Dueling Letters from the OpenAI Board

0:00Today on the AI Daily Brief, we're reading a set of warring op-eds from current and former OpenAI board members. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:24Hello, friends. Happy weekend. It being a weekend, that means, of course, that we are doing a long reads episode. And today we have something kind of interesting. At the beginning of the week, two former OpenAI board members wrote an op-ed for The Economist talking about why self-governance wasn't working. This generated a ton of conversation, some of which even happened on this show, and also led to a set of current board members from OpenAI responding. With a little help from some voices from the Eleven Labs Voice Lab, we're now going to listen to each of those op-eds, starting with the piece by former board members Helen Toner and Tasha McCauley titled, AI Firms Mustn't Govern Themselves.

1:03Can private companies pushing forward the frontier of a revolutionary new technology be expected to operate in the interests of both their shareholders and the wider world? When we were recruited to the board of OpenAI, Tasha, in 2018 and Helen in 2021, we were cautiously optimistic that the company's innovative approach to self-governance could offer a blueprint for responsible AI development. But based on our experience, we believe that self-governance cannot reliably withstand the pressure of profit incentives. With AI's enormous potential for both positive and negative impact, it's not sufficient to assume that such incentives will always be aligned with the public good.

1:42For the rise of AI to benefit everyone, governments must begin building effective regulatory frameworks now. If any company could have successfully governed itself while safely and ethically developing advanced AI systems, it would have been OpenAI. The organization was originally established as a non-profit with a laudable mission, to ensure that AGI, or Artificial General Intelligence, AI systems that are generally smarter than humans, would benefit all of humanity. Later, a for-profit subsidiary was created to raise the necessary capital, but the non-profit stayed in charge. The stated purpose of this unusual structure was to protect the company's ability to stick to its original mission, and the board's mandate was to uphold that mission.

2:29It was unprecedented, but it seemed worth trying. Unfortunately, it didn't work. Last November, in an effort to salvage this self-regulatory structure, the OpenAI board dismissed its CEO, Sam Altman. The board's ability to uphold the company's mission had become increasingly constrained due to long-standing patterns of behavior exhibited by Mr. Altman, which, among other things, we believe undermined the board's oversight of key decisions and internal safety protocols. Multiple senior leaders had privately shared grave concerns with the board, saying they believed that Mr. Altman cultivated a toxic culture of lying and engaged in behavior that can be characterized as psychological abuse.

3:13According to OpenAI, an internal investigation found that the board had acted within its broad discretion to dismiss Mr. Altman, but also concluded that his conduct did not mandate removal. Open AI relayed few specifics justifying this conclusion, and it did not make the investigation report available to employees, the press, or the public. The question of whether such behavior should generally mandate removal of a CEO is a discussion for another time. But in Open AI's specific case, given the board's duty to provide independent oversight and protect the company's public interest mission, we stand by the board's action to dismiss Mr.

3:52Altman. We also feel that development since he returned to the company, including his reinstatement to the board and the departure of senior safety-focused talent, bowed ill for the open AI experiment in self-governance. Our particular story offers the broader lesson that society must not let the rollout of AI be controlled solely by private tech companies. Certainly, there are numerous genuine efforts in the private sector to guide the development of this technology responsibly, and we applaud those efforts. But even with the best of intentions, without external oversight, this kind of self-regulation will end up unenforceable, especially under the pressure of immense profit incentives.

4:32Governments must play an active role. And yet, in recent months, a rising chorus of voices, from Washington lawmakers to Silicon Valley investors, has advocated minimal government regulation of AI. Often they draw parallels with the laissez-faire approach to the internet in the 1990s and the economic growth it spurred. However, this analogy is misleading. Inside AI companies and throughout the larger community of researchers and engineers in the field, the high stakes and large risks of developing increasingly advanced AI are widely acknowledged. In Mr. Altman's own words, successfully transitioning to a world with superintelligence is perhaps the most important and hopeful and scary project in human history.

5:17The level of concern expressed by many top AI scientists about the technology they themselves are building is well documented and very different from the optimistic attitudes of the programmers and network engineers who developed the early internet. It is also far from clear that light touch regulation of the internet has been an unalloyed good for society. Certainly many successful tech businesses and their investors have benefited enormously from the lack of constraints on commerce online. It is less obvious that societies have struck the right balance when it comes to regulating to curb misinformation and disinformation on social media, child exploitation and human trafficking, and a growing youth mental health crisis.

6:01Goods, infrastructure, and society are improved by regulation. It's because of regulation that cars have seatbelts and airbags, that we don't worry about contaminated milk, and that buildings are constructed to be accessible to all. Judicious regulation could ensure the benefits of AI are realized responsibly and more broadly. A good place to start would be policies that give governments more visibility into how the cutting edge of AI is progressing, such as transparency requirements and incident tracking. Of course, there are pitfalls to regulation, and these must be managed. Poorly designed regulation can place a disproportionate burden on smaller companies, stifling competition and innovation.

6:45It is crucial that policymakers act independently of leading AI companies when developing new rules. They must be vigilant against loopholes, regulatory moats that shield early movers from competition, and the potential for regulatory capture. Indeed, Mr. Altman's own calls for AI regulation must be understood in the context of these pitfalls as having potentially self-serving ends. An appropriate regulatory framework will require agile adjustments, keeping pace with the world's expanding grasp of AI's capabilities. Ultimately, we believe in AI's potential to boost human productivity and well-being in ways never before seen.

7:26But the path to that better future is not without peril. OpenAI was founded as a bold experiment to develop increasingly capable AI while prioritizing the public good over profits. Our experience is that even with every advantage, self-governance mechanisms like those employed by OpenAI will not suffice. It is, therefore, essential that the public sector be closely involved in the development of the technology. Now is the time for governmental bodies around the world to assert themselves. Only through a healthy balance of market forces and prudent regulation can we reliably ensure that AI's evolution truly benefits all of humanity.

8:09All right, next up, we are going to listen to the response op-ed, this one by Brett Taylor and Larry Summers, titled by The Economist, OpenAI Board Members Respond to a Warning by Former Members. Helen Toner and Tasha McCauley, who left the board of OpenAI after its decision to reverse course on replacing Sam Altman, the CEO, last November, have offered comments on the regulation of artificial intelligence, AI, and events at OpenAI in a by-invitation piece in The Economist. We do not accept the claims made by Ms. Toner and Ms. McCauley regarding events at OpenAI. Upon being asked by the former board, including Ms.

8:46Toner and Ms. McCauley, To serve on the new board, the first step we took was to commission an external review of events leading up to Mr. Altman's forced resignation. We chaired a special committee set up by the board, and Wilmer Hale, a prestigious law firm, led the review. It conducted dozens of interviews with members of Opium AI's previous board, including Ms. Toner and Ms. McCauley. OpenAI executives, advisors to the previous board, and other pertinent witnesses reviewed more than 30 ,000 documents and evaluated various corporate actions. Both Ms. Toner and Ms. McCauley provided ample input to the review, and this was carefully considered as we came to our judgments.

9:21The review's findings rejected the idea that any kind of AI safety concern necessitated Mr. Altman's replacement. In fact, Wilmer Hale found that the prior board's decision did not arise out of concerns regarding product safety or security, the pace of development, OpenAI's finances, or its statements to investors, customers, or business partners. Furthermore, in six months of nearly daily contact with the company, we have found Mr. Altman highly forthcoming on all relevant issues and consistently collegial with his management team. We regret that Ms. Toner continues to revisit issues that were thoroughly examined by the WilmerHale-led review rather than moving forward.

9:55Ms. Toner has continued to make claims in the press. Although perhaps difficult to remember now, OpenAI released ChatGPT in November 2022 as a research project to learn more about how useful its models are in conversational settings. It was built on GPT 3.5, an existing AI model, which had already been available for more than eight months at the time. That said, we share Ms. Toner's and Ms. McCauley's view, and the company and Mr. Altman have continually stated that the evolution of AI represents a major development in human history. In democratic societies, accountability to government and government regulation is essential.

10:30OpenAI has held discussions with government officials around the world on numerous issues, and the company supports effective regulation of artificial general intelligence, or AGI, AI that would outperform humans in most intellectual tasks as it evolves. Mr. Altman has implored lawmakers to regulate AI. We think that regulatory intervention by governments will be critical to mitigate the risks of increasingly powerful models, he said in his opening remarks before a Senate Judiciary Subcommittee in May 2023. OpenAI and other leading AI labs have made a set of voluntary commitments to the White House to reinforce the safety, security, and trustworthiness of AI technology and services.

11:09Additionally, Mr. Altman sits on the U.S. Department of Homeland Security's Artificial Intelligence, Safety, and Security Board, which will advise the department, private sector stakeholders, and the public on how to safely develop and deploy AI technology across America, including in its critical infrastructure. We are looking forwards, not backwards. The board is being built out with the recent addition of three new members with extensive expertise in leading global organizations and navigating complex regulatory environments, and with backgrounds in technology, running nonprofits, and board governance.

11:38The new board has adopted important improvements to OpenAI's governance structure, so the board can more effectively oversee the firm's growth and fulfill OpenAI's mission of ensuring that AGI benefits all of humanity. These improvements include clearer governance guidelines and a stronger conflict-of-interest policy. As the company's products become more advanced, the board is taking commensurate steps to ensure safety and security. Just this week, it announced the formation of a new safety committee to make recommendations to the full board on matters pertaining to critical security and safety decisions for all OpenAI projects, and we are beefing up this new committee with people with national security expertise.

12:15Last week, at a major AI summit in Seoul, OpenAI and a number of other companies agreed to make additional AI safety commitments. They also pledged to publish safety frameworks like the Preparedness Framework adopted by OpenAI last year, which is designed to close gaps in the scientific study of catastrophic risks from AI. OpenAI, with the support of the board, is proud to build and release models that are industry-leading in terms of both capabilities and safety. The company will continue to do so. As we have stated previously, we recognize the magnitude of our role in stewarding transformative technologies for the global good.

12:51Hello, friends. Quick note before we get back to the show, I'm so excited to share that Superintelligent is now live. Superintelligent is a platform for fast, fun, and super practical, useful AI learning. We have something like 300 video tutorials adding 30 to 50 each week, covering every topic in AI you can imagine from LLMs to image generators to case studies, use cases, basically everything that tells you how to use AI and what to use it on in short, fast, four to seven minute tutorial videos, which are paired with step-by-step instructions that help you actually use these tools as well. It's$20 a month for unlimited access, and I would love to see you there.

13:29Check it out at bsuper.ai. That's bsuper.ai. All right, so now we're back. I don't necessarily need to break down each of those different pieces. They're fairly self-evident, and I think you can make up your own mind. What I will note is two big conversations that I am observing happening in the AI space. The first is specific to Sam Altman with the set of different, let's call them mini-controversies that have happened over the past few weeks, the Scarlett Johansson Sky voice dust-up, this op-ed and Helen Toner's appearance on the TED AI podcast, the questions of the non-disparagement clause as it relates to former employee equity.

14:07There is certainly a lot more scrutiny around Altman himself. I think this has to do with the fact that people appreciate that OpenAI and any company like it, at the very state-of-the-art of this space, has an extremely significant role to play in the world ahead. The burden on the people who are leaders in that situation is enormously high, higher even than other private companies that we've seen before. And so I expect that scrutiny to do nothing but grow. And frankly, I think both supporters and opponents alike should be glad for the scrutiny. I do believe it will make OpenAI a better company in the long run.

14:40Now, the other thing that I've been noticing is that there is absolutely a broader question around AI safety happening right now, too. We're more than a year out now from the six-month pause letter, and it feels to me like there's a little bit of impatience with the AI safety movement among normies who felt like everything was presented as immediately dire, and who now don't necessarily see some big disastrous issue as it seemed like they might. To some extent, it feels like there is a testing of the waters going on right now, where the mainstream labs are dividing themselves a little bit, and rather than all speaking the same language of AI safety, some are doubling down on those commitments, notably Anthropic, who has basically hired all of OpenAI's former super alignment people, whereas others, OpenAI most notably, seem to me to be testing the waters of a very different strategy.

15:21It's too early to see how this trend is going to play out right now, but it's something I'm watching closely. However, that is going to do it for today's AI Daily Brief. Until next time, peace.