that the company is currently at the first level but is close to reaching the second level, termed “Reasoners.” This level signifies systems capable of basic problem-solving tasks equivalent to a human with a doctorate-level education, without external tools.

During the same meeting, OpenAI leadership showcased a research project involving the GPT-4 AI model, demonstrating new capabilities that hint at human-like reasoning. A person familiar with the discussion, who requested anonymity due to lack of authorization to speak to the press, confirmed this. The spokesperson, when asked about the demonstration, mentioned that OpenAI frequently tests new internal capabilities, a common practice in the industry.

OpenAI has long been working towards artificial general intelligence – computers that outperform humans in most tasks. Although such systems do not currently exist, CEO Sam Altman has previously expressed the belief that AGI could be achieved within this decade.

AI researchers have historically debated the criteria for achieving AGI. In a November 2023 paper, Google DeepMind researchers proposed a five-level AI framework, including tiers like “expert” and “superhuman.” This structure is similar to the classification system used in the automotive industry to evaluate the automation level of self-driving cars.

According to OpenAI’s newly defined levels, the third tier, “Agents,” refers to AI systems capable of taking actions over several days on a user’s behalf. Level 4, termed “Innovators,” describes AI that can contribute to new inventions. The highest tier, “Organizations,” signifies AI that can perform all tasks of an organization.

These levels, developed by OpenAI executives and senior leaders, are considered a work in progress. The company plans to gather feedback from employees, investors, and board members, and may adjust the levels based on this input.