A particular clause in OpenAI’s contract with Microsoft, initially regarded as a far-off concern, has emerged as a critical issue in their partnership. This clause stipulates that if OpenAI’s board asserts the development of artificial general intelligence (AGI), it would restrict Microsoft’s access to the startup’s future technologies. With Microsoft having invested over $13 billion in OpenAI, there are reports suggesting that the tech giant is attempting to have this clause removed, even weighing the option of walking away from the agreement entirely.
Tensions surrounding the definition of AGI recently permeated OpenAI, culminating in internal debates regarding an unpublished research paper titled “Five Levels of General AI Capabilities.” This paper proposes a framework for categorizing various stages of AI technology. The implications of its assertions could hinder OpenAI’s ability to claim AGI status—an essential point in negotiations with Microsoft.
OpenAI representatives emphasized their commitment to developing empirical methods to assess AGI progress. They clarified that the "Five Levels" paper was not a scientific study but an initial classification attempt. According to OpenAI, AGI refers to "a highly autonomous system that outperforms humans at most economically valuable work.”
As both companies renegotiate their agreement, Microsoft desires continued access to OpenAI’s models, expecting to utilize them even if OpenAI declares AGI before their partnership ends in 2030. However, it seems Microsoft does not anticipate that OpenAI will reach AGI by this deadline. The AGI clause is viewed by some as a critical lever for OpenAI in this negotiation.
Discussions have even approached sensitive topics, including the possibility of OpenAI publicly accusing Microsoft of anticompetitive practices. Sources suggest that OpenAI may be closer to achieving AGI than previously thought, with CEO Sam Altman predicting its arrival during Donald Trump’s current term.
The contract includes aspects that define AGI and necessitate Microsoft’s approval for OpenAI to declare its achievement. It also restricts Microsoft from independently pursuing AGI or using OpenAI’s intellectual property through third parties.
The “Five Levels” paper, which outlines a five-step scale measuring AI advancements, describes earlier models as primarily at Level 1, where AI can understand and use language fluently. Predictions about societal implications and potential risks accompanying advances in AI capabilities are also included, but the document avoids a singular definition of AGI, favoring a spectrum approach.
Despite the paper’s development nearing completion, OpenAI has postponed its release, with internal discussions often citing the Microsoft partnership as a reason for the delay. However, OpenAI maintains that the decision was rooted in failing to meet technical criteria.
Ultimately, the ongoing negotiations underscore a complex relationship between these two prominent companies, significantly shaped by the evolving understanding of what AGI entails.
