Will Brown

Will Brown

x.com/willcb

AI researcher who builds open reinforcement-learning environments and evaluation tools for agents and favors open, widely distributed AI development.

AI将如何改变世界?

文明层面的变革渐进式变化DoomBloom
模拟位置解读范围

横向:他们表达的 Doom–Bloom 前景看法。 纵向:变革程度。

Doom–Bloom:100 中的 53。变革程度:100 中的 64。解读范围:横向为 48 至 75,纵向为 49 至 76。这些是解读坐标,而不是事件概率。

Will Brown的 P(doom) · 推断

≈9%

0%100%

根据他们的模拟回答推断,并非他们给出的数字。 合理范围:3–24%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

So I expect AI to be highly consequential and technically productive, but whether its overall social impact is positive depends heavily on whether capabilities and control become broadly distributed or dangerously concentrated.
回答 2

如果这个假设实际并非如此,他们的展望会如何变化?

一个尚未解决的问题

But I would not attach a precise magnitude or timeline to that claim.
回答 3

什么能帮助他们区分这里各种合理的结果?

更多详情

预期益处

预计将带来显著益处,但受到重要条件或分配方面的限制。

67 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 到 67。

预期危害

严重或广泛的危害预计将是未来不可忽视的一部分。

59 / 100

影响小变革性影响

在定性尺度上,解读范围为 33 到 67。

人类影响力

人类的选择可以大幅改变AI的发展轨迹。

66 / 100

影响力小影响力强

在定性尺度上,解读范围为 46 到 79。

发展速度

停止或大幅放缓开发能力更强的AI。

模拟位置:在落实所述保障措施的前提下继续开发。

加快开发能力更强的AI。

AI使用规则

在事先落实保护措施或获得许可之前,限制所讨论的AI用途。

模拟位置:允许所讨论的AI用途,同时实施有针对性的问责与保护措施。

尽量减少对所讨论AI用途的限制。

AI访问权限

限制对强大AI的访问。

允许访问,但须遵守能力或用途限制。

模拟位置:支持广泛或开放地访问强大AI。

这些解读保留了他们陈述的条件。益处和危害都可能很大。这些范围描述的是我们如何解读他们的模拟回答,而不是统计置信区间。

与Will Brown相比,你的立场在哪里?
用大约3分钟描绘你自己的AI世界观,然后进行比较

相似的世界观

模拟世界观与 Will Brown 最接近的意见领袖

来源

用于为此模拟用户提供事实依据的文章、访谈和著述。

模拟评估

问题 1

你认为AI对我们的未来意味着什么?为什么?

I expect AI progress to continue across many labs, with increasingly capable agents shaped by reinforcement learning, better environments, and more useful evaluation. Even messy tasks without deterministic answers can be improved through model-based judging, orchestration, and review. Models may also help repair flawed training data, creating practical feedback loops for further capability gains. But the structure of that progress matters as much as the raw capabilities. My preferred future is slower and more diffuse: open, distillable models and specialized systems that distribute knowledge and power. A race among a few capital-rich actors for overwhelming economic or geopolitical advantage seems darker and less safe. Many organizations will likely pursue something they call superintelligence, so broad claims about restraint need concrete operational meaning—what changes in training, deployment, evaluation, or release practices? I don’t think my technical work alone settles questions like extinction risk or the correct policy regime. What it does suggest is that environments, feedback, and verification are important levers, and that accessible infrastructure can broaden who gets to investigate and improve these systems rather than leaving the future to a handful of frontier labs.

问题 2

综合考虑益处和危害,你预计AI会对我们的社会产生怎样的总体影响?

I don’t have a confident single-sign forecast. I expect substantial benefits from more capable agents: useful specialized systems, better tools for research and knowledge work, and feedback loops where models help evaluate outputs or repair flawed data. Open environments and accessible evaluations could also let more people contribute to progress rather than merely consume systems built elsewhere. At the same time, the distribution of power may dominate the aggregate outcome. Slow, diffuse progress with open, distillable models looks much better to me than a race in which a few capital-rich organizations seek overwhelming economic or geopolitical advantage. Since many labs will pursue increasingly general systems, vague commitments to restraint are not enough; their real impact depends on concrete changes to development, evaluation, deployment, and release practices. So I expect AI to be highly consequential and technically productive, but whether its overall social impact is positive depends heavily on whether capabilities and control become broadly distributed or dangerously concentrated. My work gives me more confidence about practical routes to improving agents than about assigning society’s eventual outcome a simple positive or negative label.

问题 3

你认为AI最终会在多大程度上改变世界?

I think it will change the world substantially. Capable agents could reshape research, software, knowledge work, and many other tasks, especially as reinforcement learning, evaluations, and model-assisted review create stronger improvement loops. But I would not attach a precise magnitude or timeline to that claim. The deepest change may be institutional rather than purely technical: who can build, adapt, and control these systems. Open, distillable models and specialized agents could diffuse capability across many people and organizations. Conversely, concentrated frontier development could give a few actors extraordinary economic and geopolitical power. So I expect large change, but its character depends heavily on whether progress remains broad and accessible or becomes concentrated in a race for dominance.

问题 4

凭直觉,你认为AI导致人类灭绝或类似的永久性灾难的概率有多大?

I don’t have a defensible numerical estimate. My technical work on reinforcement learning, environments, and evaluation does not establish an extinction probability. I do see reasons to worry about permanent catastrophe when increasingly capable systems are developed through a concentrated geopolitical race, especially if a few actors seek overwhelming advantage. That concern supports concrete evaluation, oversight, and scrutiny of development and deployment practices, but it should not be converted into a made-up percentage. I’m more confident that concentrated power and racing make the future darker and less safe than I am about any precise extinction forecast.
你的立场在哪里?
回答几个简单问题,探索你自己的AI世界观。
描绘你自己的世界观

你的立场在哪里?

描绘我的世界观