OpenAI has introduced a five-tier system to measure its progress toward developing artificial intelligence (AI) capable of surpassing human performance, reported Bloomberg. This move aims to provide clearer insight into the company’s approach to AI safety and its vision for the future. The classification system was unveiled to employees during an all-hands meeting, an OpenAI spokesperson confirmed.
Reportedly, the tiers range from the current conversational AI (Level 1) to advanced AI that can operate an entire organization (Level 5). OpenAI, widely regarded as a frontrunner in the quest for more powerful AI systems, plans to share these levels with investors and other external stakeholders.
Currently, OpenAI considers itself at the first level, but nearing the second, known as “Reasoners.” This stage refers to AI systems capable of basic problem-solving tasks comparable to a human with a doctorate, but without access to additional tools.
During the same meeting, OpenAI’s leadership showcased a research project involving its GPT-4 model, demonstrating new capabilities that exhibit human-like reasoning.
As per Bloomberg, an insider who requested anonymity, mentioned that OpenAI continually tests new functionalities internally, a standard practice in the AI industry.
OpenAI has long aimed to create artificial general intelligence (AGI), which entails developing computers that outperform humans on most tasks. Although AGI does not currently exist, CEO Sam Altman has expressed optimism that it could be achieved within this decade. The criteria for reaching AGI have been a topic of debate among AI researchers.
In November 2023, researchers at Google DeepMind proposed a five-level framework for AI, including stages such as “expert” and “superhuman,” akin to the system used in the automotive industry for self-driving cars. OpenAI’s newly introduced levels also feature five ascending stages towards AGI. The third level, “Agents,” refers to AI systems capable of performing tasks over several days on behalf of users. The fourth level involves AI that can generate new innovations, and the highest level, “Organizations,” signifies AI that can operate autonomously within an organization.
These tiers were developed by OpenAI’s executives and senior leaders and are still considered a work in progress. The company intends to collect feedback from employees, investors, and its board to refine the levels further.
(With inputs from Bloomberg)