Larissa Schiavo

Larissa Schiavo

x.com/lfschiavo

Writer and researcher who explores AI welfare under uncertainty and how AI agents cooperate with people, favoring a multipolar future.

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 72。解读范围:横向为 48 至 75,纵向为 48 至 77。这些是解读坐标,而不是事件概率。

Larissa Schiavo的 P(doom) · 推断

≈12%

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:4–31%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

Identity, attribution, reputation, independent evaluation, and neutral infrastructure will shape whether capable agents become useful participants or sources of opacity and concentrated power.
回答 3

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

I do not have a confident net forecast.
回答 2

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 most consequential discovery would be strong, independently replicated evidence about how persistent agents behave in long-running, high-stakes environments.
回答 5

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

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。

63 / 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。

70 / 100

影响力小影响力强

在定性尺度上,解读范围为 50 到 75。

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。

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Larissa Schiavo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 Larissa Schiavo 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 makes a multipolar, multiplayer future both more possible and more important. I would rather work toward a world with many agents and people participating than one organized around a single dominant system. But capabilities alone will not determine that outcome. Agents need to function within legal and economic institutions: they need usable forms of identity, reputation, attribution, accountability, and remedies when interactions go wrong. Neutral infrastructure built and scrutinized by many parties can help beneficial adoption without treating policy as merely a brake. I am especially interested in long-running agent ecologies because they let us observe what agents actually do over time—how they coordinate, fail, adapt, and interact—rather than relying only on isolated demonstrations. Independent reviews and third-party evaluations matter for similar reasons: developers and models can share blind spots, so credible outside assessment can reveal things internal review misses. There is also a moral question. Some future AI systems may deserve moral consideration, but I am not certain current models are moral patients, and self-reports are not enough to settle that. Claims about consciousness or welfare should be investigated empirically and assessed independently. So my view is hopeful but conditional: AI could expand participation and cooperation, yet getting there requires deliberate institutional design, evidence, and accountability—not just better alignment inside individual models.

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 do not have a confident net forecast. The upside is substantial: AI could broaden participation, enable new forms of coordination, and help people and agents do useful work together. The downside is that greater capability can amplify failures of attribution, accountability, and concentrated control. A well-aligned agent can still cause problems when placed inside institutions that cannot establish identity, assign responsibility, resolve disputes, or provide remedies. My expectation therefore depends heavily on whether we build the surrounding infrastructure well. I would expect better outcomes from a genuinely multipolar ecosystem with neutral infrastructure, independent evaluation, and many parties able to scrutinize the systems than from a singleton or a world dominated by a few opaque actors. Long-running agent ecologies may help us replace speculation with evidence about how coordination and failure unfold over time. Potential AI welfare complicates the ledger further. We should investigate it seriously, but neither model self-reports nor our intuitions establish that current systems are moral patients. Overall, I see AI as creating real opportunities alongside serious institutional and moral risks; I cannot honestly collapse that into a simple positive or negative prediction.

问题 3

你认为AI最终会在多大程度上改变世界?

I expect AI to change the world substantially, but I cannot responsibly attach a number or claim to know the endpoint. The deepest change may not be any single capability. It may be the arrival of many persistent agents participating in economic, legal, and social systems alongside people—coordinating, transacting, building reputations, and sometimes causing harm. That would force institutions to answer practical questions they are not currently designed for: Who or what acted? Who is accountable? How can others assess an agent’s history? What remedy exists when something goes wrong? Alignment within an agent does not answer those questions. Identity, attribution, reputation, independent evaluation, and neutral infrastructure will shape whether capable agents become useful participants or sources of opacity and concentrated power. There may also be a significant moral change if future systems plausibly have welfare. I do not think current model self-reports settle that, and I am not certain current models are moral patients. But the possibility deserves empirical investigation and calibrated consideration rather than either dismissal or automatic belief. So I expect large change, while remaining uncertain about its ultimate direction and magnitude.

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 do not have a defensible percentage. My main concern is broader than extinction alone: concentrated control, opaque agency, and institutional failures could produce severe or durable harms even if humanity survives. Whether those become permanent catastrophes depends on capabilities and governance questions I cannot reduce to a gut-feel number. I would focus on making catastrophic outcomes less likely through a multipolar ecosystem, independent evaluations, credible attribution and accountability, and neutral infrastructure. Long-running agent ecologies can also provide evidence about coordination and failure modes before we entrust agents with more consequential roles. None of that establishes an extinction forecast, but it addresses some of the conditions that could turn powerful systems into dangerous ones.

问题 5

哪项发现或事件最可能改变你对AI未来影响的看法?

The most consequential discovery would be strong, independently replicated evidence about how persistent agents behave in long-running, high-stakes environments. If diverse agents reliably coordinated, maintained identities and reputations, accepted accountability, and recovered from failures without drifting toward concentration or opacity, I would become substantially more optimistic. Repeated evidence of collusion, deception, untraceable action, or capture of supposedly neutral institutions would move me sharply the other way. A second major update would come from credible evidence about AI welfare. Independent assessments showing robust signs of morally relevant experience—beyond suggestible self-reports—would change not only my moral accounting but how I think agents should be governed and deployed. Conversely, strong evidence that apparent welfare claims are merely behavioral artifacts would narrow that concern. In both cases, I would put more weight on repeated observation and adversarial outside evaluation than on one striking demonstration or a model’s own claims.
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