OpenAI's founding mission is to “enable artificial general intelligence (AGI) to benefit all humanity,” and the company defines AGI as an autonomous system that “outperforms humans at the most economically rewarding tasks.” From this, we can infer that the company will at least attempt to actually develop AGI at some point.
Yesterday, Bloomberg reported that OpenAI has come up with a five-level classification system to track progress toward AGI. On the one hand, the five-level formulation may make it seem like Sam Altman and co. are plowing toward AGI in a systematic, metrics-driven way. An OpenAI spokesperson said the stages the company shared with employees at an all-hands meeting range from conversational chatbots (Level 1), which are available today, to AI that can run tasks across an entire organization (Level 5). In between, Level 2 tackles human-level problem solving, Level 3 is about agents that allow the system to take action, and Level 4 moves to AI that can help invent.
The problem is, the term AGI is nowhere to be found on this list. Will OpenAI claim AGI when it reaches level 5, “AI that can do organizational work”? Is level 4 when AI helps invent things like curing cancer? Or is level 6 on the horizon? What about artificial superintelligence (ASI), which OpenAI talks about as a kind of AI system more intelligent than all humans combined? Where does ASI fall on this five-level scale?
To be fair, even OpenAI's definition of AGI is not widely accepted by others in the AI research community, and there is no widely accepted definition of intelligence, which makes the entire attempt to define AI capabilities in terms of being “more intelligent” than humans problematic.
Last year, OpenAI rival Google DeepMind published a research paper outlining stages of AI progress that were quite different from OpenAI's. The list did not include AI doing “organizational tasks.” Instead, it listed “emerging” (which includes today's chatbots), “competent,” “expert,” “virtuoso,” and “superhuman” — that is, performing a wide range of tasks better than all humans, including tasks that are beyond human capabilities, such as deciphering human thoughts or predicting future events. Google DeepMind researchers emphasized that a level beyond “emerging” has not yet been achieved.
OpenAI executives reportedly told employees that the company is currently at Level 1 in the classification hierarchy, but is close to reaching Level 2, the “reasoner” level, “which refers to a system that can perform basic problem-solving tasks as well as a PhD-level educated human who does not have access to the tools.”
But how close will this get OpenAI to AGI? There's no way to know, and maybe that's exactly the point.
After all, even if OpenAI is keen that AGI “benefit all humanity,” it certainly wants to benefit. And to do so requires a thoughtful strategy. For example, the term “AGI” is not up for debate because it is so loaded with meaning. Why does it scare people? This is another good reason to say that the company is on the brink of Level 2. It shows that the company is neither lagging behind nor unnecessarily ahead.
While the five tiers may seem to imply a slow and steady climb up the ladder of capability to level 5, it’s just as likely that OpenAI will keep their AGI card secret and suddenly announce “Eureka!”, leapfrogging a level or two to achieve AGI.
That's because once OpenAI reaches AGI, everything will change in its favor: according to the company's organization, “the Board of Directors will determine when AGI is achieved…Such systems will be excluded from the IP license and other commercial terms with Microsoft that apply only to pre-AGI technologies.”
This is where it gets really interesting, if we remember that Sam Altman was ousted from OpenAI in November 2023. Before that, the six-person OpenAI board that would have made that decision was very different from the current one. It remains to be seen at what stage the current board (Altman and board chair Brett Taylor, Adam D'Angelo, Dr. Sue Desmond-Hellman, retired US Army General Paul M. Nakasone, Nicole Seligman, Fiji Simo, and Larry Summers) will decide that OpenAI's version of “AGI” has arrived.
CEO Daily provides essential context for the news business leaders need to know. More than 125,000 readers trust CEO Daily every weekday morning for insight on and from the C-level. Subscribe now.
