Bloomberg recently scored an inside scoop from OpenAI, unveiling a brand-new internal scale designed to track the progress of its AI capabilities. Picture it as a ladder with five rungs, each representing a distinct level of AI prowess:
Level 1: Chatbots – These AI have conversational skills akin to the likes of ChatGPT.
Level 2: Reasoners – Here, AI starts flexing human-like reasoning and problem-solving muscles.
Level 3: Agents – At this stage, AI can operate autonomously, without a human chaperone.
Level 4: Innovators – Now, we’re talking about AI capable of inventing or significantly aiding the process of invention.
Level 5: Organizations – This is where AI can handle the workload of an entire organization.
Conceived by the visionaries at OpenAI, this scale is a living document, subject to tweaks and modifications as AI continues to evolve. Amid the hype from OpenAI, Microsoft, Meta, Google, and others, all jostling to release the latest versions of their language models, the stark reality is that we remain firmly planted at Level 1. This is the same message OpenAI conveyed to its own employees.
The holy grail for OpenAI, and pretty much every AI enterprise, is achieving Artificial General Intelligence (AGI) – an AI that can outshine humans across the board. Curiously, the scale doesn’t pinpoint which level marks the arrival of AGI. One might assume that Level 4, where AI can invent and innovate, would already surpass the average human in some areas. Level 5, undoubtedly, would demand AGI due to the sheer breadth and depth of work required to support an entire organization.
Yet, the precise definition of AGI remains elusive. Sure, everyone agrees it should outclass humans, but there’s a vast spectrum of human capability to consider. Even an AI at Level 3 might already be eclipsing some individuals. Definitions aside, the burning questions are: how close are we to AGI, and what will it take to get there?
Road to AGI: When Will We Achieve It?
OpenAI’s CEO, Sam Altman, envisions a timeline of about five years, maybe a tad more, to achieve AGI, provided there’s a steady stream of investment—up to $50 billion annually.
"Whether we burn $500 million a year or $5 billion—or $50 billion a year—I don’t care, I genuinely don’t…..As long as we can figure out a way to pay the bills, we’re making AGI. It’s going to be expensive."
Sam Altman, CEO, OpenAI
Echoing this sentiment, Dario Amodei, CEO of Anthropic, offers a similar forecast for developing language models that can outperform humans. He, too, underscores the hefty price tag.
"I think if we go to $10 or $100 billion, and I think that will happen in 2025, 2026, maybe 2027... then I think there is a good chance that by that time we'll be able to get models that are better than most humans at most things..."
Dario Amodei, CEO, Anthropic
Dario notes that current training costs for some language models already exceed a billion dollars, and these are just Level 1 capabilities. So, $50 to $100 billion a year to train AGI might not be as crazy as it sounds.
McKinsey shares this optimism, projecting that by 2030, general AI will perform at a median level of human capability across most domains. This aligns with the chart below from McKinsey & Company, illustrating various AI capabilities.
While there’s a chorus of voices suggesting AGI could be realized by the end of this decade, it’s worth remembering that AGI remains a broad term. If the goal is an AI that matches median human performance in various areas, as McKinsey suggests, then it seems plausible.
However, if we’re aiming for an AI at Level 5 on OpenAI’s scale—one that can run an entire organization—that’s a different beast entirely. Such an AI would need to surpass not just average human performance but also excel in specialized functions and executive decision-making. And don't forget, the massive amount of funding required would be something the world has yet to see.
Unless we settle on a definition of AGI that doesn’t require the capabilities of a Level 5 AI on OpenAI’s scale, achieving AGI might remain a lofty goal for the next decade.
Comments