Can AI Help Solve the Climate Crisis? | Sims Witherspoon | TED

譯者: C Leung 審譯者: 麗玲 辛

You may have had the experience of unboxing furniture

你可能有拆開家具包裝的經驗，

and come across instructions that go something like this:

並遇到類似這樣的說明：

"Assemble the bookshelf according to the provided diagram."

「根據示意圖，組裝書架。」

Yes, I know what a bookshelf looks like.

我當然知道書架是什麼樣子。

Probably wouldn't be reading the assembly instructions

要不是在過程中需要一點幫助，

if I didn't need a little more help with the process.

我才不會去看組裝說明。

Or maybe you've opened a cookbook

或許你曾打開一本烹飪書，

with an author who thinks you're already somewhat of a chef.

而該作者假設你已有些廚藝基礎。

"Deglaze the pan."

「洗鍋收汁。」

What?

什麼？

(Laughter)

（笑聲）

OK, off I go on a separate search to understand whatever that means.

好吧，我就另外去搜尋一下 以理解它的意思。

Instructions that tell you what to do and not how to do it

那些告訴你該做什麼， 而不是如何做的使用說明

are pretty useless.

是毫無用處的。

And yet, even when we're talking

然而，即使我們在談論

about something as important as climate change,

像氣候變化如此重要的事情時，

we hear them all the time.

我們總會聽到這些聲音：

“Transition to renewable energy.”

「過渡到再生能源。」

“Electrify everything else.”

「將一切電氣化。」

“Deploy solutions that are equitable and fair.”

「部署公平公正的解決方案。」

Yes, let's do all of that.

是的，讓我們來做這一切。

But how?

但怎樣做呢？

Answering how is where we understand which solutions are actually feasible,

回答「怎樣做」，我們就了解 哪些解決方案實際上可行，

whether that be with today's infrastructure,

無論是當今的基礎設施，

our evolving regulatory environment

不斷發展的監管環境，

or any of the other number of dependencies and constraints

還是其他我們必須考慮的 依賴關係和限制。

that we have to consider.

How we solve climate change

我們如何解決氣候變化

also depends on our very definition of the problem.

也取決於我們對這個問題的定義。

It's a scientific challenge, a sociopolitical issue,

這是科學挑戰、社會政治問題、

an economic problem and so much more.

經濟問題等等。

And how we solve it will depend on how we frame it.

我們如何解決它 將取決於我們如何構建它。

There is no single answer.

沒有單一答案。

I'm a scientist,

我是科學家，

so I approach climate change as a scientific challenge.

所以我將氣候變化視為一項科學挑戰。

I'm also a techno-optimist

我也是技術樂觀主義者

and artificial-intelligence product manager,

和人工智能慧產品經理，

so I also approach it as a technological one.

因此我也視其為技術挑戰。

When it comes to a sustainable future,

談到永續發展的未來，

artificial intelligence can help us do three critical things.

人工智慧可以幫助我們 做三件關鍵事情。

First, it can help us understand climate change

首先，它可以幫助我們了解氣候變化

and its effects on Earth's ecosystems.

及其對地球生態系統的影響。

Second, it can help us optimize current systems and infrastructure,

其次，它可以幫助我們最佳化 當前的系統和基礎設施，

because we can't just start over from scratch today.

因為我們不能在這刻才從頭起步。

And third, it can help us accelerate the breakthrough science we need,

第三，它可以幫助我們加速 所需的科學突破，

such as fusion as a carbon-free energy source.

例如作為無碳能源的核聚變。

Today, I'd like to talk about that second one,

今天，我想談談第二個關鍵：

optimizing current systems,

最佳化當前系統，

and specifically, how we can use AI to harness a superpower

特別是，我們如何利用 AI 去駕馭

我們在這場戰鬥中已擁有的巨大力量：

we already have in this fight:

wind energy.

風能。

Renewables are unquestionably a key to a sustainable future,

再生能源無疑是永續發展未來的關鍵，

but the problem is they're unpredictable.

但問題在於它們是不可預測的。

Sometimes, the sun shines and the wind blows,

有時候，風和日麗；

and sometimes, it just doesn't.

有時候，靜如止水。

Now, for an electricity systems operator,

對於電力系統營運商來說，

who needs supply to meet demand in real time, 24-7,

它們需要全天候實時滿足供電需求，

this is hugely problematic.

這是非常大的問題。

Renewables can't be 100 percent reliably scheduled.

再生能源無法百分之百 可靠地預先安排。

Now, unfortunately, fossil-fuel plants are the opposite.

不幸的是，化石燃料發電廠卻可以。

You can burn a specific amount of coal at a set time

你可以在預設的時間內， 燃燒特定數量的煤炭，

to deliver exactly the amount of electricity you want

以便在可預測的時間範圍內

in a predictable time window.

準確提供你想要的電量。

So ...

所以...

if you're a power systems manager

如果你是電力系統經理，

whose job is to literally keep the lights on,

工作就是確保電燈保持亮著，

which source are you more confident depending on?

你對哪個來源更有信心？

But here's one of the places where AI can come in.

這是 AI 可以發揮作用的地方。

It is a powerful tool for forecasting.

它是強大的預測工具。

AI systems can ingest vast amounts of historical data

AI 系統可以擷取大量歷史數據，

and help us predict future events.

並協助我們預測未來事件。

So, while we can't eliminate the variability of wind,

雖然我們無法消除風的可變化性，

we can use AI to more accurately predict its availability.

但我們可以使用 AI 更準確地預測其可用性。

That was my team’s “what” to do.

這就是我的團隊要做的「事情」，

Use AI to accelerate the transition to renewables, like wind energy.

使用 AI 加速過渡到 再生能源，如風能，

The tough part was the “how” to do it.

困難的是「怎樣」做到這點。

First, we researched the challenge.

首先，我們研究了這個難題。

We read papers, we spoke to domain experts,

我們閱讀論文，與領域專家交流，

we found out everything we could about the problem.

我們儘力找出關於這問題的一切面向。

Our team, which is a mix of research scientists,

我們的團隊由研究科學家、

engineers, a product manager, a program manager

工程師、產品經理、專案經理

and an impact analyst,

和影響力分析師組成，

decided that a neural net trained on historical weather data

他們認為，根據歷史天氣資料 和渦輪發電生產資訊

and turbine power-production information

訓練而成的神經網路，

would likely help us accomplish our goal.

可能會幫助我們實現目標。

Next, we needed to find two core elements:

接下來，我們需要找到兩個核心要素：

data to train the system

用於訓練系統的數據

and a partner who was willing to deploy it.

和願意部署系統的合作夥伴。

Both of these can be major obstacles

在實際情境中部署 AI 時，

when it comes to deploying AI in real-world scenarios.

這兩者都可能是主要障礙。

Let's start with data.

讓我們從數據開始。

There are massive gaps in climate-critical data --

關鍵氣候數據存在巨大差距，

not just in electricity,

不僅在電力方面，

but in agriculture, transportation, industry and many other sectors.

而且在農業，運輸， 工業和許多其他領域。

Some of our data, we could purchase or download for free --

我們可以購買或免費下載某些數據，

weather forecasts, for instance.

例如，天氣預報。

But some of the data we needed was proprietary,

但我們需要的一些數據是專有的，

and this would be, like, turbine power-production information

例如渦輪機發電資訊

and other operational data from the wind farms.

和風力發電場的操作數據。

Now, we needed that proprietary data so that we could train our models

我們需要這些專有數據 來訓練我們的模型，

to learn the relationship between historical weather

學習了解過去天氣 和電力生產之間的關係，

and historical power production,

so it could then then make predictions about future power availability

然後它可以根據未來天氣的數據，

based on what data said about future weather.

預測未來的電力可用性。

Now it's probably worth mentioning here

在此或值得一提的是，

that we were looking at a few years of data

我們正查看幾年來， 每小時分辨率的數據，

on hourly resolution,

而不是時間刻度上的歷史數據，

not historical data at a timescale

that would have massive climactic differences from present day.

因它們與當今數據 存在巨大氣候差異。

In addition to data,

除了數據外，

we needed to find a partner with domain expertise

我們還需要尋找具備領域專業知識，

and the willingness and scale to test new systems.

有意願、也有規模的合作夥伴 來測試新系統。

You know, surprisingly, this can be a major hurdle

出人意外地， 在現實世界中部署 AI，

when it comes to deploying AI in the real world.

這可能是個主要障礙。

Believe it or not, it's not every wind-farm manager

信不信由你， 並不是每個風電場經理

that wants to let a bunch of AI researchers

都想讓一群 AI 研究人員，

test on their multimillion- or multibillion-dollar systems.

在他們價值數億或數十億美元的 系統上進行測試。

But the thing is, in order to prove that AI works,

但問題是， 為了證明 AI 有效，

we have to have deployment opportunities in the real world.

我們必須在現實世界中有部署機會。

Luckily for us, Google was a ready and willing partner.

幸運地，谷歌是個現成 而樂意的合作夥伴。

OK, yes, DeepMind is a part of Google,

是的，DeepMind 是 谷歌的一部分，

but it's not a given that they would let us test on their systems.

但他們不一定會讓我們 在其系統上測試。

Yet they let us test on 700 megawatts of their wind-power capacity,

然而，他們讓我們測試 700 兆瓦 的風力發電能力，

which is equivalent to a large wind farm in the United States.

相當於美國的一個大型風力發電場。

This made them an excellent proxy for external wind-farm operators.

這使它們成為外部 風電場營運商的優秀代理。

They also lent us an expert team to advise on metrics and benchmarks

他們還借給我們一個專家團隊， 就指標和基準測試提供建議，

and to share the data that we needed.

並共享我們所需的數據。

This is another critical component of AI for the real-world deployments.

這是 AI 在現實世界部署的 另一關鍵組件。

Working with a domain-expert team that can tell you what they need,

與領域專家團隊合作， 他們可告訴你他們的需求、

how they need it to work,

如何運作、

which constraints keep the system safe,

哪些限制可確保系統安全、

what quantifiable metrics to use to measure AI performance

使用哪些量化指標 來衡量 AI 性能，

and how much better that AI performance needs to be

以及 AI 性能需要 比之前的系統要好多少，

than their previous systems

to make the cost of switching over even worth it.

才能讓切換成本物超所值。

And that's just to name a few.

而這只是幾個例子。

So at this point, we have our idea,

到這時候，我們已經有想法、

we have our data, we have our deployment partner.

有數據、有部署合作夥伴。

Now, to test and deploy our system.

現在談談測試和部署我們的系統。

Improving the accuracy of electricity-supply forecast

提高電力供應預測的準確性

is incredibly important.

極其重要。

If predictions are higher than actual generation,

如果預測值高於實際發電，

renewable electricity managers may not have enough supply to meet demand.

再生電力管理者未必 有足夠的電力供應來滿足需求。

This, in turn, drives the purchase of carbon-intensive fossil fuels

這反而會迫使他們 購買碳密集型化石燃料，

to cover that gap,

以彌補缺口，

because they're largely what makes up backup generation.

因為它們主要組成備用發電。

Now, the good news.

好消息是，

Our AI system performed 20 percent better than Google's existing systems.

我們的 AI 系統效能 比谷歌現有系統好 20％。

Even better news is that Google decided to scale this technology.

更好的是，谷歌決定 擴展這項技術。

And scaling is so important.

規模化是如此重要。

We will run out of time in the climate countdown

如果我們不去部署 廣泛適用的解決方案 ，

if we aren't deploying solutions that are widely applicable.

我們的時間就會於氣候倒數中耗盡。

This particular solution is being developed

這種特殊的解決方案 正在開發成為軟體產品，

into a software product

that French company Engie is among the first to pilot.

而法國公司 Engie 是首批試點公司。

But, you know,

但你知道，

it doesn't even take a major research organization to do this kind of work.

我們甚至不需要 大型研究組織來做這種工作。

Where we focused on AI for supply-side forecasting,

當我們專注於 AI 進行供給面的預測，

a small UK-based nonprofit called Open Climate Fix

有一家名為 Open Climate Fix 的 英國小型非營利組織

is focusing on AI for demand-side forecasting.

則專注於需求面預測的 AI 。

They found a willing partner in the UK National Grid,

他們在英國國家電網 找到了願意合作的夥伴，

and are currently deploying forecasts that are two times more accurate

目前正在部署的預測，

比英國電網先前使用的系統準確兩倍。

than the UK grid's previously used systems.

Now, all of this is to say is that AI can help us

這一切都在告訴我們，

AI 能幫助我們過渡到再生能源，

with the transition to renewable energy,

but scientists and technologists,

但科學家和技術人員，

we're not going to be able to do that alone.

並不能單獨做到這點。

We need to be working with partners and experts

我們要和合作夥伴及專家一同工作，

who can teach us the “how.”

因他們能教導我們「怎樣」做。

So for those of you interested in this space,

因此，對這個領域感興趣的人來說，

if you're a domain expert,

如果你是領域專家，

please share the problems you face and the challenges that you have

請分享你面對的問題和挑戰，

so that our sector can ensure

以便我們的行業可以確保

that AI pursuits will have impact in the real world

AI 的追求將在現實世界中產生影響，

and not be purely academic.

而不是純理論的。

Even better,

更好的是，

if you want to incentivize ML researchers to work on your problems,

若你想激勵機器學習的 研究人員去解決你的問題，

I'll let you in on a little secret:

告訴你一個小秘訣：

build a competition, and they will come.

舉辦競賽，他們就來。

(Laughter)

（笑聲）

It's true.

是真的。

Just don't forget the datasets and metrics.

不過，不要忘記數據集和指標。

If you are a data holder, where it’s safe and responsible to do so,

如果你是數據持有者， 在安全且負責的情況下，

please share data related to those challenges.

請分享與這些問題相關的數據。

If you're not sure whether the data you have is even climate-critical,

如果你不確定自己擁有的數據 是否對氣候至關重要，

you can check out Climate Change AI's website,

可以查看 Climate Change AI 的網站，

where they have published a wish list of climate-critical datasets.

在那裡他們發布了 氣候關鍵數據集的願望清單。

Access to these datasets

若能取得這些數據集，

would unblock crucial research and innovation in AI for climate.

將掃除 AI 在氣候方面 的關鍵研究和創新的障礙。

If you're a deployment partner,

如果你願意成為部署合作夥伴，

please, let us know who you are,

請告訴我們，

特別是如果你願意測試創新系統。

especially if you're willing to test innovative systems.

And for everyone who's interested in this space,

至於每一位對這個領域感興趣的人，

please know you do not have to be technical to work in tech.

請知道你不必具備技術背景， 也能從事技術工作。

AI for climate action requires a variety of skill sets

應用於氣候行動的 AI

需要各種技能和多元的背景，

and a diversity of backgrounds

that, yes, includes research scientists and engineers,

是的，包括研究科學家和工程師，

but it also includes ethicists and policy experts,

但也包括倫理學家和政策專家、

communication teams, product managers,

溝通團隊、產品經理、

program managers and so many more folks.

項目經理及其他人才等等。

Now for the warning label.

現在談談警告。

AI is not a silver bullet.

AI 不是靈丹妙藥。

It will not solve all problems driving climate change.

它無法解決 導致氣候變化的所有問題。

It isn't even the right tool for many of the challenges that we face.

對於我們面臨的許多挑戰， 它甚至不是正確的工具。

AI is also not a technology without tensions.

AI 也不是一門輕鬆科技。

It needs to be deployed safely and responsibly.

它需要安全地、負責任地部署。

Not to mention,

更不用說，

until our grids are run on clean energy, AI itself will carry a carbon footprint,

除非我們的電網使用清潔能源， AI 本身將帶來碳足跡，

as will any energy-intensive technology we use.

我們使用的任何 能源密集型科技也一樣。

But AI can be a transformational tool in our fight against climate change --

但是，AI 可以成為 我們應對氣候變化的轉型工具－

it's just on all of us to wield it effectively.

只是需要我們所有人有效地運用它。

The “why” we need to is absolutely harrowing.

我們要做的「原因」絕對痛苦，

The “what” we can do is really exciting.

而能做的「事情」確實令人興奮。

But it’s the “how” we can do it

但正是我們能「怎樣」做到，

that will illuminate feasibility and help us drive impact.

將啟發可行性，並幫助我們產生影響。

So, in your next climate action conversations,

因此，下一次，在氣候行動對話中，

when someone presents you with an exciting "what,"

當有人向你提出 讓人振奮的「事情」時，

please help to advance the conversation

請幫忙將對話推進，

to the impactful "how."

朝向有影響力的「怎樣」做。

Thank you.

謝謝。

(Cheers and applause)

（歡呼聲及掌聲）