Name: AI機器人竟然開始面試了！我親自體驗了！ (AI bots are now conducting job interviews. I tried it)
Uploaded: 2026-03-11T16:33:25.000Z
Duration: 9 min 14 s

What challenges did you encounter and how did you address them?

你遇到了什麼挑戰，又是怎麼解決的呢？

Sure, that's a couple different questions, so let me start with the first one when it comes to I hear you got it.

沒問題，這有兩個不同的問題，所以我就先回答第一個，關於你說的「我聽說你搞定了」。

程度副詞

I'm being interviewed by an AI and it's just as unsettling as you'd expect.

我竟然要被 AI 面試，這感覺就像你預期的那樣，有點讓人不安。

With AI everywhere in today's job search, it seems inevitable that we'd eventually get here.

在現今的求職過程中，AI 無所不在，我們最終走到這一步似乎是不可避免的。

But still, AI avatars running job interviews is pretty next level.

不過，由 AI 頭像來進行面試，這真的太超現實了。

It seems like the process is on the rise.

Wanted to understand why and whether they can ever be helpful.

我想了解原因，以及它們到底有沒有幫助。

So you made it to the interview stage of your job search.

所以你已經進入了求職的面試階段了。

但劇情出現了意想不到的轉折。

Before you meet with an actual human, you have to make it past an interview with an AI.

在你跟真人面試之前，你得先通過AI的面試。

A ton of companies apparently, from Meta and Netflix to mastercard and Domino's, have started to use these AI interviewers as part of their initial screening processes.

很多公司，像是Meta、Netflix、Mastercard和Domino's，似乎都開始使用這些AI面試官來做初步篩選。

The the value add here, according to the creators of these AI tools, is that it allows some companies to talk in some way, shape or form to virtually everyone who applies for a role instead of just some of those people.

這些AI工具的創作者認為，這樣做的好處是讓公司能夠跟幾乎所有申請者進行某種形式的對話，而不是隻跟部分人談。

Most of our customers are unable to get humans out to about 95% of applicants.

我們的大部分客戶無法接觸到大約95%的求職者。

And you hope that the lucky 5% that you somehow picked out is the 5% that you should in fact be talking to.

而你只能寄望你幸運挑出的那5%，確實是你應該談的對象。

All of them make the claim that these tools operate with less bias and prejudice because after all, you're talking to an emotionless bot about why you deserve the job.

他們都聲稱這些工具的運作比較沒有偏見，畢竟你是在跟一個沒有感情的機器人談論你為何值得這份工作。

AI interviews usually look like a one on one video call with an AI agent.

AI面試通常看起來就像是跟一個AI代理進行一對一的視訊通話。

I tried out one from Codesignal, which was a static photo of an AI avatar.

我試用了 Codesignal 的一個，它是一個 AI 頭像的靜態照片。

Kind of like when someone turns off their camera on Zoom and you just see their headshot.

有點像是有人在 Zoom 上關閉鏡頭，你只能看到他們的頭像。

Another one, Humanly, was a fully built out AI avatar designed to look like a human.

另一個是 Humanly，它是一個設計成真人樣貌的完整 AI 頭像。

So you go through a 30 minute, 45 minutes.

所以你進行一個 30 分鐘，45 分鐘的訪談。

Sometimes it's even 15 minute interview companies can decide the length, they can decide the tone.

有時候甚至只有 15 分鐘，公司可以決定長度，他們可以決定語氣。

Could you walk me through your typical workflow from the initial deep dive conversations to the final audience friendly product.

你能帶我走一遍你從最初的深入對話到最終的觀眾友善產品的典型工作流程嗎？

Compared to the other tools I tried, Humanly's AI avatar asked pretty good questions and we had an okay back and forth, but

跟我試用的其他工具相比，Humanly 的 AI 頭像問的問題還不錯，我們也有一些來回的互動，但是

the appearance of the avatar itself was approaching the uncanny valley.

頭像本身的外觀已經接近恐怖谷了。

對我來說，這可能會讓人分心。

I had to look away from her face the whole time or I wouldn't have been able to focus on my answers.

我整個過程都得避開她的臉，不然我根本沒辦法專心回答。

Humanly says their AI interviewer is unique because it gets candidates talking more.

Humanly 表示，他們的 AI 面試官很特別，能讓應徵者講更多話。

They'll average about 200 words per response to the avatar, which is more than a human doing a phone screen.

他們平均每段回應 AI 頭像大約有 200 個字，比真人進行電話初試還多。

Then there was code signal which I did the demo with using a photo of an AI avatar, but

然後是 Codesignal，我用 AI 頭像的照片做了示範，但是

not one conversing with me in video form.

How would you describe your technical fluency around AI and the hardware that powers it?

你怎麼形容你對 AI 和驅動它的硬體的技術熟練度？

And how do you stay current on that side?

你又是怎麼在這方面保持更新的呢？

I definitely am more comfortable with the software rather than the hardware.

我對軟體絕對比對硬體更熟悉。

Codesignal was the best one I tried because the interview questions were the most realistic.

Codesignal 是我試過最好的，因為面試問題最貼近真實情況。

We do not currently look at how you set things mostly what you said, so really focused on the words that you use to describe and answer the questions.

我們目前不太看你怎麼設定，主要看你說了什麼，所以真的專注在你用什麼詞語來描述和回答問題。

It even had some of the newsroom lingo down, like asking about my toughest experience with a comms person who didn't like the outcome of an article and how I handled it.

它甚至還學會了一些新聞界的術語，像是問我跟一位不喜歡報導結果的公關人員打交道時，最艱難的經驗是什麼，以及我如何處理。

in practice it did allow me to answer how I would normally answer any other type of interview questions, but

實際上，這讓我能夠像回答其他類型的面試問題一樣回答，但是

I still wouldn't want to do an interview this way and just making sure that we're like framing it in the right.

我還是不想用這種方式進行面試，只是要確保我們是以正確的方式來呈現。

When you get that kind of pushback from a source or.

當你遇到那種來自消息來源或。

But under the surface of any AI tool there are questions about the effect of a technology that scales so quickly and the subtle or not so subtle ways it could impact the people it's analyzing.

但在任何 AI 工具的表面之下，都存在著關於這項快速擴張的技術的影響，以及它可能對被分析對象產生的微妙或不那麼微妙的影響的問題。

In theory or according to these companies pitches at least, AI interviews should reduce bias and level the playing field.

理論上，或者至少根據這些公司的說法，AI 面試應該能減少偏見，讓競爭更公平。

But that's only if the systems themselves are bias free and that is a standard that is impossible to achieve.

但前提是這些系統本身沒有偏見，而這是一個不可能達到的標準。

The reason for that is because these systems are trained on large swaths of the Internet and obviously large swaths of the Internet are human made and they contain sexism, racism and a ton of other biases.

原因在於這些系統是在網路上大量的資料上訓練出來的，而網路上的資料顯然大部分是由人類創造的，其中包含了性別歧視、種族歧視以及大量的其他偏見。

What's less clear is exactly how these AI algorithms work to evaluate and rank candidates.

比較不清楚的是，這些AI演算法究竟是如何評估和排序求職者的。

It doesn't help that with a lot of large AI models, parts of them are so called black boxes, but

而且，很多大型AI模型的某些部分被稱為「黑盒子」，這點也無濟於事，

meaning that we don't always know how or why the AI systems make the decisions they do.

也就是說，我們並不總是知道AI系統是如何或為何做出決定的。

And many times the companies can't fully explain it either.

而且很多時候，公司也無法完全解釋。

One company, Eightfold, is currently being sued by two plaintiffs who want greater transparency into these black boxes.

其中一家公司 Eightfold，目前正被兩名原告起訴，他們希望對這些黑盒子有更多的透明度。

According to their website, Eightfold works with companies like Activision, Salesforce, Dolby and more.

根據他們的網站，Eightfold 與 Activision、Salesforce、Dolby 等公司合作。

I actually got to try out Eightfold's software and out of the three I tried, it did feel the most robotic and least human.

我 actually 試用了 Eightfold 的軟體，在我試用的三個軟體中，它確實感覺最像機器人，最不像真人。

Like I Will begin asking you questions shortly.

例如，我很快就會開始問你問題了。

I did appreciate that the default setting was just a pulsating circle for me to talk to, so it wasn't distracting from an Uncanny Valley standpoint, at least.

我其實很欣賞預設設定只是一個讓我對話的脈動圓圈，至少從恐怖谷效應來看，它不會讓人分心。

Personally, pretty passionate about the way technology is covered.

我個人對科技報導的方式很有熱情。

In your view, what are the most important AI trends to cover in 2026?

在你看來，2026年最重要的 AI 趨勢有哪些？

When we spoke to the company, they said their algorithm sets them apart.

我們跟公司談的時候，他們說他們的演算法讓他們與眾不同。

And they kept talking about their consistent approach, which they say leads to less bias.

他們一直強調他們一致的做法，說這樣可以減少偏見。

Every candidate will be asked the same question, same rating score, very consistent during the interview, and

每位應徵者都會被問到相同的問題，相同的評分標準，面試過程中非常一致，而且

So based on the job description, the response, what was that level of, I would say, competency of that capability from 0 to 5 being, you know, mastery of that skill to not aware.

所以根據職位描述、回覆，以及從 0 到 5 的能力等級，也就是從精通到完全不知道，你認為是怎麼樣的？

But the plaintiffs in the lawsuit against Eightfold said that in scoring candidates, these tools should operate more like credit agencies, which are required to report how sensitive consumer data is handled and evaluated, as well as inform consumers of their scores.

但針對 Eightfold 的訴訟原告表示，在評分應徵者時，這些工具應該更像信用機構，需要報告敏感的消費者數據如何被處理和評估，並告知消費者他們的得分。

We asked Eightfold for comment via email and they responded with, we use information applicants choose to submit and data authorized by our customers under contract.

我們透過電子郵件向 Eightfold 尋求評論，他們回覆說，我們使用應徵者選擇提交的資訊，以及我們客戶根據合約授權的數據。

We do not scrape social media and the like to assess an applicant's fit for a specific role.

我們不會抓取社群媒體等來評估應徵者是否適合特定職位。

During our interview prior to this statement, they also expressed confidence in their bias procedures, saying they haven't had to change anything.

在我們這次聲明之前的訪談中，他們也對他們的偏見處理程序表示信心，說他們沒有需要做任何修改。

We haven't had anything that we had to go back and change because there was bias in the system.

我們沒有任何需要回頭修改的地方，因為系統裡有偏見。

When it comes to our algorithm and the integrity of our algorithm, we haven't had to go back and course correct the outcome of the lawsuit could change that and it'll likely have a ripple effect on Eightfold's competitors.

談到我們的演算法和演算法的完整性，我們沒有需要回頭修正，訴訟結果可能會改變這一點，而且很可能會對 Eightfold 的競爭對手產生連鎖反應。

We asked codesignal and humanly how they handle the bias in their systems.

我們詢問 codesignal 和 humanly 他們如何處理系統中的偏見。

All three companies said they conduct bias audits or a way to test out the bias in AI systems and how they treat minorities and different groups of people.

這三家公司都表示，他們會進行偏見審計，或以某種方式測試 AI 系統中的偏見，以及它們如何對待少數族裔和不同群體的人。

We've been audited by a third party for several years, so we went in our early days and these are audits that are done yearly.

我們已經由第三方審計了好幾年，所以我們在早期就開始了，這些審計都是每年進行的。

We have an industrial and organizational psychology team internally as well as subject matter experts, which we call the talent science team.

我們內部有一個工業與組織心理學團隊，以及我們稱為人才科學團隊的專家。

And they conduct regular bias audits in partnership with customers.

他們會與客戶合作進行定期的偏見審計。

And the bias audits look at all sorts of different cuts, right where we would look at performance on the AI interview and the rate at which people get the job.

偏見審計會檢視各種不同的指標，像是 AI 面試的表現，以及人們獲得錄用的比率。

Ethics and security have been part of just the way we build.

道德和安全一直是我們建置方式的一部分。

Whether it's SoC2 or ISO certification for security, privacy and all of the AI controls, the executives we spoke with said that candidates are only graded on what they say in their actual responses.

無論是安全、隱私和所有 AI 控制的 SoC2 或 ISO 認證，我們訪談的執行長們都表示，候選人僅根據他們在實際回應中的內容來評分。

But that means that just like when an AI parses your resume, cover letter, or another part of your job application, keywords here are important.

但這意味著，就像 AI 解析你的履歷、求職信或其他求職申請的一部分一樣，關鍵字在這裡很重要。

So if this is the future of job interviews, I'm apprehensive.

所以，如果這就是未來求職面試的樣貌，我感到很不安。

It's already such a tough process to apply for a job, to feel like a number in the system already, to feel nervous, to have to talk about your experience and why you're the best fit.

求職已經是個艱難的過程了，感覺自己已經是系統中的一個數字，感到緊張，還要談論你的經驗和你為何是最適合的人選。

And I just can't see it being easier to do all of that with an AI system on the other end rather than a human.

我實在無法想像，在另一端是 AI 系統而不是人類，會讓所有這些事情變得更容易。

So what should you do if you're faced with an interview like this?

那麼，如果你遇到這樣的面試，你該怎麼辦？

It's important to speak clearly, use keywords and metrics when you can your responses, and

清楚地表達自己、在回答中盡量使用關鍵字和數據，

maybe ask a question or two at the end just to be on the safe side.

並且為了保險起見，在最後問一兩個問題。

I unfortunately don't have those details on hand in a reliable way since this is a first round screen.

很可惜，我手邊沒有可靠的這些細節，因為這是第一輪的篩選。

You know, one thing that really surprised me in doing all these interviews was that these companies said that a lot of people tell them that they prefer talking to an AI interviewer rather than a human.

說到這，我在進行這些面試時，有一件事真的讓我很驚訝，那就是這些公司說，很多人告訴他們，他們寧願跟 AI 面試官談話，也不願跟真人談。

Let us know in the comments if that's true or if you think that's bogus.

如果你也是這樣，或者你覺得這很扯，都可以在留言區告訴我們。