王峰 AI × 经营 约一次诊断

总目 / 文章 / WF-0383

文章 · 存档

AI永远无法取代人,自动化才会

因为AI永远都是概率性的,只要是概率性的,就需要决策,只要需要决策就离不开人。

自动化是100%确认的流程和结果,除非故障和误差,否则就是100%的,这类工作就完全不需要决策,不需要决策的工作流才是取代人的。

AI永远无法取代人,自动化才会。

不是给AI唱反调,也不是给人类发奖状,而是在划一条很多人故意模糊的线:到底什么能被彻底拿走,什么永远拿不走。

先说最直白的那层意思:AI永远都是概率性的。其实人的思考本身就是概率性的,这一点永远都不会被打破,真正被打破的那一天,人也就不是人了。

只要是概率性的,就需要决策,只要需要决策就离不开人。

这不是哲学,是数学。

今天的模型再大,归根结底还是在下一个token的分布里抽样。它给出的答案永远带着一个隐形的大概率标签,哪怕99.999%也还是概率。

你把医生换成AI,它能把肺结节的诊断准确率提升到99.99%,但最后做出要不要动手术的是人,要不要接受手术的也还是人。

把法官换成AI,它能算出99.99%的量刑建议,但签不签判决书的终究还是人。

那0.01%的尾巴看起来小,却正好卡在人类生命、自由、尊严这些不能赌的地方。

反过来,自动化是100%确认的流程和结果,除非故障和误差,否则就是100%的,这类工作就完全不需要决策,不需要决策的工作流才是取代人的。

装配线上拧六个螺丝,仓库里按坐标码垛,银行后台对一笔跨行转账做三要素校验,这些事一旦写死规则,跑通流程就再也不需要有人盯着点头。

机器人不偷懒、不生气、不撂挑子、不内耗,流水线24小时亮着灯,效率直接碾压。

过去三十年被机器吃掉的岗位基本上都是这种确定性工作:打字员、接线员、柜台出纳、胶片洗印技师……它们消失的时候我们甚至没来得及伤感,因为生活确实变得更方便了。

所以真正的分水岭不是聪明不聪明,而是要不要赌,赌概率结果谁负责,负不负得起这个责。AI可以比人聪明100倍,但只要它还在赌概率就得有人在最后一步签字、扛雷。这就是为什么华尔街最好的量化基金仍然需要人类交易员坐在旁边,为什么特斯拉FSD再牛也得有人把手放在方向盘上,当然特斯拉已经做了没有方向盘的汽车,那么当道路上真的遇上“电车难题”的时候,正常行驶撞死5人,方向往左打一点点撞死另外的1个人,AI如何决断?按照概率,不言而喻,损失最小化,但左边那个无辜的人招谁惹谁了?凭什么?不打方向盘直接撞,不符合最小损失逻辑。这类决策是上升到伦理的瞬时决策,人类自己也没解决。

抛开这些极端事件,人类决策的所有结果考量里都有一个绕不过去的东西,叫做“运气”。运气好的人做的决策总是有好结果,别人没话可说。但运气不好的人即便是做了很正确的决策,结果也有可能是糟糕的,谁负责?决策人负责,负什么责?降职降薪开除甚至判刑,都行。但如果这个决策人是机器呢?谁让计算机降薪?人家没薪!让AI降职?人家没岗!开除AI?那你公司就停摆。不是因为模型不够好,而是因为一旦出事没人敢替一个概率模型去坐牢。

有点意思,对吧。

再往深里看,这件事正在悄悄重塑组织的权力结构。凡是能被自动化掉的基本上已经没议价权了,凡是离不开人的往往集中在最稀缺的两种能力上:

一是敢为概率买单的决策力,

二是能把模糊问题翻译成可执行方案的定义力。

前者是高管、投资人、医生、法官,后者是产品经理、策略分析师、咨询顾问、创意总监。他们不是在跟AI比谁算得快,而是在做AI永远做不了的事:把99.9%的确定性和0.1%的不确定性缝在一起,变成一个可以落地的选择。

我最近跟一个做供应链AI的朋友聊,他讲得特别透彻:他们的模型能把库存预测误差压到1%以内,但最后决定要不要多备10%的安全库存的一定只能留给老板。因为万一断货损失是几个亿,模型不会赔钱,老板会。他们花几千万买的AI归根结底是给老板买了一个敢拍板子的底气,而不是买了一个替代老板的机器人。

同样的故事在医疗、在金融、在司法、在战争里反复上演。AlphaFold把蛋白质折叠速度提高了1万倍,但新药最后敢不敢进三期临床,还是药厂高管签字;

chatGPT能在30秒写出比99%实习生更好的财报分析,但敢不敢把这篇分析直接塞进招股书,还是CFO和投行说了算。

技术越接近100%,最后0.01%的责任就越贵,越往金字塔顶端集中。

所以别担心AI抢饭碗,真正该担心的是自己干的活到底属于哪一类。

如果你的工作可以被写成输入A→输出B且B永远只有一种正确答案,那确实迟早要被自动化干掉,而且干掉得理直气壮。

但如果你的日常里经常出现“我也不确定,但我觉得应该这么干”或者“虽然数据说A,但这次我赌B”,那你反而越来越值钱——因为你做的事,AI一辈子都学不会。

自动化在吃掉确定性,AI在放大不确定性,而人活在两者之间那条最细的缝隙里。

那条缝隙看起来越来越窄,其实越来越贵。

未来最吃香的岗位可能不是会用AI的人,而是敢为AI的结果负责的人。

最值钱的技能可能不是写prompt,而是写免责声明和兜底方案。

所以下次再有人焦虑AI会不会把我取代,可以直接让他问自己一个问题:我每天干的活到底是100%确定该怎么干,还是99%确定但最后还得我拍板的。答案不同,命运完全两样。

AI永远无法取代人,自动化才会。这不是安慰剂,这是过去五年我亲眼看到的血淋淋现实,也是未来二十年最残酷也最公平的筛选器。

接受它并找准自己的位置,要么去写那100%的规则,要么去签那0.01%的字。别的路,早就被没了。

AI Can Never Replace People. Only Automation Can.

Because AI is always about chances. If something is about chances, someone has to make a choice. If a choice has to be made, you still need a person.

Automation is 100% sure steps and results. Unless it breaks or makes a mistake, it is always 100%. This kind of job needs no choices at all. Jobs that need no choices are the ones that can fully replace people.

AI can never replace people. Only automation can.

This is not hating on AI. It is not giving humans a gold star. It is drawing a clear line that many people try to make blurry: what can be taken away forever, and what can never be taken away.

Here is the simplest meaning: AI is always about chances. Even human thinking is about chances. That will never break. The day it really breaks, humans stop being humans.

If something is about chances, someone has to make a choice. If a choice has to be made, you still need a person.

This is not just big thinking. This is math.

Today's models can be super huge, but in the end they still pick the next word by looking at chances. Their answers always come with a hidden “most likely” tag. Even if it is 99.999%, it is still not 100%.

If you let AI be the doctor, it can say a lung spot is bad 99.99% of the time, but the person who says “yes, let's do surgery” is still a human. The person who says “yes, I will have the surgery” is still a human.

If you let AI be the judge, it can give a perfect sentence 99.99% of the time, but the person who signs the paper is still a human.

That tiny 0.01% looks super small, but it sits right on things we cannot bet with: life, freedom, and dignity.

Now look at the other side. Automation is 100% sure steps and results. Unless it breaks or makes a mistake, it is always 100%. This kind of job needs no choices. Jobs that need no choices are the ones that can fully replace people.

Putting six screws on a car part on the factory line, stacking boxes in the warehouse by exact spots, checking three things when money moves between banks—these jobs have fixed rules. Once the rules work, no one has to watch and say “okay.”

Robots do not get lazy. They do not get mad. They do not quit. They do not fight inside. The factory lights stay on 24 hours, and they beat humans at speed every time.

In the last thirty years, the jobs machines ate were almost all these sure jobs: people who typed letters all day, phone operators, bank tellers at the window, people who washed camera film… When they disappeared, we did not even have time to feel sad because life got easier.

So the real line is not “who is smarter.” The line is “do we have to bet?” Who pays if the bet is wrong? Can they pay that price?

AI can be 100 times smarter than people, but as long as it is betting on chances, a human has to sign at the last step and take the blame. That is why the best money-trading funds on Wall Street still keep human traders right there. That is why even the coolest Tesla self-driving still wants a hand near the wheel (Tesla already made cars with no wheel, but what happens on the road when it is the trolley problem—keep going straight and hit five people, or turn a tiny bit and hit one person? By chance numbers, it should turn and save more lives. But that one innocent person did nothing wrong. Why him? If it does not turn, it breaks the “save the most lives” rule. This choice is about right and wrong in a split second. Even humans have not solved it).

Even away from crazy examples, every human choice has one thing you cannot escape: luck. Lucky people make choices that turn out great, and everyone says good job. Unlucky people can make the right choice and still get a bad result. Who gets in trouble? The person who chose. They can lose their job, lose money, or even go to jail. All okay. But what if the chooser is a machine? How do you take away a computer's paycheck? It has no paycheck! How do you fire AI? The whole company stops. It is not because the model is bad. It is because when something bad happens, no one wants a chance-model to go to jail for it.

Kind of funny, right?

Now look deeper. This is quietly changing who has power in companies. Jobs that can be fully automated already have almost no bargaining power. Jobs that still need humans are usually the rarest two skills:

1. The courage to pay for a chance bet.

2. The skill to turn a messy problem into clear steps.

The first group: top bosses, investors, doctors, judges.

The second group: product managers, strategy people, consultants, creative directors.

They are not racing AI to see who calculates faster. They are doing what AI can never do: sew 99.9% sure things and 0.1% unsure things together into one real choice.

I talked to a friend who makes supply-chain AI not long ago. He explained it super clear: their model can make inventory guesses wrong less than 1% of the time, but the final call—“do we keep 10% extra just in case?”—can only be the boss. If they run out, they lose billions. The model does not pay. The boss does. They spent tens of millions on AI, but really they bought the boss more courage to make the final call, not a robot to replace the boss.

You see the same story in hospitals, banks, courts, and wars. AlphaFold made protein folding 10,000 times faster, but the drug company boss still signs to start the big expensive human tests. ChatGPT can write a financial report better than 99% of interns in 30 seconds, but the CFO and bankers decide if it goes into the big company selling papers.

The closer tech gets to 100%, the more expensive that last 0.01% responsibility becomes. It moves up to the top of the pyramid.

So do not worry that AI will steal your lunch. Worry about what kind of job you really have.

If your job can be written as “put in A, always get the one right B,” then yes, automation will take it one day, and it has every right.

But if your day is full of “I'm not totally sure, but I think we should do this” or “data says A, but this time I pick B,” then you are getting more valuable—because AI will never ever learn that.

Automation is eating all the 100% sure stuff. AI is making the unsure stuff bigger. Humans live in the tiny crack between them.

That crack looks smaller and smaller, but it is worth more and more.

The hottest jobs in the future might not be “people who are good at using AI.” They might be “people who dare to take the blame for AI's answers.”

The most expensive skill might not be writing prompts. It might be writing “who pays if this goes wrong” plans.

So next time someone is scared “will AI take my job,” tell them to ask one easy question: Is my daily work 100% sure how to do it, or 99% sure but I still have to make the final call?

The answer decides everything.

AI can never replace people. Only automation can.

This is not a feel-good pill. This is the bloody truth I have seen with my own eyes in the last five years. It will also be the toughest and fairest filter for the next twenty years.

Accept it. Find your spot. Either write the 100% rules, or sign your name on that 0.01% call.

Every other road is already gone.