The Deep View

Courtesy of Deep View

OpenAI exists really for one singular reason: to invent artificial general intelligence, which OpenAI has defined as a system that is generally “smarter” than humans. 

Beyond being the driving force for just about every ounce of hype that has proliferated within this industry, AGI would mark a game-changing moment for OpenAI; based on the rather strange structure of the company’s capped for-profit/nonprofit mix, “the board determines when we’ve attained AGI…Such a system is excluded from IP licenses and other commercial terms with Microsoft, which only apply to pre-AGI technology.”

At an all-hands meeting last week, OpenAI shared a new set of tiers designed to track its progress toward reaching AGI, Bloomberg reported. 

• The first tier involves chatbots, the second involves human-level problem solving, nicknamed “reasoners,” the third involves “agents,” the fourth involves AI that can aid in new inventions and the fifth tier involves AI that can “do the work of an organization.” 
• OpenAI said it is currently at Level One, but it is on the brink of Level Two. 

The term AGI is not present on this list of tiers. 

Stages of Artificial Intelligence	OpenAI
Level 1	Chatbots, AI with conversational language
Level 2	Reasoners, human-level problem solving
Level 3	Agents, systems that can take actions
Level 4	Innovators, AI that can aid in invention
Level 5	Organizations, AI that can do the work of an organization

The context: As we’ve mentioned before, there is no universal scientific definition of AGI, nor is it universally accepted that AGI will ever be possible (some researchers have compared the belief that AGI is on its way to religious ferver). What scientists do know, however, is that large language models are not AGI and likely do not represent a pathway to it, as they are incapable of extrapolating beyond their training data. 

• A recent paper by Google DeepMind similarly attempted to carve out five levels designed to track AGI progress, and they took a wildly different approach than OpenAI did, categorizing capabilities on a scale of ‘emerging’ to ‘superhuman.’
• The five tiers, as Bloomberg pointed out, seem quite similar to the five levels of autonomous driving used to describe autonomous capabilities. 

My view: OpenAI is in the business of AI. This company has a tremendous amount of incentive to continue fueling hype around the tech. Hype drives investment, which is good for business. 

These tiers make very clear that OpenAI’s focus is on enticing the excitement of the corporate community; its categories are all about business application — if OpenAI turned around tomorrow and said it achieved Level 5, plenty of corporate executives would be thrilled at all the labor costs they could cut. 

But this is all divorced from reality and steeped in problematic ethics. I think these tiers make clear that what OpenAI is interested in is not curing cancer or solving climate change (which are naively simplistic goals, anyway) but instead seeing some return on its investment.