TUESDAY BROUGHT CINCO DE MAYO and the sixth day in the trial of Elon Musk’s claims against Sam Altman, Greg Brockman and OpenAI, the San Francisco developer of artificial intelligence.
The press and crowds at the federal courthouse in Oakland have thinned as the celebrity factor diminished after Musk testified last week.
The trial is addressing Musk’s allegations that OpenAI was chartered as a nonprofit corporation to develop artificial general intelligence for the benefit of humanity, but that under Altman’s leadership, the company has abandoned its charitable mission — with the help of partner and investor Microsoft, also named in Musk’s suit.
The jury heard more testimony Tuesday from Greg Brockman, defendant in the case and president of OpenAI, who alleged that Musk not only supported creating a for-profit arm of the company, but wanted to have complete control over it.
Brockman returned to the stand to complete his examination by Sarah Eddy, one of the lawyers representing the OpenAI defendants.
In a jury trial, the plaintiff (Musk) typically presents their evidence first, followed by the defendants. But this time, Musk’s lawyers called Brockman to testify as part of presenting Musk’s case, even though Brockman is Eddy’s client. That reversal of the usual order gave Eddy the opportunity to examine Brockman at length and essentially present OpenAI’s overall narrative in the middle of Musk’s case. She took full advantage of the opportunity.
Brockman was polished, focused and seemingly anxious to get his side of the story in front of the jury.
In his testimony the day before, Brockman told the origin story of OpenAI. On Tuesday, Eddy jumped forward to a critical six-week period in August and September of 2017 when Musk, Altman, Brockman and Ilya Sutskever — the four members of the founding group — considered how to solve what had emerged as a major hurdle for the startup: how to raise the hundreds of millions or billions that would be needed to obtain the “compute” required for the development of artificial general intelligence, or AGI. “Compute” is tech slang for computer processing time and capacity. AGI is frequently defined as machine intelligence able to reason as well or better than humans at all human endeavors.
Fulfilling the mission
In the fall of 2017, OpenAI, a nonprofit corporation, had been living on Musk’s charitable donations and trying to expand its pool of donors. OpenAI was formed as a nonprofit with the mission of developing artificial intelligence for the benefit of humanity (and not for individual profit). According to Brockman’s testimony Tuesday, the founders concluded that to fulfill “the mission” they had to raise much more money than anticipated, and the nonprofit structure was holding them back.
The discussions — characterized by Brockman as wide-ranging and intense — considered a number of options to address the issue, including the creation of a for-profit subsidiary of the nonprofit OpenAI, or combining OpenAI with Musk’s Tesla, and even an initial offering of a cryptocurrency.
None of the ideas were implemented at the time. However, two years later, in 2019, after Musk was no longer a member of the OpenAI board of directors, a for-profit subsidiary was created.
Eddy walked Brockman through that period with a granularity and focus that allowed Brockman to build context for comments that he made in his personal journal. Musk had relied on select comments from the journal to demonstrate that Brockman and fellow founders were insincere with Musk and pursuing a path inconsistent with the nonprofit mission.
Brockman’s journal was provided to Musk’s lawyers during the “discovery” process, where the parties exchange relevant documents. The portion of Brockman’s journal that covered the six-week period was roughly a hundred typed pages.
The appearance of a detailed personal journal is not a regular occurrence in business litigation, where cases generally feature corporate records and emails or texts between the parties. Brockman described the journal as the place where he recorded issues and events in his personal and professional life. He said it was “stream of thought” and a place where he tried to “puzzle things out.” He said it was deeply personal and even contained contradictory statements.
In a series of questions, Eddy allowed Brockman to paint a picture of Musk that was very different than the one Musk offered in his testimony. According to Brockman, Musk was hard to reach, did not understand AI deeply, and — in his opinion — was unlikely to set aside time to learn it thoroughly. He said Musk’s aggressive approach created friction at OpenAI. He recounted a story about how Musk’s harsh comments about a project in early development crushed the engineer so thoroughly that he “almost left the field.”
On the key issue of using a for-profit vehicle to raise money, Musk was, according to Brockman, agreeable to the creation of a for-profit subsidiary, but only if he had complete control of it and only if his share of the equity was far larger than those of his co-founders.
Brockman found Musk hard to reach, lacking a deep understanding of AI and unlikely to set aside time to learn it thoroughly. He said Musk’s aggressive approach created friction.
In Brockman’s version, Musk bridled when Brockman, Sutskever and Altman pushed to get more equity for themselves than Musk thought appropriate, and again when they did not want to give him the “absolute control” that Brockman said Musk sought.
Brockman challenged Musk’s claim that OpenAI was all about humanity, not individual profits. He said that in one discussion about equity in the for-profit, Musk said he wanted a bigger stake that would, if AGI was achieved, help fund his efforts to build a space colony on Mars.
The equity issue was important, Brockman said, but in the end, Musk’s desire for control was an unsolvable issue. Brockman said he did not believe that control of AGI should be in the hands of a single person. He recalled Musk saying that Demis Hassabis of Google’s DeepMind shouldn’t be allowed to be the “Dictator of AGI,” and Brockman believed the same thing about Musk.
Raising the stakes
Brockman wrote in an email presented to the court that if the company ultimately achieves AGI, the conversation about who controls OpenAI will “turn out to have been the highest stakes conversation the world has seen.”
He described a meeting in August 2017, when Musk became so angry that Brockman and the others did not agree to his proposal for control and equity that he stood up and stormed around the table. Brockman said he was worried that Musk was going to hit him.
Musk cut off quarterly charitable donations to OpenAI in 2018, in what Brockman said was a move to put pressure on the others to agree to his terms.
Brockman testified that after they rejected Musk’s proposal, Musk gave an ultimatum to the three that they should either leave to start a for-profit operation or accept the existing nonprofit structure.
After they said they would accept the existing nonprofit arrangements, Musk would not restart charitable contributions without conditions, Brockman continued. Altman, Brockman and Sutskever had to agree to commit to staying at OpenAI for two years, accept a non-solicitation obligation after they left (stopping them from recruiting OpenAI employees), and they had to agree Musk would get appointments to two additional seats on the nonprofit board.
The other three founders did not agree to those conditions, but they did not leave OpenAI. The quarterly contributions did not recommence.
Brockman said Musk did little to help the charitable fundraising that was still the primary method for raising money. Brockman added that at some point Musk lost interest and “gave up on OpenAI.”
Musk left OpenAI’s board in February 2018.
Brockman testified it was the “end of an era” but that his principal feeling was “relief” because they were free to pursue the mission of developing artificial intelligence for humanity.
Eddy asked Brockman if he ever promised Musk that OpenAI would not create a for-profit subsidiary. Brockman said he did not make such a promise.
She followed by asking whether he told Musk that OpenAI would always remain a for-profit. Brockman said he did not.
On redirect of Brockman, Musk’s lead attorney Steven Molo focused on Brockman’s personal profits from the decision to create a for-profit subsidiary and obtain stock that is today worth $30 billion.
The trial will continue Wednesday with the highly anticipated testimony of Shivon Zilis, who served as Musk’s senior advisor and on the nonprofit OpenAI board from 2019 to 2023. Because Musk was hard to reach, Brockman said he and Sutskever referred to her as “Proxy Elon.” Musk is also the father of four of her children.