Sam Altman, CEO of OpenAI, touts the importance of the non-profit ownership structure of his company, which is claimed to foster openness and safety and ensure benefits are distributed widely. There is little evidence that these qualities are actually embedded in OpenAI LLC.
The Wall Street Journal reported that OpenAI will likely be put up for sale. This news would be quite a departure from the claim made by OpenAI CEO, Sam Altman, that he would be unwilling to take the company public.
Altman implied that the company’s non-profit ownership structure protects the company from the short-term interests of shareholders. The non-profit ownership structure also ensures that the benefits accrued from artificial intelligence (AI) would be distributed broadly, AI systems’ safety would be assured, and OpenAI would work to serve the “best interests of humanity.”
I have previously written that such a structure is a viable way to achieve commercial success, while upholding societal values, as demonstrated by Patagonia and Novo Nordisk.
Given this previous research, I was expecting that OpenAI would also be influenced by its non-profit values, even though OpenAI LLC had pivoted to a capped-profit structure in 2019. Beyond Altman’s rhetoric, I was unable to uncover any evidence that OpenAI is operating any differently from its competitors.
OpenAI’s Non-Profit Ownership
OpenAI was founded as a non-profit by research engineers and scientists, with major donations from highly successful entrepreneurs, including Altman, Elon Musk (xAI founder), Reid Hoffman (LinkedIn co-founder), and Peter Thiel (PayPal co-founder). OpenAI’s original press release said OpenAI’s non-profit structure would ensure the company would “prioritize a good outcome for all over its own self-interest.”
OpenAI’s original ambition was audacious and exciting. It was to unlock artificial general intelligence (AGI). Its mission as a non-profit was to ensure the safety of its application. And, Altman remains committed to this ambition even today, having posted on Reddit on September 26, 2023 that OpenAI had achieved AGI internally (but later claimed to be “just memeing”).
The company was going to create machines that simulated human learning and reasoning by hiring the best people in the industry. It intended to spend “only a tiny fraction” of its $1-billion initial investment in the first few years. The company called itself OpenAI to reflect its openness to sharing research and collaborating in deploying the technologies.
Nathan Lambert, an AI researcher and podcaster of The Retort AI, told me OpenAI had three cultures that cemented the company in the early days. One: a researcher culture interested in new ideas and openness. Two: A non-profit culture interested in safety and wealth distribution. And three: A Bay Area technologist culture interested in pushing cool ideas. These cultures melded together well – until the company shifted its ownership structure.
A Critical Turn To ‘Capped’ Profits
OpenAI caught both industry insiders and outsiders off guard by its about-turn to a “capped-profit” company in March 2019. The non-profit (OpenAI Inc.) would manage the for-profit company (OpenAI LLC). A figure that illustrates OpenAI’s ownership structure is available on its website.
OpenAI President and Co-Founder, Greg Brockman, defended the move to a capped-profit company, saying the company had to find a way to generate more capital more quickly. OpenAI’s competitors were doubling their computational resources every three to four months, and OpenAI needed to keep up.
Under the new structure, investors’ profits would be capped at 100 times their investment. Any additional profits would be directed to the supervising non-profit company, which would then distribute the profits for ventures intended for societal good.
Later investors received much smaller profit multipliers. For example, Microsoft’s $1-billion investment in 2019 was rumored to be capped at 20 times. Forbes.com reported that Microsoft was to receive 75 per cent of OpenAI’s profits for its $10-billion investment in January 2023, until it secures its investment returns. After it breaks even on that investment, Microsoft would secure a 49-per-cent stake in the company.
This capped-profit structure was a significant departure from OpenAI’s initial non-profit structure. It would allow OpenAI to compete more aggressively in a hypercompetitive industry by channeling its revenues back into the company.
What remains unclear is how the “100-times profit” is calculated and how profits beyond the cap will be dispersed. My desktop research and interviews with industry experts yielded few details. A company that advocates the principles of openness has kept its ownership and governance structure rather closed.
Owned By A Non-Profit, Acting Like a For-Profit
There are several indicators that OpenAI’s non-profit ownership structure has very little influence in its current operations.
First, OpenAI no longer appears to follow the principles of ‘openness’ especially in its training of algorithms and data, which calls into question what is open about OpenAI. In fact, Ilya Sutskever, Chief Scientist and OpenAI co-founder, told The Verge OpenAI was wrong in saying it would make all its research open. “If you believe, as we do, that at some point, AI — AGI — is going to be extremely, unbelievably potent, then it just does not make sense to open-source,” he said. Yet, there is the counter argument that if AI, or AGI, is an arms race, then the technology will be made safer through openness and transparency.
Second, the company scrapes and processes all content on the web without permission from those who own the data. In other words, OpenAI profits from content that others have loaded. Whereas OpenAI benefits from contributions made broadly, it has not yet upheld its promise to distribute benefits broadly.
Third, most non-profit-owned companies typically contribute a fraction of their annual profits to the non-profit. According to TechCrunch, OpenAI’s non-profit owner received only $3,066 in donations in 2021. The Wall Street Journal reported OpenAI’s profits were $1 billion annually and would be significantly more in 2024. The high cap to OpenAI’s profits ensures no contributions are actually made to its non-profit owner.
Finally, employees are paid exceptionally well at OpenAI, which can make it difficult for employees to uphold a social mission when they are being rewarded for their commercial outputs. According to Ritwik Gupta, a Berkeley PhD student who also advises the U.S. Department of Justice on AI and AI policy, OpenAI is known among leading AI scientists and engineers for hiring the best and paying them the most.
Non-profit Ownership Structures Do Not Always Matter
Almost all AI companies tout safety concerns. Google claims it will develop technology that will “be socially beneficial” and Musk signed an open letter that called for pausing AI because of the inherent risks. Altman also recognizes the importance of AI safety, but points to OpenAI’s non-profit ownership structure to offer assurances to the company’s commitment to safety.
When I embarked on this research, I believed that non-profit ownership provided social guardrails for commercial firms. Whereas this may be true for many firms, it cannot be generalized to all non-profit owned firms.
Note: I would like to thank Minali Giani for her research support.
Read the full article here