What skills do you need to survive for ai era?

As AI continues to evolve and potentially replace jobs across industries, a crucial question arises: What skills should we focus on to thrive in this new era? While AI can automate a wide range of tasks, it struggles in areas where human complex problem-solving abilities excel — domains rooted in intangible variables such as motivation, intention, culture, emotions, and social dynamics.

Skills like analytics, creativity, and critical thinking are essential — not only because AI struggles to replicate the nuanced authenticity of these human faculties, but also because these skills require a systematic approach to addressing problems that are subjective, unstructured, and context-dependent.

To fully answer this question, it is crucial to first understand AI’s capabilities and limitations about the skills demanded by today’s jobs.

AI’s capability

To begin with, let’s examine the skills where AI excels at replicating human capacities. One study links 10 AI applications — abstract strategy games, real-time video games, image recognition, visual question answering, image generation, reading comprehension, language modeling, translation, speech recognition, and instrumental track recognition — with 52 human abilities, such as oral comprehension, oral expression, inductive reasoning, and arm-hand steadiness. The study ranks the overlapping jobs as follows.1

Top Overlapping Skills:

• Language-focused tasks and information delivery, such as those performed by professors, rank at the top.

• Programming and writing also show strong alignment with AI’s capabilities, particularly in areas like automated coding and content generation.2 For instance, OpenAI’s o1-preview with AIDE scaffolding — achieves at least the level of a Kaggle bronze medal in 16.9% of competitions.3

Bottom Overlapping Skills:

• Physical activities like dancing require not only coordinated movement but also emotional expression and social interaction — elements that AI cannot replicate. Additionally, when we consider physical activities more deeply, I hypothesize that AI struggles with understanding or measuring intangible human experiences, such as motivation, intention, and the cultural context of real-world interactions.

• Science and critical thinking skills are strongly negatively associated LLM capability because these tasks often involve subjective reasoning, intuitive insights, and an understanding of context that AI is not yet capable of replicating (GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models).

Ultimately, it’s clear that AI is better suited to tasks involving structured information and defined processes, while human skills that rely on physical dexterity, creativity, and nuanced judgment and data flowing in the real world are less likely to be replaced soon.

The pitiful of AI capability

Despite the remarkable productivity improvements AI can bring to tasks, it still struggles with jobs that require a deeper understanding of context and critical thinking. Research has shown that even BCG consultants using AI were 12.2% more productive and completed tasks 25.1% faster than those who did not use AI, when tasks required context-based problem-solving, such as case interviews, consultants using AI were 19% less likely to produce correct solutions.4 In Singularity is Near, the definition of complexity is the minimum amount of meaningful, non-random, but unpredictable information needed to characterize a system of process. Based on this definition, as data and context complexity grow, the presence of random or irrelevant information may skew AI’s ability to make accurate decisions.

Similar findings were observed with recruiters who, when working with low-quality AI, adapted to interact more effectively with the tools. The key takeaway is that over-reliance on AI can weaken critical thinking and create a bias toward AI-generated solutions.5 While AI may analyze a candidate’s skill set, it cannot accurately assess cultural fit, motivation, or personal character — factors critical to team dynamics. Effective collaboration between humans and AI happens when experts apply their own knowledge to critically evaluate AI outputs.6

In the short term, although AI can boost productivity, we cannot rely on it entirely. Human evaluation, active engagement, and critical dialogue with AI are essential to achieving the best outcomes.

What Should We Focus on Learning?

First, we need to understand AI’s impact on the job market — whether it enhances current responsibility or replaces the role entirely. For example, Devon AI aims to mimic a junior software engineer, while Cursor AI focuses on enhancing the productivity of existing engineers.

I argue that we should prioritize learning skills that AI struggles to replace — particularly those that involve complex problem-solving where judgment is subjective, problems are unstructured, and context matters, with involvement of uncharacterizable information, such as motivation, emotions, and interpersonal dynamics. While AI can automate many tasks, it cannot address the nuanced challenges that arise in areas shaped by intangible variables mingled with the complexity of the existing problem context.

Take my experience building a Duolingo-like leaderboard, for example. AI might be able to generate the basic code for a leaderboard, but the design and strategic decisions behind it are much more complex. System design is an inherently unstructured problem — there is no single “correct” solution. Decisions about scaling, optimizing for user experience, and ensuring smooth operation in varying conditions involve judgment calls that depend on a deep understanding of the context.

While AI can handle technical elements like infinite scrolling or efficient ranking, the priority of implementation is based on constraints of engineering resources and feature priority. In addition, it cannot fully grasp how user motivation might change based on how the leaderboard is structured. Is a simple ranking system motivating useful, or would an other system — like a progress bar — better encourage user engagement? These questions require human insight, as they depend on understanding how different users may feel or be motivated by their experience, which AI cannot anticipate with the same depth.

Additionally, when designing a system, you must consider the social dynamics of your users. For example, how can a database design be considered that can allow more user competition in the future? Is it by leaderboard leagues? Daily vs weekly vs all-time competition? Or compete against your schools? These decisions are rooted in a deeper understanding of user behavior, business strategy, and the context — areas where AI lacks the insight needed to make meaningful choices. The broader thinking required combining product insights and system design — such as scaling, optimization, and user experience considerations — is more complex. Let alone as user needs and strategic goals evolve, so too must these considerations.

Conclusion

In conclusion, while AI is transforming industries, its limitations in contextual understanding, nuanced judgment, and addressing intangible human factors highlight the enduring value of human-centric skills like creativity and critical thinking. While machines may surpass human intelligence in the distant future, the immediate focus should be on areas where AI falls short — unstructured problem-solving, contextual analysis, and incorporating human elements like social dynamics — while leveraging AI’s efficiency to enhance human potential.

隨著人工智慧（AI）不斷發展，AI 可能取代各行各業的工作，這時候最關鍵的問題變成：我們應該專注於學習哪些技能，才能在這個新時代中茁壯成長？儘管AI能自動化各種工作，但它在複雜問題解決能力的領域仍然存在困難，特別是當這些領域涉及無形變量時。像是動機、意圖、文化、情感和社交關係。

分析、創造力和批判性思維這些技能是必須的 — 不僅是因為AI難以模仿這些人類能力的微妙真實性，還因為這些技能需要一種系統化的方法來處理主觀、他們無結構分散且依賴於當時特定的情境。

如果要充分回答關鍵問題，我們首先需要了解AI的能力和局限性，並與當今工作所需的技能進行比較。

AI的能力

首先，讓我們來看看AI在哪些技能上能夠成功複製人類的能力。一項研究將10個AI應用程序 — 抽象策略遊戲、實時視頻遊戲、圖像識別、視覺問題回答、圖像生成、閱讀理解、語言建模、翻譯、語音識別和樂器曲目識別 — 與52種人類能力（如口語理解、口語表達、歸納推理和手臂穩定性）進行了對比。該研究對重疊的工作進行了排名。

主要重疊技能：

以語言為主的任務和信息傳遞，例如教授所做的工作，排名居首。

編程和寫作也顯示出與AI能力的高度契合，尤其是在自動編碼和內容生成等領域。例如，OpenAI的o1-preview配合AIDE搭建框架 — 在16.9%的比賽中，至少達到了Kaggle銅獎水平。

最低重疊技能：

像舞蹈這樣的體育活動，不僅需要協調的動作，還需要情感表達和社交互動 — 這些是AI無法複製的元素。此外，當我們更深入地考慮實體互動時，我假設AI在理解或衡量無形的人類經歷（如動機、意圖以及現實世界互動中的文化背景）方面仍然存在困難。

科學和批判性思維技能與大型語言模型（LLM）的能力有很強的負相關，因為這些任務通常涉及主觀推理、直覺洞察以及對特定背景的理解，這是AI尚無法複製的（《GPTs是GPTs：大型語言模型對勞動市場影響潛力的初步觀察》）。

總體來說，很明顯，AI更適合處理涉及結構化信息和已定義過程的工作，而依賴於實體互動、創造力、判斷力以及現實世界中數據等人類技能則不太可能很快被取代。

AI的不足

儘管AI能顯著提高任務的生產力，它仍然在需要深入理解上下文和批判性思維的工作中遇到困難。研究顯示，即使是使用AI的BCG顧問，也比未使用AI的人更具生產力（提高了12.2%），並且完成任務的速度快了25.1%，但當任務需要基於上下文的問題解決時，例如案例面試，使用AI的顧問解決正確問題的可能性反而降低了19%。在 Singularity is Near 中，複雜性的定義是描述系統過程所需的最低有意義的、非隨機的、但不可預測的信息量。根據這一定義，隨著數據和問題背景的複雜性增長，隨機或無關的信息可能會負面影響AI做出準確決策的能力。

類似的發現也出現在招聘人員身上，當他們使用低質量的AI時，適應了與工具更有效的互動。關鍵的教訓是，過度依賴AI可能會削弱批判性思維，並產生對AI生成解決方案的偏見。儘管AI可以分析候選人的技能集，但它無法準確評估文化契合度、動機或個人性格 — 這些對團隊動態至關重要的因素。研究發現人類和AI之間的有效協作發生在專家運用自身知識來批判性地評估AI輸出的時候。

短期內，儘管AI可以提高生產力，我們不能完全依賴它。人類的評估、積極參與和與AI的批判性對話對實現最佳成果至關重要。

我們應該專注於學習什麼？

首先，我們需要了解AI對勞動市場的影響 — 它是增強現有職責還是完全取代這一角色。例如，Devon AI旨在模擬初級軟件工程師，而Cursor AI則專注於提升現有工程師的生產力。

我認為，我們應該優先學習AI難以取代的技能 — 特別是那些涉及複雜問題解決的技能，這些問題的判斷是主觀的、無結構的，並且依賴於特定情境，涉及無法量化的信息，如動機、情感和社交動態。儘管AI能夠自動化許多任務，但它無法解決在受無形變量影響的領域中出現的微妙挑戰，這些變量與現有問題背景的複雜性交織在一起。

以我在建構類似 Duolingo 排行榜的經驗為例。AI可能能夠生成排行榜的基本代碼，但其設計和背後的戰略決策則要複雜得多。系統設計本質上是一個無結構的問題 — 沒有單一的“正確”解決方案。關於擴展、優化用戶體驗和確保在不同條件下平穩運行的決策，都涉及到需要深刻理解公司背景及策略的判斷。

儘管AI可以處理像無限滾動或高效排名這樣的技術元素，但實現的優先級是基於工程資源的限制和功能優先順序。此外，它無法完全理解用戶動機如何根據排行榜的結構而改變。簡單的排名系統是否有效激勵用戶？還是另一種系統，如進度條，更能促進用戶參與？這些問題需要人類的洞察力，因為它們依賴於理解不同用戶如何感覺或受到他們的體驗激勵，而這是AI無法以同樣的深度預測的。

此外，當設計系統時，您必須考慮用戶的社交關係。例如，如何設計一個數據庫，能夠在未來允許更多的用戶在排行榜上競爭？要不要做排行榜聯盟？每日、每週還是全時競爭？還是與學校競爭？這些決策根植於對用戶行為、商業策略和背景的深入理解 — 這是AI無法提供的有意義的選擇。更廣泛的思維結合產品洞察和系統設計 — 如擴展、優化和用戶體驗考量 — 是更複雜的。更不用說，隨著用戶需求和策略目標的演變，這些考量也必須隨之調整。

結論

總之，儘管AI正在改變各行各業，它在理解上下文、判斷能力和了解無形人類因素方面的局限性突顯了人類中心技能（如創造力和批判性思維）的持久價值。儘管機器在遙遠的未來可能超越人類智慧，但當前的重點應該放在AI無法勝任的領域 — 無結構問題解決、情境分析，以及融入社交動態等元素 — 同時利用AI的效率來提升人類潛力。

1 Occupational Heterogeneity in Exposure to Generative AI

2 GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models

3 MLE-BENCH: EVALUATING MACHINE LEARNING AGENTS ON MACHINE LEARNING ENGINEERING

4 Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality

5 Falling Asleep at the Wheel: Human/AI Collaboration in a Field Experiment on HR Recruiters

6 To Engage or Not to Engage with AI for Critical Judgments: How Professionals Deal with Opacity When Using AI for Medical Diagnosis