Could AI help communities navigate complex challenges like the US data center boom?
As the artificial intelligence (AI) boom brings large data center proposals to small and rural communities across the United States and beyond, the most resounding story to emerge is one of mismatched agency.
随着人工智能（AI）的繁荣给美国及其他地区的小型和农村社区带来了大型数据中心的建议，出现的最响亮的故事是机构不匹配的故事之一。
In journalist Jasmine Sun’s words , “big companies descend on a small town and run roughshod over small-d democracy.”
用记者孙茉莉的话来说，“大公司降临小镇，践踏小民主”。
Many people do not want data centers in their communities.
许多人不希望自己的社区中有数据中心。
And for good reasons: the costs are material, if still not fully understood, spanning noise and visual pollution, land-use change, and rising electricity demand with knock-on effects for water withdrawals and utility affordability .
这是有充分理由的：即使尚未完全理解，成本也是巨大的，涵盖噪音和视觉污染、土地利用变化以及不断增长的电力需求，对取水和公用事业负担能力产生连锁反应。
As with most large-scale developments, data centers also offer economic upsides —a surge of construction jobs , considerable tax revenue , and improvements in infrastructure.
与大多数大规模开发一样，数据中心也能带来经济上的好处——建筑工作岗位的激增、可观的税收收入和基础设施的改善。
These can be significant but are not always guaranteed .
这些可能很重要，但并不总是有保证。
Given the scale and impact of data centers, people should have a voice.
考虑到数据中心的规模和影响，人们应该有发言权。
However, decisionmaking about data center siting is often fragmented , noninclusive , and based on criteria other than their impact on communities.
然而，有关数据中心选址的决策通常是分散的、非包容性的，并且基于的标准而不是对社区的影响。
In many parts of the country, s tates set tax exemptions and approve utility contracts, while local officials, often working part-time, negotiate one siting approval at a time, up against experienced corporate counsel imposing time pressure and confidentiality agreements.
在该国许多地区，各州制定免税政策并批准公用事业合同，而地方官员通常是兼职工作，一次只能协商一项选址批准，而经验丰富的公司法律顾问则施加时间压力和保密协议。
Local citizens are rarely directly involved in these processes and, where confidentiality provisions apply, may learn the terms of a proposed deal only after negotiations are well advanced.
当地居民很少直接参与这些过程，并且在适用保密条款的情况下，只有在谈判进展顺利后才能了解拟议交易的条款。
The result is that residents feel sidelined, fueling skepticism as to whether data center proponents are sharing all relevant information and whether public officials are acting in the community’s best interests.
结果是，居民感到被边缘化，加剧了人们对数据中心支持者是否共享所有相关信息以及公职人员是否以社区最佳利益为出发点的怀疑。
Because data center developments are closely associated with the rapid diffusion of generative AI in the U.S. and elsewhere, uncertainty about the projects themselves can also become entangled with broader concerns about AI’s economic and social impacts , as well as its existential risks.
由于数据中心的发展与生成式人工智能在美国和其他地方的快速扩散密切相关，因此项目本身的不确定性也可能与对人工智能的经济和社会影响及其存在风险的更广泛担忧纠缠在一起。
Known remedies for these challenges exist—namely, community-led processes that surface varied priorities, align on a common vision, and help turn it into action.
针对这些挑战的已知补救措施是存在的，即由社区主导的流程，这些流程可以呈现不同的优先事项，与共同的愿景保持一致，并帮助将其转化为行动。
Historically, such processes have proven costly and complex, requiring time to build the trusting relationships on which collaboration depends—time that can be particularly scarce when communities are responding to fast-moving development proposals.
从历史上看，此类流程已被证明成本高昂且复杂，需要时间来建立协作所依赖的信任关系——当社区响应快速变化的开发提案时，时间可能特别稀缺。
Deliberative processes become more difficult to run when issues are emotionally and politically charged, as is increasingly the case with data center developments in the U.S.
当问题带有情感和政治色彩时，审议过程就会变得更加难以进行，美国数据中心开发的情况越来越如此。
Given this context, we undertook to explore whether the same technological advances driving the need for large data centers—generative AI—could be leveraged as a community resource for problem-solving.
在这种背景下，我们着手探索是否可以利用推动大型数据中心需求的相同技术进步（生成式人工智能）作为解决问题的社区资源。
We began with a simple question: What if communities could use AI to help reach better collective decisions about data center developments and other similarly complex local challenges?
我们从一个简单的问题开始：如果社区可以使用人工智能来帮助就数据中心开发和其他类似复杂的当地挑战达成更好的集体决策，会怎么样？
Those worried about AI’s impacts, or doubtful of its usefulness to society, may be skeptical.
那些担心人工智能影响或怀疑其对社会有用性的人可能会持怀疑态度。
Our proposal may sound like a contradiction: If AI is seemingly—in contexts like the U.S. at least—being designed and deployed to replace people in work, how can it enhance people’s shared agency?
我们的建议听起来可能有些矛盾：如果人工智能（至少在美国这样的背景下）似乎是为了取代人们的工作而设计和部署的，那么它如何增强人们的共享机构呢？
We argue that it can.
我们认为可以。
When designed to support, not replace, human-to-human collaboration , and deployed in ways that communities own, govern, and control , generative AI shows considerable potential for enhancing shared problem-solving and collective intelligence .
当生成式人工智能旨在支持而不是取代人与人之间的协作，并以社区拥有、治理和控制的方式进行部署时，它在增强共享问题解决和集体智慧方面显示出巨大的潜力。
AI can collapse the time and costs that have traditionally kept broad, sustained participation in such problem-solving out of reach.
人工智能可以消除传统上使广泛、持续参与此类问题解决遥不可及的时间和成本。
An AI facilitator, unlike a human, has no problem with everyone talking at the same time.
与人类不同，人工智能辅导员对于每个人同时讲话没有任何问题。
The key question is how to design AI systems that support more inclusive decisionmaking and that communities will adopt and trust.
关键问题是如何设计支持更具包容性的决策并让社区采用和信任的人工智能系统。
For communities to answer an AI buildout with AI applications of their own may sound odd given the warranted skepticism of AI’s usefulness and the fact that the technology has not been designed or deployed on community terms.
考虑到对人工智能有用性的怀疑以及该技术尚未按照社区条款设计或部署的事实，社区用自己的人工智能应用程序来应对人工智能扩建可能听起来很奇怪。
But the broader proposition of AI’s value to communities extends beyond data centers and the U.S. institutional context in which this work is situated.
但人工智能对社区的更广泛的价值主张超出了数据中心和这项工作所处的美国机构背景。
Communities across different national and development contexts regularly face complex decisions that cut across policy domains, institutions, and stakeholder interests while lacking the capacity to integrate what people know and value into a shared course of action.
不同国家和发展背景的社区经常面临复杂的决策，这些决策跨越政策领域、机构和利益相关者的利益，同时缺乏将人们所知道和重视的内容整合到共同行动方针中的能力。
Managing the data center boom is a clear and relevant example of this type of community decisionmaking.
管理数据中心的繁荣是此类社区决策的一个明显且相关的例子。
Our motivating assumption was that c ommunities can reach better decisions and better address shared challenges by transforming a complex issue (like whether to build a data center) from a discrete decision to an iterative process of community problem-solving that includes the perspectives, interests, and expertise of all stakeholders, not just elites and elected officials.
我们的激励性假设是，通过将复杂问题（例如是否建立数据中心）从离散决策转变为社区问题解决的迭代过程，其中包括所有利益相关者（而不仅仅是精英和民选官员）的观点、利益和专业知识，社区可以做出更好的决策并更好地解决共同的挑战。
Informed by existing literature, community problem-solving can be understood as including four phases that we summarize here as surface, align, implement, and learn, or “SAIL” (see Figure 1).
根据现有文献，社区问题解决可以理解为包括四个阶段，我们在这里总结为表面、调整、实施和学习，或“SAIL”（见图 1）。
The cycle synthesizes standard loops in adjacent literatures, including the policy cycle’s stages of agenda-setting, formulation, implementation, and evaluation; the plan-do-study-act loop from improvement science; the observe-orient-decide-act loop from strategy science; and the plan-act-monitor-adjust loop from adaptive ecological management.
该周期综合了相关文献中的标准循环，包括政策周期的议程设置、制定、实施和评估阶段；改进科学中的计划-实施-学习-行动循环；战略科学中的观察-导向-决定-行动循环；以及适应性生态管理的计划-行动-监测-调整循环。
Where these cycles typically begin from observation or agenda-setting by a single actor or organization, these phases emphasize the need for surfacing values and perspectives across many stakeholders and issues and aligning them before decision and action across multiple domains.
这些周期通常从单个参与者或组织的观察或议程设置开始，而这些阶段强调需要在许多利益相关者和问题上表达价值观和观点，并在跨多个领域的决策和行动之前将它们协调一致。
SAIL is aligned with decades of behavioral science research on group dynamics and collective intelligence .
SAIL 与数十年来关于群体动态和集体智慧的行为科学研究相一致。
This research finds that groups that succeed in coordinating do so through establishing shared mental models and transposing them into technologies for shared action like maps and stories .
这项研究发现，成功协调的群体是通过建立共享的心理模型并将其转化为地图和故事等共享行动的技术来实现这一目标的。
In these settings, high – performing groups navigate a consensus-diversity tradeoff.
在这些环境中，高绩效团队需要在共识与多样性之间进行权衡。
Groups need consensus around shared goals to coordinate action .
团体需要围绕共同目标达成共识以协调行动。
Yet groups solve problems better when they draw on more diverse perspectives .
然而，当团队利用更多样化的观点时，他们就能更好地解决问题。
Generating a shared vision and developing affordances for shared action is commonplace in small teams and organizations.
在小型团队和组织中，产生共同愿景并为共同行动提供支持是很常见的。
However, doing so within larger and more diffuse communities can be prohibitively expensive , time-consuming, and often contentious .
然而，在更大、更分散的社区内这样做可能会非常昂贵、耗时，而且常常引起争议。
A long line of scholarship, from Ronald Coase on transaction costs to Elinor Ostrom on how communities govern shared resources , holds that gains that can arise from community-wide coordination go unrealized because the costs of establishing and sustaining coordination are prohibitive.
从罗纳德·科斯（Ronald Coase）关于交易成本的问题到埃莉诺·奥斯特罗姆（Elinor Ostrom）关于社区如何管理共享资源的一系列学术研究都认为，社区范围内的协调可能带来的收益无法实现，因为建立和维持协调的成本过高。
Gathering what people know and want, mapping how issues interact, and keeping agreements on track take time, staff, and skills most public institutions cannot support.
收集人们所知道和想要的信息，绘制问题如何相互作用的图谱，并使协议保持在正轨上，需要时间、人员和技能，而大多数公共机构都无法支持。
Even institutions with capacity for cross-issue convening, such as community foundations, find it difficult to systematically connect across issues.
即使是具有跨问题召集能力的机构，例如社区基金会，也发现很难系统地连接跨问题。
As a result, community problem-solving often narrows and focuses on single-issue initiatives or yes-or-no ballot measures.
因此，社区解决问题的范围往往会缩小，并集中于单一问题的倡议或是或否的投票措施。
Effective community problem-solving processes of surfacing, aligning, implementing, and learning are therefore defined by the challenge of managing two demands at once: maximally including participant perspectives while advancing a minimally complex set of shared actions.
因此，有效的社区问题解决过程的浮现、协调、实施和学习是由同时管理两个需求的挑战所定义的：最大限度地纳入参与者的观点，同时推进一组最简单的复杂的共享行动。
A township weighing a data center proposal, for example, must balance concerns about economic development, environmental sustainability, and social equity all at once.
例如，一个正在权衡数据中心提案的乡镇必须同时平衡对经济发展、环境可持续性和社会公平的担忧。
It must then carry what residents value through to implementation.
然后，它必须将居民所重视的内容付诸实施。
And, hardest of all, it must neutrally evaluate whether its actions are working as planned and how they can be improved.
而且，最困难的是，它必须中立地评估其行动是否按计划发挥作用以及如何改进。
Generative AI has the capacity to increase the effectiveness and lower the costs of each core phase of community problem-solving.
生成式人工智能有能力提高社区问题解决的每个核心阶段的效率并降低成本。
AI can transcribe and summarize large volumes of input in seconds rather than weeks.
人工智能可以在几秒钟而不是几周内转录和总结大量输入。
It can help synthesize thousands of perspectives—and incorporate more of the nuance of each perspective—into a single navigable picture.
它可以帮助合成数千个视角，并将每个视角的更多细微差别融入到一个可导航的图片中。
AI can map how proposals interact across issues, making integration across stakeholders more tractable.
人工智能可以绘制提案如何跨问题交互的方式，使利益相关者之间的整合更容易处理。
It can also quickly judge ideas for feasibility and affordability as well as check facts.
它还可以快速判断想法的可行性和承受能力以及检查事实。
No one questions whether a plane can transport someone across the country faster than a horse can.
没有人质疑飞机能否比马更快地运送人穿越全国。
The plane is much faster.
飞机速度快得多。
A similar comparison holds for AI versus humans in aggregating information.
人工智能与人类在聚合信息方面也存在类似的比较。
AI is faster.
人工智能更快。
Early applications of AI to community problem-solving demonstrate this potential.
人工智能在解决社区问题方面的早期应用证明了这种潜力。
A preliminary landscape analysis of existing tools includes examples such as Taiwan’s vTaiwan process (which turned citizen input into consensus statements that shaped legislation), U.N.
对现有工具的初步景观分析包括台湾的 vTaiwan 进程（将公民的意见转化为形成立法的共识声明）、联合国
consultations in conflict settings, citizens’ assemblies , and large-scale listening .
冲突环境下的协商、公民集会和大规模倾听。
Several of these tools rely on established machine learning methods such as clustering and dimension reduction; generative AI extends them by working end-to-end in natural language and by letting nontechnical users produce new software artifacts to support collaboration and decisionmaking.
其中一些工具依赖于现有的机器学习方法，例如聚类和降维；生成式人工智能通过以自然语言进行端到端工作并让非技术用户生成新的软件工件来支持协作和决策来扩展它们。