人工智能需要第一性原理思維

正確的方向比速度更重要。

The right vector matters more than velocity.

作者： 庫(kù)爾特·唐納爾
管理顧問彼得·德魯克曾教導(dǎo)《從優(yōu)秀到卓越》的作者吉姆·柯林斯一條徹底改變他經(jīng)商方式的原則：能做一件事就不要做一百件事。這個(gè)洞見看似簡(jiǎn)單，實(shí)則不然——看似無數(shù)個(gè)互不相干的決定，往往只是同一個(gè)決定的不同偽裝。

史蒂夫·喬布斯曾因每天穿同一套衣服而臭名昭著，他認(rèn)為這樣可以省去很多穿衣打扮的麻煩。然而，這種思維方式的真正力量遠(yuǎn)不止于此。它關(guān)乎于做出能夠徹底排除未來決策類別的戰(zhàn)略性決定。

能夠有效應(yīng)對(duì)決策疲勞的領(lǐng)導(dǎo)者，其績(jī)效比同行高出22%，由此可見，這一原則至關(guān)重要。但在當(dāng)今人工智能驅(qū)動(dòng)的世界里，它更是不可或缺。

決策疲勞的影響

據(jù)估計(jì)，普通人每天要做出35000個(gè)決定。等到真正需要做決定的時(shí)候，他們的腦力往往已經(jīng)耗盡。這種認(rèn)知超負(fù)荷會(huì)損害自我調(diào)節(jié)能力，導(dǎo)致做出次優(yōu)選擇，每年給全球經(jīng)濟(jì)造成數(shù)十億美元的生產(chǎn)力損失。

當(dāng)認(rèn)知負(fù)荷超過大腦的承受能力時(shí)，大腦調(diào)節(jié)沖動(dòng)和情緒的能力就會(huì)下降，這往往會(huì)導(dǎo)致我們做出下意識(shí)的反應(yīng)，變得過度規(guī)避風(fēng)險(xiǎn)，或者干脆避免做決定。

對(duì)錯(cuò)誤問題的正確答案
人工智能并沒有減輕我們的決策負(fù)擔(dān)，反而放大了我們做出錯(cuò)誤決策的后果。例如，如果你決定在A和B之間做出選擇，你可能會(huì)構(gòu)建一個(gè)人工智能系統(tǒng)，它能夠出色地在A和B之間做出選擇，考慮所有相關(guān)因素并進(jìn)行完美優(yōu)化。

然而，你或許根本無需做出選擇，因?yàn)橥瑫r(shí)選擇A和B，或者兩個(gè)選項(xiàng)都不選，可能反而更好。你只是花費(fèi)了大量的時(shí)間、精力和資源，卻在為一個(gè)完全錯(cuò)誤的問題進(jìn)行優(yōu)化。例如，一家轎車銷量下滑的汽車制造商可能會(huì)構(gòu)建一個(gè)復(fù)雜的、多變量的AI增強(qiáng)系統(tǒng)，來優(yōu)化在每個(gè)市場(chǎng)生產(chǎn)和向經(jīng)銷商交付哪些顏色的轎車，而更重要的問題是：應(yīng)該生產(chǎn)轎車、SUV還是卡車？

速度會(huì)放大誤差
人工智能讓我們行動(dòng)更快。應(yīng)用程序原型設(shè)計(jì)、假設(shè)驗(yàn)證和解決方案生成正以前所未有的速度進(jìn)行。但飛機(jī)偏離航線一度，每飛行 60 海里，最終就會(huì)偏離航線一海里。以人工智能的速度進(jìn)行開發(fā)時(shí)，一度的誤差會(huì)呈指數(shù)級(jí)增長(zhǎng)。

危險(xiǎn)不在于人工智能本身，而在于人們更容易鉆研細(xì)節(jié)，為一些根本不需要解決的問題構(gòu)建復(fù)雜的解決方案。我們加快了進(jìn)程，但除非我們都朝著正確的目標(biāo)前進(jìn)，并擁有正確的基本假設(shè)，否則加快速度又有什么意義呢？

我親眼見過這種情況。你可以產(chǎn)出驚人的成果——編寫大量代碼、構(gòu)建令人印象深刻的功能、創(chuàng)建漂亮的儀表盤。但如果這些成果沒有帶來預(yù)期的結(jié)果，那就是浪費(fèi)時(shí)間和精力。

運(yùn)用第一性原理思考
這就是第一性原理思維成為你競(jìng)爭(zhēng)優(yōu)勢(shì)的地方。它是一種解決問題的方法，能夠?qū)?fù)雜問題分解成最基本、最根本的真理。它摒棄了假設(shè)，從零開始構(gòu)建解決方案，而不是通過類比推理。

當(dāng)事情開始變得復(fù)雜時(shí)，就應(yīng)該停下來，回到最初。我們最初的假設(shè)是什么？驗(yàn)證它，如果它是錯(cuò)誤的，就能解釋為什么后續(xù)所有環(huán)節(jié)都受到影響。

在實(shí)施人工智能解決方案或要求團(tuán)隊(duì)優(yōu)化流程之前，請(qǐng)回答以下三個(gè)問題：

1. 我們真正想要解決的問題是什么？不是癥狀，不是眼前的痛點(diǎn)，而是根本問題。

2. 我們對(duì)這個(gè)問題的原因做了什么假設(shè)？質(zhì)疑這個(gè)假設(shè)是否正確。

3. 解決這個(gè)問題對(duì)我們的核心目標(biāo)有何幫助？如果你無法將這個(gè)解決方案與你的主要目標(biāo)直接聯(lián)系起來，那么你很可能是在追求錯(cuò)誤的目標(biāo)。

指標(biāo)錯(cuò)位會(huì)使人工智能失去作用
最陰險(xiǎn)的錯(cuò)誤決策莫過于不同團(tuán)隊(duì)朝著不同的目標(biāo)努力，并使用不同的指標(biāo)來衡量成功。我曾親眼目睹一些公司部署了令人驚嘆的人工智能技術(shù)，但他們的團(tuán)隊(duì)成員對(duì)成功的定義卻各執(zhí)一詞。最終，他們反而成為了實(shí)現(xiàn)目標(biāo)的絆腳石。

如果你的組織內(nèi)部不同部門各自為政，那么你的人工智能技術(shù)再先進(jìn)、執(zhí)行速度再快都無濟(jì)于事。在加速推進(jìn)其他所有工作之前，你需要明確真正重要的決策——你的方向。

這并不是說你不應(yīng)該進(jìn)行一些充滿創(chuàng)意的編程探索，探索各種可能性。但要把這些探索當(dāng)作概念驗(yàn)證。進(jìn)行壓力測(cè)試，確保概念驗(yàn)證朝著解決挑戰(zhàn)的方向發(fā)展，然后充分發(fā)揮你那支人工智能賦能團(tuán)隊(duì)的全部潛能。

贏得規(guī)模化發(fā)展的權(quán)利
這里有點(diǎn)違反直覺：有時(shí)候你需要先獲得許可，才能做到效率低下。正如身著黃色燕尾服的薩凡納香蕉隊(duì)創(chuàng)始人杰西·科爾所說：“你需要先做那些無法規(guī)模化的事情，才能做那些可以規(guī)模化的事情。”

在 Freestar，我們特意安排人手處理一些顯然更適合自動(dòng)化解決的問題。當(dāng)時(shí)感覺效率不高，但正是這種先摸索后學(xué)習(xí)的過程，讓我們明白了高效的方法。我們必須先深入理解問題，才能系統(tǒng)化地構(gòu)建解決方案。這并非浪費(fèi)時(shí)間，而是第一性原理思維的實(shí)踐。

做出更好的初始決定
如果使用得當(dāng)，人工智能可以幫助你更快地檢驗(yàn)基礎(chǔ)假設(shè)。你可以在幾天內(nèi)（而不是幾個(gè)月）完成解決方案原型設(shè)計(jì)、數(shù)據(jù)收集和理論驗(yàn)證（或推翻）。在加速應(yīng)用人工智能之前，請(qǐng)先掌握以下三個(gè)決策框架：

1. 一次決定，處處適用
哪些可以一概而論的重復(fù)性決策？記錄下來，系統(tǒng)化。這樣可以解放團(tuán)隊(duì)的認(rèn)知資源，讓他們專注于真正全新的情況。

2. 對(duì)這個(gè)問題提出質(zhì)疑
當(dāng)有人讓你在幾個(gè)選項(xiàng)中做出選擇時(shí)，停下來問問自己，整個(gè)框架是否正確。通常，最佳答案就隱藏在對(duì)前提的質(zhì)疑之中。

3. 衡量真正重要的事
確保整個(gè)組織對(duì)成功的定義保持一致。如果各個(gè)團(tuán)隊(duì)追求的指標(biāo)各不相同，那就已經(jīng)偏離了方向。

一旦方向正確，人工智能就能成為你的放大器。但如果方向錯(cuò)誤，你只會(huì)更快地迷失方向。因此，當(dāng)人工智能讓我們?cè)谝郧白鲆淮螞Q定所需的時(shí)間內(nèi)做出100次決定時(shí)，你無需成為速度最快的人；你只需要專注于做出正確的第一次決定。

Kurt Donnell 是 Freestar 的首席執(zhí)行官。

關(guān)于作者
自 2019 年以來，庫(kù)爾特·唐納爾 (Kurt Donnell) 一直擔(dān)任 Freestar 的總裁兼首席執(zhí)行官。Freestar 是一家為數(shù)字媒體出版商提供廣告技術(shù)服務(wù)的公司。在他的領(lǐng)導(dǎo)下，F(xiàn)reestar 從一家總部位于美國(guó)的 25 人團(tuán)隊(duì)發(fā)展成為一家遍布 15 個(gè)國(guó)家/地區(qū)、擁有近 200 名員工的公司，收入增長(zhǎng)超過 500%，并被評(píng)為美國(guó)增長(zhǎng)最快的私營(yíng)公司。

AI requires first principles thinking

The right vector matters more than velocity.

Management consultant Peter Drucker once taught Jim Collins, author of Good to Great, a principle that transformed how he approached business: Don’t make a hundred decisions when one will do. The insight is deceptively simple—what appear to be countless disparate decisions are often the same decision wearing different disguises.

ADVERTISEMENT

Steve Jobs famously wore the same outfit daily to eliminate fashion decisions, although the real power of this thinking went far beyond wardrobe choices. It was about identifying strategic decisions that categorically remove entire categories of future decisions.

Leaders who effectively manage decision fatigue outperform peers by 22%, so we know this principle is relevant. But in today’s AI-powered world, it’s become absolutely critical.

THE IMPACT OF DECISION FATIGUE

Subscribe to the Daily newsletter.Fast Company's trending stories delivered to you every day

Privacy Policy

Fast Company Newsletters

ADVERTISEMENT

The average person makes an estimated 35,000decisions per day. By the time they reach the important ones, their mental fuel tank is often empty. This cognitive overload impairs self-regulation and can lead to suboptimal choices, costing the global economy billions annually in lost productivity.

When cognitive load exceeds capacity, the brain’s ability to regulate impulses and emotions is diminished, often leading to us making knee-jerk reactions, becoming overly risk-averse, or avoiding decisions altogether.

THE RIGHT ANSWER TO THE WRONG QUESTION

AI hasn’t reduced our decision load, but it’s amplified the consequences of getting our foundational decisions wrong. For instance, if you decide to choose between A or B, you might then build an AI system that does an amazing job selecting between A and B, taking in all the relevant factors and optimizing beautifully.

However, you may not have needed to choose in the first place, because choosing both A and B, or picking neither option may have been better. You’ve just spent enormous time, energy, and resources optimizing for the wrong question entirely. For example, a car manufacturer with flagging sedan sales could build a complex, multi-variate AI-enhanced system to optimize which sedan colors to manufacture and send to dealers in each market, when the more important question is whether to manufacture sedans, SUVs, or trucks.

SPEED MAGNIFIES THE ERROR

AI allows us to move faster. Prototyping apps, testing assumptions, and generating solutions is happening at unprecedented speed. But a plane off course by one degree ends up a full nautical mile off course for every 60 nautical miles flown. When you’re building at AI speed, a one-degree error compounds exponentially.

The danger isn’t AI, but that it is easier to go down rabbit holes, building sophisticated solutions to problems that don’t need solving. We’ve accelerated accomplishment. But unless we’re all heading toward the correct goal with the right foundational assumptions, what’s the point of going faster?

I’ve seen this firsthand. You can show tremendous output—generate reams of code, build impressive features, create beautiful dashboards. But if it’s not driving the right outcome, it’s a waste of time and energy.

USE FIRST PRINCIPLES THINKING

This is where first principles thinking becomes your competitive advantage. It’s a problem-solving method that breaks complex problems down into their most basic, fundamental truths. It eliminates assumptions and builds solutions from scratch rather than reasoning by analogy.

When something starts feeling complex, that’s your signal to stop and go back to the beginning. What was the first assumption we made? Test that, and if it’s wrong, it explains why everything downstream is compromised.

Before implementing an AI solution or asking a team to optimize a process, answer these three questions:

1. What problem are we actually trying to solve?Not the symptom, not the immediate pain point, but the foundational problem.

2. What assumption are we making about why this is a problem?Challenge whether that assumption is true.

3. What would solving this accomplish for our core goals?If you can’t draw a straight line from this solution to your primary objectives, you’re likely optimizing for the wrong thing.

MISALIGNED METRICS MAKE AI USELESS

The most insidious form of wrong first decisions is when different teams are working toward different goals and using different metrics to measure success. I’ve watched companies implement incredible AI capabilities while their teams speak different languages about what success means. They become their own barrier to achievement.

It doesn’t matter how sophisticated your AI is or how fast you can execute if different parts of your organization are on different vectors. You need clarity on the decision that matters—your true north—before accelerating everything else.

This isn’t to say that you shouldn’t have creative vibe coding sessions to explore the art of the possible. But treat the sessions as proofs of concept. Pressure test, ensure the proof of concept is heading the direction that will solve your challenge, then unleash the full power of your AI-supercharged team.

EARN THE RIGHT TO SCALE

Here’s the counterintuitive part: Sometimes you need permission to be inefficient first. As Jesse Cole, the yellow tuxedo-clad founder of the Savannah Bananas said, “You need to do the unscalable to do the scalable.”

At Freestar, we’ve purposefully thrown people at problems that undoubtedly would be better handled by automation. It felt inefficient in the moment. But doing things the hard way first taught us what the efficient way looks like. We had to understand the problem deeply before we could systematize the solution. That’s not wasted time—that’s first principles thinking in action.

MAKE BETTER FIRST DECISIONS

AI can help you test foundational assumptions faster if you use it correctly. You can prototype solutions, gather data, and validate (or invalidate) theories in days instead of months. Before you accelerate with AI, anchor yourself with these three decision-making frameworks:

1. DECIDE ONCE, APPLY EVERYWHERE

What recurring decisions can you make categorically? Document them. Systematize them. Free your team’s cognitive resources for genuinely novel situations.

2. CHALLENGE THE QUESTION

When someone asks you to choose between options, pause and ask whether the entire framing is correct. Often, the best answer is hidden in questioning the premise.

3. MEASURE WHAT MATTERS

Align your entire organization on shared definitions of success. If teams are optimizing for different metrics, you’re already off course.

Once you’re pointed in the right direction, AI becomes your amplifier. But if you’re aimed wrong, you just get lost faster. So, when AI lets us make 100 decisions in the time it used to take to make one, you don’t need to be the fastest; you just need to focus on getting the first decision right.

Kurt Donnell is the CEO of Freestar.

The final deadline for Fast Company's Brands That Matter Awards is Friday, May 15, at 11:59 p.m. PT. Apply today.