Detangling OpenAI’s Palace Intrigue

In a world where personal privacy has essentially vanished, it’s ironic that we may never know the full story of OpenAI’s mid-November boardroom coup. John Best and I attempted to detangle the saga’s various threads on a wide-ranging episode of The BIGCast. As a companion piece, here’s a somewhat tidier synopsis of the complex and mysterious path leading to CEO/Co-Founder Sam Altman’s brief exile.

Bear in mind these thoughts are based on information publicly available through December 1- there will undoubtedly be further revelations and developments. Our reflections are informed by some fine journalistic reporting, links to which can found in our BIGCast show notes. Thanks also to our own Melissa Showen for additional research, particularly regarding Elon Musk’s inevitable fingerprints.

ChatGPT, OpenAI’s breakout consumer product, is generating additional headlines for celebrating its first anniversary. More accurately, it’s the anniversary of a free user-friendly version based on OpenAI’s GPT-3.5 large language model (since upgraded to GPT-4). GPT-1 was introduced in 2018, and the formation of OpenAI’s research entity itself dates back to December 2015.

An understanding of the initiative’s origins is important to grasp the current tension. OpenAI began life as a non-profit research organization launched by a group of Silicon Valley visionaries including Reid Hoffman (LinkedIn), Peter Thiel (PayPal), Elon Musk (Tesla) and representatives of the startup accelerator Y Combinator, including Altman himself.

OpenAI’s charter, published in 2018, declares a mission to “ensure that artificial general intelligence (AGI) benefits all of humanity. We will attempt to directly build safe and beneficial AGI, but will also consider our mission fulfilled if our work aids others to achieve this outcome.”

According to TechCrunch, Musk was the largest donor toward OpenAI’s early operations. He resigned from the board- and ceased his funding- in 2018, however, citing potential conflicts with Tesla’s own AGI development. Musk has since criticized OpenAI, accusing it of reneging on its “open” credo.

The plot thickened further in 2019 when OpenAI established a for-profit arm (which houses ChatGPT), situated under the non-profit structure. Two factors likely explain this move. One was the need to attract operating funds, particularly in the absence of Musk’s largesse. As a concession to its altruistic mission, OpenAI capped returns on equity to… wait for it… 100 times the initial investment.

A stated rationale for the for-profit subsidiary was recruitment. The going rate for skilled top-notch AI programmers had skyrocketed. Those attracted to OpenAI’s mission willingly accepted below market pay packages, but the gap has widened to unsustainable levels. OpenAI needed to be able to offer the prospect of equity upside in order to attract the talent necessary to fulfill its mission.

These two points share a common thread- “Money Changes Everything,” to quote a song by new wave band the Brains- I’m sure ChatGPT can tell you about the song’s subsequent cover versions.

OpenAI soon had the funding it needed thanks to a $13 billion injection from Microsoft- mostly in the form of in-kind data services, which along with salaries comprise ChatGPT’s largest expense- in exchange for a 49% stake in the for-profit subsidiary. Amazingly, Microsoft did not demand a board seat as a part of the deal- a glitch that played into November’s turmoil.

Bear in mind, OpenAI’s board was assembled with the non-profit parent’s altruistic mission as its primary tenet. As such, board composition differed from that of a typical tech startup. Two members- Tasha McCauley and Helen Toner- are advocates of the “effective altruism” movement, both with ties to the Center for the Governance of AI- a logical fit with the non-profit side of OpenAI’s mission. Other members came and went with surprising regularity, and the firm showed little apparent regard for maintaining a consistent number of seats.

At the time of November’s coup the board was down to four- Toner, McCauley, OpenAI’s Chief Scientist Ilya Sutskever, and Quora CEO/Co-Founder Adam D’Angelo. The foursome issued a statement that it had lost confidence in Altman because “he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities.”

It can reasonably be assumed that McCauley and Toner supported Altman’s ouster- the latter recently published a paper on AGI in her role at Georgetown’s Center for Security and Emerging Technology that many read as a cautionary tale implicating ChatGPT. But even if not unanimous, who provided the decisive third vote? Days after the firing Sutskever was among the signatories to a letter threatening to resign if Altman were not reinstated. He also expressed “regret” for any role he played in the board’s actions. D’Angelo is the reconstituted board’s sole holdover; it’s hard to imagine he’d be retained (or would stay) after a no confidence vote.

Multiple reports, including a recent TED Talk by Sutskever himself, imply that OpenAI is on the verge of the sort of breakthrough that poses a threat to the “safe and beneficial AGI” the organization was designed to ensure. Lost amid the chaos is the distinct possibility that the board may well have acted appropriately, poor execution notwithstanding, given OpenAI’s stated charter. It will be interesting to watch whether any changes to the charter or bylaws are forthcoming.

Meanwhile, it’s terrible optics for OpenAI to have dismissed its only two female directors, creating an all-male board with Salesforce Co-CEO Bret Taylor and former Harvard President and US Secretary of the Treasury Larry Summers joining D’Angelo. Presumably more appointments are forthcoming- plus Microsoft secured a seat, albeit a non-voting one, after being blindsided by the actions impacting its $13 billion investment. Microsoft could earn some ethical capital by appointing a female board observer.

Despite the current détente, we still have the awkward situation of a for-profit entity reportedly valued at more than $80 billion governed under a non-profit structure- a key reason this story probably isn’t over.