Report: Microsoft, OpenAI Tied AGI Metric to Profits, Not Intelligence

With OpenAI barreling towards its shift to being a for-profit business, a new report claims that the company’s partnership with Microsoft is even weirder than we knew.

Based on public disclosures, Microsoft has so far invested about $13 billion in OpenAI. Under the terms of this partnership, OpenAI hosts its resource-hungry AI language models on Microsoft Azure, and the software giant gets unfettered but non-exclusive access to those models to use in its own products and services.

Less well understood is that OpenAI has an escape hatch. If OpenAI ever obtains artificial general intelligence (AGI), it can stop providing Microsoft with access to its models. AGI is a fuzzy concept–it’s defined as AI so strong that it meets or exceeds human cognitive capabilities–but it’s understood to be vastly superior to the so-called narrow AI we use today to accomplish specific tasks. Reaching AGI is the reason OpenAI exists in the first place, though many scientists and researchers believe it to be technically impossible or at least something that will not occur for centuries or more. But whatever the opinion, AGI is fundamentally a technical term, something that can be proven or disproven in use.

Given this, one would assume that the Microsoft/OpenAI partnership will remain in place until (or unless) OpenAI creates a language model that achieves human cognitive capabilities. But according to a new report in The Information, that is not the case. Instead, the two firms have defined AGI as OpenAI generating at least $100 billion in profits (I assume that’s annual profits, but I can’t access the original article.) Aside from being patently ludicrous, that means that the Microsoft/OpenAI partnership will remain in place basically forever, given how unprofitable OpenAI is. The company has never made a profit, and is on track to earn roughly $4 billion in revenues in 2024.

This raises all kinds of questions, most notably about the intelligence of the humans who forged this agreement. Instead of pretending to tie their partnership to AGI, Microsoft and OpenAI could have simply tied it explicitly to profits.

From what I can tell, this clause is decidedly in Microsoft’s favor. Microsoft collects 20 percent of OpenAI’s revenues off the top, and because OpenAI’s cloud-hosted models are so expensive to run, it’s unlikely the company can be profitable any time soon. And OpenAI can’t shop around for a better deal with one of the few other companies that could meet its processing needs–Amazon and Google, basically–in a bid to become profitable more quickly. This resets our understanding of the partnership dramatically, as it seemed that Microsoft was assuming a lot of risk since OpenAI is free to license its services to other companies. But OpenAI is, if anything, more stuck with Microsoft than the reverse.

This report comes at an interesting time, too. OpenAI is in the early stages of a corporate restructuring that will remove its for-profit subsidiary from the control of the non-profit parent company. Yesterday, it announced that the OpenAI board of directors is evaluating different structural plans to “ensure artificial general intelligence (AGI) benefits all of humanity” while making the non-profit “sustainable” and the for-profit a “long-term success.”

“We have to become an enduring company,” OpenAI explains. “The world is moving to build out a new infrastructure of energy, land use, chips, datacenters, data, AI models, and AI systems for the 21st century economy. We seek to evolve in order to take the next step in our mission, helping to build the AGI economy and ensuring it benefits humanity.”