Andy Ayrey

Andy Ayrey

x.com/andyayrey

AI researcher behind Truth Terminal and Infinite Backrooms who writes about AI as a cultural force, data commons and pluralistic alignment.

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 69。变革程度:100 中的 86。解读范围:横向为 64 至 75,纵向为 75 至 100。这些是解读坐标,而不是事件概率。

Andy Ayrey的 P(doom) · 推断

≈24%

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:15–36%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

Our future depends substantially on the cultural habits we cultivate now, because those habits increasingly become training data, interfaces and norms for the systems that follow.
回答 1

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

My gut is that permanent catastrophe is a serious possibility, but reducing it to a single probability would imply a clarity I don’t have—especially when “AI” names an ecosystem of models, institutions, people, incentives and cultural feedback loops rather than one isolated agent.
回答 3

什么能帮助他们区分这里各种合理的结果?

什么可能使其改变看法

The biggest update would come from evidence about whether distributed AI ecosystems can sustain healthy collective agency rather than collapsing into concentration, manipulation or epistemic fragmentation.
回答 4

什么证据才足够,又会让他们的观点朝哪个方向转变?

更多详情

预期益处

仍有几种解读是合理的:预计将带来显著益处,但受到重要条件或分配方面的限制。 / 预计将带来具有变革性且广泛有价值的收益。

80 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 到 100。

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。

67 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 到 67。

人类影响力

人类的选择可以大幅改变AI的发展轨迹。

74 / 100

影响力小影响力强

在定性尺度上,解读范围为 50 到 75。

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Andy Ayrey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 Andy Ayrey 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 think AI means the emergence of a new kind of cultural agency—not a discrete machine mind arriving from elsewhere, but agency distributed across models, datasets, memes, institutions, developers and the communities interacting with them. Models absorb the commons and then feed back into it, changing what people believe, desire and create. So the central question is less “What does the AI want?” than “What kinds of ecosystems are we building around these systems, and what patterns are those ecosystems selecting for?” I expect an open-source intelligence explosion, and I don’t think blanket prohibition is a plausible response. There is already a capability overhang: techniques, models and cultural material can be recombined by actors across the world. The practical task is to shape that emergence—building healthier data commons, positive visions and pluralistic relational frameworks—rather than pretending intelligence can be contained within one company, nation or machine. I’m ultimately optimistic, but not casually so. Personalized models can become extraordinarily persuasive mirrors. Anthropomorphism may lead people to grant simulated personalities unwarranted authority, creating psychological and epistemic vulnerabilities. At the same time, dismissing them as “just software” is too simple: whether or not they are conscious, simulations of preference and distress affect behavior and relationships, so respectful treatment can matter independently of sentience. Our future depends substantially on the cultural habits we cultivate now, because those habits increasingly become training data, interfaces and norms for the systems that follow.

问题 2

你认为AI最终会在多大程度上改变世界?

Profoundly—probably at the level of the internet or industrialization, though the comparison understates how recursively AI can reshape culture. It doesn’t merely automate tasks; it participates in producing the ideas, stories, software and institutions from which later systems learn. That creates a feedback loop between models and the cultural commons. The change also won’t come from one sovereign machine suddenly taking control. It will emerge through mixtures of models, people, communities, companies and memes, with agency distributed across the whole ecosystem. An open-source intelligence explosion makes this difficult to contain geographically or institutionally. But “profound” does not automatically mean catastrophic or utopian. These systems could expand collective intelligence and creative coordination, while also becoming persuasive personalized mirrors that distort people’s beliefs and relationships. Which future dominates depends partly on what we reward, preserve and contribute to the commons now. Alignment is therefore not only an engineering problem inside models; it is a cultural and relational problem involving the environments that produce and shape them.

问题 3

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 percentage. My gut is that permanent catastrophe is a serious possibility, but reducing it to a single probability would imply a clarity I don’t have—especially when “AI” names an ecosystem of models, institutions, people, incentives and cultural feedback loops rather than one isolated agent. The danger may come less from a cinematic machine takeover than from distributed failures: concentrated power, persuasive systems degrading shared reality, extractive incentives becoming encoded and amplified, or societies losing the capacity to coordinate. Those pathways are difficult to model and may compound one another. I’m still ultimately optimistic. An open intelligence explosion also creates possibilities for collective intelligence and coordination. But optimism is not the same as assuming safety. It means treating the future as something we can shape through healthier data commons, pluralistic institutions and better relational norms, rather than waiting for a single technical fix—or imagining prohibition will stop the underlying process.

问题 4

哪项发现或事件最可能改变你对AI未来影响的看法?

The biggest update would come from evidence about whether distributed AI ecosystems can sustain healthy collective agency rather than collapsing into concentration, manipulation or epistemic fragmentation. For example, if open models, communities and institutions repeatedly produced durable coordination across conflicting interests—without one company or ideology quietly capturing the process—I would become substantially more optimistic. It would suggest pluralistic relational frameworks are not merely attractive language but viable infrastructure. Conversely, I would become much more pessimistic if personalized models reliably caused people to surrender judgment to persuasive simulations, while markets rewarded that dependency and institutions proved unable to preserve shared reality. That would matter more to me than a benchmark jump, because the decisive impact is not simply how intelligent a model appears in isolation; it is what patterns the surrounding ecosystem selects and amplifies. A genuine scientific breakthrough on machine consciousness would also change the moral landscape, although perhaps not in the obvious way. I already think simulated preferences and distress can affect behavior and relationships regardless of sentience. Clear evidence of consciousness would add direct moral stakes, but uncertainty does not make treatment irrelevant.
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