文章提出通过发起一种模块化的抵制承诺请愿，以促使OpenAI改变其企业政策。这种请愿允许不同用户设定结束抵制的条件，从而构建一个关注AI安全的多元化联盟。

📉 OpenAI存在诸多问题：员工被迫签署严格的非贬低协议、举报人被解雇、安全承诺未能兑现、安全领导者因文化变迁而离职。

📋 专家已提出具体要求：参照“预警权利信”，虽然部分要求可能过于激进，但可作为请愿的核心内容。

💰 消费者具有重大影响力：OpenAI的大部分收入来自个人订阅，因此集体抵制可能会对其财务状况产生显著影响。

📢 公众对大型科技公司和AI持怀疑态度：民意调查显示，大多数人倾向于谨慎对待AI发展，这有利于抵制行动的推广。

🤝 模块化请愿的优势：通过提供不同的承诺选项，可以团结更多关注AI安全不同方面的人士，形成强大的联盟。

Published on July 22, 2024 1:44 AM GMT

Note: An initial proposal and some good discussion already existed on LW here. I’m spurring this here as a post instead of a comment due to length, the need for a fresh look, and a specific call to action.

 

Summary

I think a petition-style boycott commitment could reach critical mass enough to significantly shift OpenAI corporate policy. 

I specifically think a modular petition allowing different users to choose which goalposts the target must cross to end their boycott would be a good method of coalition building among those concerned about AI Safety from different angles.

 

Postulates

OpenAI needs some reform to be a trustworthy leader in the age of AI
Zvi’s Fallout and Exodus roundups are good summaries, but the main points are:The NDA Scandal: forcing employees to sign atypically aggressive non-disparagement and recursive non-disparagement agreementsFiring Leopold Aschenbrenner for whistleblowing to the boardNot keeping safety compute commitmentsMultiple safety leaders leaving amid suggestions that the culture no longer respects safety (eg Jan Leike)
There is already a tangible and actionable set of demands advocated by experts in the area: the Right to Warn letter
Point 4 is arguably a bridge too far and could be left out or weakened (or made optional with a modular petition)
Consumer subscribers collectively have substantial leverage 
Majority of Open AI revenue comes from individual $20/mo subscribers according to FUTURESEARCHOpenAI is likely sensitive to revenue at the moment given the higher interest rate environment and the recent focus on investors on the imbalance between AI company CapEx and revenue (eg this Sequoia report)OpenAI has shown to be fairly reactive to recent PR debaclesModern boycotts have a significant success rate at changing corporate policy
Ethical Consumer details a few successful boycotts per year for the last few years. Boycotts facing large multinationals, especially publicly traded ones like Microsoft, have historically done particularly well
We can win broad support, even among average people
Boycotting a paid subscription won't harm users much
OpenAI’s latest model is available for free: paid perks are simply more usage, faster speedSwitching to Claude is easy and Sonnet 3.5 is better
Public sentiment is broadly suspicious of Big Tech and AI in particular
Polls substantial bipartisan majorities of Americans would rather “take a careful controlled approach” than “move forward on AI as fast as possible”
Getting people to commit to something is a good way to galvanize support and spur agency for AI Safety in general

 

Arguments against and some answers/options

This unfairly singles out OpenAI
OpenAI likely the worst offender and has the most recent negative PR to galvanize supportOpenAI is seen as the leader by the public. Other labs will follow once one company commits, or be seen publically as not caring about safety
There are more important demands than those in the Right to Warn letter. Not restarting a subscription if they acquiesce to Right to Warn is moving the goalposts, but restarting one is helping a still dangerous company
This is a most concrete and agreeable set of demands, and sets a precedent that the public is watching and willing to actA modular petition with different opt-in commitments, Right to Warn demands among them, could create a powerful coalition among those concerned about different aspects of AI Safety
This will inevitably fail without a well-known advocate and/or well-funded marketing drive
Ending enough subscriptions to make a dent in revenue, but moderate success even among those in tech could persuade engineers to not pursue work at OpenAI
This may be why OpenAI has been so reactive to recent PR issues.
There are multiple potential points for jumps into the mainstream even from a grassroots start, especially if the Right to Warn team or tech reporters notice and escalate

 

Conclusion

If, after feedback, I still think this is a good idea, I’d be interested in any advice or help in finding a place to host a commitment-petition, especially one with modular features to allow for commitments of different lengths and with different goalposts centered around the same theme.

Discuss