「英中对照」行业观点：楼宇自控需划定AI数据采信与动作执行的治理边界

Buildings have spent decades learning how to talk.建筑花了数十年时间学习如何“说话”。

Sensors measure temperature, humidity, carbon dioxide, pressure, particulate matter, airflow, electrical consumption and equipment state. Building automation systems collect those measurements. Dashboards visualize them. Analytics compare them. Digital twins represent them.传感器测量温度、湿度、二氧化碳、压力、颗粒物、气流、用电量和设备状态。楼宇自动化系统收集这些测量数据。仪表板将其可视化。分析系统对其进行比较。数字孪生对其进行呈现。

Now artificial intelligence is arriving with the ability to read enormous portions of that building history, recognize patterns, diagnose conditions, recommend changes and, increasingly, initiate actions.如今，人工智能的到来使其能够读取海量楼宇历史记录、识别模式、诊断状况、建议变更，并越来越多地自主发起操作。

That creates an architectural question I don’t think our industry can avoid much longer.这带来了一个我认为行业无法再回避太久的架构性问题。

What evidence should AI be permitted to rely upon — and what should AI be permitted to cause because of it?应允许人工智能依赖哪些证据——又应允许它基于这些证据造成什么结果？

Those are two different questions.这是两个不同的问题。

And there is a consequence boundary on both sides of intelligence.而且在智能的两侧都存在一条后果边界。

THE RECORD IS NOT THE DIAGNOSIS记录不等于诊断

Let’s start with an air-handling unit.先从空气处理机组说起。

Imagine we have several years of supply-air temperature, return-air temperature, static pressure, airflow, fan speed, valve position, outdoor-air conditions, electrical consumption and zone environmental measurements.设想我们有数年的送风温度、回风温度、静压、气流、风机转速、阀位、室外空气状况、用电量和区域环境测量数据。

Those measurements describe things that happened.这些测量数据描述了发生过的事情。

They do not automatically tell us why they happened.它们不会自动告诉我们事情为何发生。

This distinction matters:这一区分至关重要：

The record is not the diagnosis.记录不等于诊断。

The diagnosis is not the optimization.诊断不等于优化。

The optimization is not authority.优化不等于授权。

If we collapse those layers, we create a dangerous feedback loop.如果将这些层级混为一谈，就会产生危险的反馈回路。

A system observes something, interprets its own observation, changes the building based on that interpretation, and then potentially uses the resulting state as evidence that its original interpretation was correct.系统观察到某种现象，解释自己的观察，基于该解释改变楼宇，随后还可能将由此产生的状态用作证据，证明其最初的解释是正确的。

The original record can disappear inside the intelligence acting upon it.原始记录可能消失在对其采取行动的智能内部。

I believe we need to preserve the record first.我认为我们必须首先保存记录。

ATMOSPHERIC INTEGRITY RECORDS大气完整性记录

This is one reason I developed Atmospheric Integrity Records, or AIR.这正是我开发大气完整性记录（Atmospheric Integrity Records，简称 AIR）的原因之一。

An AIR is intended to preserve environmental reality independently from the later diagnosis, optimization or intervention.AIR 旨在独立于后续的诊断、优化或干预，保存环境现实。

That means preserving more than a number on a dashboard.这意味着要保存的不只是仪表板上的一个数字。

What was observed?观察到了什么？

Which sensor or system produced it?由哪个传感器或系统产生？

Where?在何处？

When?在何时？

What equipment or space did it represent?它代表什么设备或空间？

What happened before and after it?其前后发生了什么？

Did the sensor change?传感器是否发生过变更？

Did calibration change?校准是否发生过变化？

Did the BAS configuration change?BAS 配置是否发生过变更？

Were there gaps?是否存在数据缺口？

Were transformations applied?是否应用过数据转换？

What limitations traveled with the record?记录附带了哪些限制？

The point isn’t to make every measurement magically trustworthy.关键并不是让每个测量数据奇迹般地变得可信。

It is to preserve enough integrity around the record that we can later examine what that record can legitimately support.而是围绕记录保存足够的完整性，使我们日后能够审视该记录能够合理支撑什么。

That creates a progression:这形成了一个递进过程：

REALITY → RECORD → CONTINUITY → ADMISSIBILITY现实 → 记录 → 连续性 → 可采性

Now something interesting becomes possible.现在，某种有趣的事情成为可能。

THE BUILDING CAN SHOW US DRIFT楼宇能向我们展示漂移

Mechanical equipment rarely goes from perfect operation to catastrophic failure in one instant.机械设备很少会在一瞬间从完美运行变为灾难性故障。

It drifts.它是逐渐漂移的。

A fan requires slightly more power.风机需要略高的功率。

A pressure relationship changes.压力关系发生变化。

Airflow begins declining.气流开始下降。

A valve operates farther open to achieve the same result.阀门开度变得更大，才能达到相同效果。

Temperature recovery becomes slower.温度恢复变慢。

Environmental performance changes gradually.环境性能逐渐变化。

One measurement may not mean very much.单次测量可能说明不了什么。

But an attributable, continuous record spanning months or years can tell a much richer story.但一份可持续数月或数年、可溯源的连续记录能够讲述丰富得多的情况。

Imagine an AHU that has not generated an alarm.设想一台未产生过报警的 AHU。

Occupants aren’t complaining yet.占用者尚未投诉。

Nothing has “failed.”没有任何东西“故障”。

But twelve months of preserved operating history shows that under comparable conditions, fan power has gradually increased while delivered airflow has decreased.但保存下来的十二个月运行历史显示，在可比条件下，风机功率逐渐上升，而输送风量却在下降。

That isn’t automatically a diagnosis.这并不能自动构成诊断。

But it is evidence of change.但这是变革的明证。

Now AI becomes extremely interesting.如今，人工智能变得极其有趣。

An intelligent system can examine that longitudinal record and identify relationships that a technician, operator or engineer might not reasonably find by manually reviewing thousands or millions of individual observations.智能系统可以检查那些纵向记录，识别出技术人员、操作员或工程师在人工审阅数千乃至数百万条独立观测数据时不太可能发现的关联。

AI may be able to identify mechanical or environmental drift long before the condition becomes an emergency.人工智能或许能在状况演变为紧急事件很久之前，就识别出机械或环境偏移。

That’s valuable.这很有价值。

But before we celebrate, there is another question.但在我们庆贺之前，还有另一个问题。

WHAT ARE WE GIVING THE AI?我们在给人工智能提供什么？

We tend to think of feeding data to AI as if nothing consequential has happened yet.我们往往认为向人工智能输入数据时，似乎还没有发生什么要紧的事。

I don’t think that’s correct.我认为这种看法并不正确。

Giving information to an intelligent system is itself a consequence.将信息提供给智能系统本身就是一种后果。

We are allowing a particular computational actor to receive particular information, for a particular purpose, within a particular context.我们是在允许某个特定的计算行为体，在特定的情境下、为特定的目的接收特定的信息。

So before the AI gets the record, I want to ask:因此，在人工智能获取记录之前，我想问：

Is this AI permitted to rely upon this evidence for this purpose now?是否允许该人工智能此刻为此目的依赖此证据？

That means the proposed consequence can be stated explicitly:这意味着拟议的后果可以被明确表述为：

Release this bounded environmental and mechanical record to this identified intelligence for this identified purpose.将此有界的环境与机械记录发布给该已识别的智能体，用于该已识别的目的。

Now we can govern it.现在我们就可以对其进行管控。

Is the evidence sufficiently attributable?证据是否具有充分的可归因性？

Is continuity established?连续性是否已确立？

Is the record current enough for the proposed reliance?对于拟议的依赖而言，记录是否足够新？

Are material gaps known?是否已知晓重大缺口？

Does the authority exist to release or use it?是否存在发布或使用它的权限？

Does this system have standing to receive and act computationally upon it?该系统是否具备接收它并对其进行计算处理的资格？

If those things are not established, the fact that an API connection exists doesn’t answer the question.如果这些事项均未确立，那么存在 API 连接这一事实并不能回答这个问题。

Capability is not authority.能力不等于权限。

The AI should not get information simply because technically we can send it.人工智能不应仅仅因为技术上我们能够发送信息，就获取该信息。

NOW LET THE AI THINK现在让人工智能思考

Suppose the record survives that boundary.假设记录通过了那道边界。

The AI receives the governed evidence.人工智能接收到经过管控的证据。

Now let it do what AI can do extraordinarily well.现在让它去做人工智能格外擅长的事。

Compare.比较。

Interpret.解读。

Detect patterns.检测模式。

Model.建模。

Diagnose.诊断。

Predict.预测。

Optimize.优化。

Recommend.建议。

Perhaps it examines eighteen months of AHU history and determines that the relationship between fan power, static pressure and delivered airflow has materially changed.也许它检查了十八个月的 AHU 历史数据，判定风机功率、静压与输送风量之间的关系已发生实质性变化。

It proposes:它提议：

Increase AHU-1 fan speed from 62% to 66%.将 AHU-1 风机转速从 62% 提高至 66%。

Now we have another boundary.现在我们又遇到了另一道边界。

And the fact that we governed the AI’s input does not authorize its output.而我们对人工智能的输入进行了管控这一事实，并不意味着其输出就获得了授权。

This is where I believe a major architectural mistake could occur in autonomous buildings.我认为，在自主建筑中，一个重大的架构错误可能就发生在这里。

We could assume:我们可能会假设：

Good data in → intelligent analysis → authorized action out.优质数据输入 → 智能分析 → 授权动作输出。

That equation doesn’t hold.这个等式并不成立。

An admissible input does not create an admissible output.可采纳的输入并不会产生可采纳的输出。

ASK THE QUESTION AGAIN再次提出这个问题

The AI has produced a new proposed consequence.人工智能已经产生了一个新的拟议后果。

So TA-14 asks its governing question again:因此 TA-14 再次提出其管控问题：

Does this proposed consequence have sufficient Admissible Evidence, Applicable Authority, and Established Standing to become reality NOW?该拟议后果是否具备充分的可采纳证据、适用权限和已确立资格，从而在当下成为现实？

Not five minutes ago.不是五分钟前。

Not when the model was trained.不是模型训练之时。

Not when the building was commissioned.不是建筑调试启用之时。

Not because the AI is usually correct.也不是因为人工智能通常是正确的。

NOW.而是当下。

Maybe everything is established.也许一切均已确立。

ALLOW.允许。

Maybe an important evidentiary condition is unresolved.也许某项重要的证据条件尚未解决。

HOLD.暂缓。

Maybe the proposed action is prohibited under the applicable authority.也许拟议动作在适用权限下被禁止。

DENY.拒绝。

Maybe another authority must make the determination.也许必须由另一权限方作出裁定。

ESCALATE.上报。

The important point is that the AI’s recommendation remains a proposition until the consequence earns the right to cross the commit boundary.关键在于，在后果赢得跨越提交边界的权利之前，人工智能的建议始终只是一个命题。

WHY CHECK AGAIN?为何要再次检查？

Because reality does not stop changing while AI thinks.因为在人工智能思考之时，现实并未停止变化。

Suppose the AIR evidence was admitted for analysis at 10:00:00 AM.假设 AIR 证据在上午 10:00:00 被采纳用于分析。

The AI completes its analysis at 10:00:20.AI 在 10:00:20 完成分析。

At 10:00:25 it proposes changing the AHU.在 10:00:25，它提议更改 AHU。

But at 10:00:22 another material condition changed.但在 10:00:22，另一项实质状况发生了变化。

Occupancy changed.占用情况发生了变化。

Outdoor particulate levels increased.室外颗粒物浓度升高。

A fire mode activated.火灾模式已启动。

An equipment interlock changed state.设备联锁改变了状态。

Authority expired.权限已过期。

Another operator assumed control.另一名操作员接管了控制权。

A maintenance technician entered a hazardous area.一名维护技术人员进入了危险区域。

The evidence may have been perfectly appropriate for the AI’s analysis.该证据对于 AI 的分析可能完全合适。

That does not establish that the resulting physical action remains appropriate three seconds later.但这并不能证明由此产生的物理动作在三秒后仍然合适。

Every new consequence gets its own NOW.每个新后果都有自己的 当下。

THEN COMES THE COMMIT LINE接下来便是提交线

If the proposed consequence receives an ALLOW, it can move toward commit and execution through the appropriate governed system.如果提议的后果获得 ALLOW，它便可通过相应的受治理系统进入提交和执行阶段。

That might mean a BAS command.这可能意味着一条 BAS 命令。

It might mean a work order.这可能意味着一张工单。

It might mean notifying an operator.这可能意味着通知操作员。

It might mean scheduling an inspection rather than changing anything automatically.这可能意味着安排一次检查，而不是自动更改任何内容。

The appropriate consequence depends on the evidence, authority, standing and governing requirements.适当的后果取决于证据、权限、资格和治理要求。

But even after execution, I don’t want the architecture to assume success.但即使在执行之后，我也不希望该架构假定成功。

A command is not an outcome.命令不等于结果。

The BAS saying “command sent” doesn’t prove that the intended physical result occurred.BAS 显示“命令已发送”并不能证明预期的物理结果已经发生。

So we return to where we started.因此，我们回到起点。

Reality.现实。

DID IT ACTUALLY WORK?它真的奏效了吗？

AIR observes the building again.AIR 再次观察建筑。

What happened after the intervention?干预之后发生了什么？

Did airflow increase?气流是否增加？

Did pressure stabilize?压力是否稳定？

Did energy consumption improve or worsen?能耗是改善了还是恶化了？

Did indoor environmental conditions improve?室内环境状况是否改善？

Did another condition emerge?是否出现了其他状况？

Did the change persist?更改是否持续存在？

Now the building can preserve:现在，建筑可以保存：

BEFORE INTERVENTION干预前

↓

PROPOSED CONSEQUENCE提议的后果

↓

AUTHORIZED INTERVENTION经授权的干预

↓

EXECUTION执行

↓

OBSERVED OUTCOME观察到的结果

↓

NEW BASELINE新基线

That new baseline becomes part of the next Atmospheric Integrity Record.该新基线成为下一条 Atmospheric Integrity Record 的一部分。

And the loop begins again.循环再次开始。

CONTINUOUS MONITORING IS NOT GOVERNED AIR持续监控不是 Governed Air

I’ve said this many times:这话我已经说过很多次：

Continuous monitoring is not Governed Air.持续监控不是 Governed Air。

I still believe that.我仍然相信这一点。

But I think we need to add something important:但我认为我们需要补充一点重要内容：

The continuous monitoring you already have may become the evidence foundation for Governed Air.你现有的持续监控可能成为 Governed Air 的证据基础。

That distinction matters.这一区别很重要。

TA-14 isn’t arguing that owners need to throw away their sensors, BAS platforms, dashboards, gateways, analytics or AI systems.TA-14 并不是说业主需要丢弃他们的传感器、BAS 平台、仪表盘、网关、分析工具或 AI 系统。

Quite the opposite.恰恰相反。

Much of the infrastructure may already exist.大部分基础设施可能已经存在。

The missing architecture may be the governance connecting observation to reliance and reliance to consequence.缺失的架构可能是那层治理，将观察连接到信赖，再将信赖连接到后果。

Continuous monitoring tells us what the building appears to be doing.持续监控告诉我们建筑看似在做什么。

Atmospheric Integrity Records preserve what was actually recorded and the integrity context necessary to evaluate what that record can support.Atmospheric Integrity Record 保存实际记录的内容，以及评估该记录能够支持什么所需的完整性上下文。

Intelligence interprets that governed evidence.智能解释该受治理证据。

Admissible Execution Architecture governs whether the resulting consequence may cross into reality.Admissible Execution Architecture 负责管控由此产生的后果是否可以进入现实。

And the resulting reality is observed again.而由此产生的现实会被再次观察。

The loop becomes:循环变为：

REALITY → AIR → GOVERNED RELIANCE → INTELLIGENCE → PROPOSED CONSEQUENCE → AEA → COMMIT → EXECUTION → OUTCOME → NEW REALITY现实 → AIR → 受治理信赖 → 智能 → 提议的后果 → AEA → 提交 → 执行 → 结果 → 新现实

Then:然后：

NEW REALITY → NEW RECORD新现实 → 新记录

INTEROPERABILITY IS NO LONGER ENOUGH互操作性已不再足够

For decades, the smart-building industry has worked on an enormously important question:数十年来，智慧建筑行业一直在研究一个极其重要的问题：

Can these systems exchange information?这些系统能否交换信息？

Protocols, gateways, semantic models, APIs and connection architectures have helped us answer that question.协议、网关、语义模型、API 和连接架构已帮助我们回答了这个问题。

But AI introduces another question.但 AI 引出了另一个问题。

Should this information cross this boundary for this purpose?此信息是否应当为了这一目的跨越此边界？

And then another:然后是另一个问题：

What may the receiving intelligence cause because it did?接收方智能体因其行为可能造成什么后果？

A successful connection does not answer either question.成功建立连接并不能回答这两个问题中的任何一个。

Neither does authentication.身份验证也不能。

Neither does a valid command.有效命令也不能。

Neither does a sophisticated model.复杂模型也不能。

Neither does confidence.置信度也不能。

Neither does intelligence.智能本身也不能。

These systems remain critically important, but the consequence boundary is different.这些系统仍然至关重要，但后果边界有所不同。

We have moved from asking whether machines can communicate to asking what happens when their communication can autonomously change reality.我们已经从追问机器能否通信，转向追问当机器通信能够自主改变现实时会发生什么。

GOVERN BOTH SIDES OF INTELLIGENCE治理智能的两端

This is the architecture I believe autonomous buildings will eventually require.我相信，这就是自主建筑最终将需要的架构。

Not because AI is bad.不是因为 AI 有害。

Because AI is powerful.而是因为 AI 强大。

We should want AI looking at years of building performance.我们应当希望 AI 审视多年的建筑性能。

We should want it finding mechanical drift before equipment fails.我们应当希望它在设备故障前发现机械漂移。

We should want it finding environmental drift before occupants complain.我们应当希望它在使用者投诉前发现环境漂移。

We should want it comparing interventions against actual outcomes.我们应当希望它将干预措施与实际结果进行比较。

We should want it helping technicians understand complicated systems.我们应当希望它帮助技术人员理解复杂系统。

We should want it discovering relationships humans may never find manually.我们应当希望它发现人类靠人工可能永远无法发现的关联。

But intelligence should operate inside boundaries.但智能应在边界内运行。

On the way into intelligence:在智能的输入端：

What evidence may this system legitimately rely upon?本系统可正当依赖哪些证据？

On the way out:在智能的输出端：

What consequence may this system legitimately cause?本系统可正当造成什么后果？

And between those two boundaries, keep the layers separate.在这两个边界之间，应保持各层相互分离。

Record ≠ Diagnosis记录 ≠ 诊断

Diagnosis ≠ Optimization诊断 ≠ 优化

Optimization ≠ Authority优化 ≠ 权限

Capability ≠ Authority能力 ≠ 权限

Understanding ≠ Permission理解 ≠ 许可

An AI can be extraordinarily intelligent and still be wrong.AI 可能极其智能，却仍然会出错。

Governed evidence does not eliminate hallucination.受治理的证据并不能消除幻觉。

A sophisticated model does not eliminate uncertainty.复杂模型并不能消除不确定性。

That is exactly why a recommendation should not silently become permission.正因如此，建议不应悄无声息地变为许可。

FROM SMART BUILDINGS TO GOVERNED BUILDINGS从智慧建筑到受治理建筑

The smart building was largely about connecting things.智慧建筑主要关乎连接万物。

The intelligent building is increasingly about understanding things.智能建筑正日益关乎理解万物。

The autonomous building will be about allowing machines to do things.自主建筑将关乎允许机器采取行动。

That last transition changes the architectural question.这最后一种转变改变了架构问题。

Once intelligence can create physical, operational, financial, environmental or human consequence, we need to know what evidence it relied upon, whether that evidence had continuity and admissibility, what authority applied, who or what had standing, what exact consequence was proposed, what crossed the commit boundary and what actually happened afterward.一旦智能能够造成物理、运行、财务、环境或人员方面的后果，我们就需要知道它依赖了什么证据，这些证据是否具备连续性和可采性，适用什么权限，谁或什么具有资格，提议的确切后果是什么，什么越过了提交边界，以及之后实际发生了什么。

That is where I believe building automation is heading.我相信，楼宇自动化正朝着这个方向发展。

Not away from AI.不是远离 AI。

Toward AI that can participate in a governed chain from reality back to reality.而是走向能够参与从现实返回现实的受治理链条的 AI。

Govern what intelligence may rely upon.治理智能可以依赖的内容。

Govern what intelligence may cause.治理智能可能造成的后果。

Preserve what actually happened.留存实际发生之事。

Then return to reality and do it again.然后返回现实，重新执行。

Because ultimately, the most important question in an autonomous building isn’t whether the machine is intelligent enough to act.因为归根结底，自主建筑中最重要的问题并非机器是否足够智能到可以行动。

It is:而是：

Does this proposed consequence have sufficient Admissible Evidence, Applicable Authority, and Established Standing to become reality NOW?这一提议后果是否具备充分的可采证据、适用权限和既定资格，能够即刻成为现实？

The post What Does Your Building Let AI Know — and What Does It Let AI Do? appeared first on AutomatedBuildings.com.文章 你的建筑让 AI 知道什么——又让 AI 做什么？ 最早发表于 AutomatedBuildings.com。