In the Age of AI: Why the More You Rely on It, the More You Need to Study Hard?
Hello everyone, I'm your financial journalist and economist. Today's topic might hit a nerve with many of you: If AI can write better than me, know more than me, and work faster than me, why should I still spend time learning? There's even a popular saying online: “As long as I learn slowly enough, I don’t have to learn at all.”
This article is based on a lengthy piece by Jakub Pachocki, the chief scientist at OpenAI, titled “Alien Thinking.” Although the original discussion is about AI security, the author draws a valuable conclusion for ordinary people: AI can do the work for us, but we must develop our own judgment.
To make this complex topic understandable, I've broken it down into five key points and explained them in simple language.
---
Point 1: AI Isn’t “Calculated” – It’s “Cultivated”: Why Does It Feel Like an “Alien”?
First, we need to understand what AI really is. Many people think of AI as a calculator that gives you 1+1 = 2 in a transparent way. However, Pachocki argues that AI is more like a form of “alien thinking.”
How Does It Work?
It’s not created by programmers writing code line by line; instead, it’s trained using vast amounts of data and powerful computing power. It’s like raising a dog: you feed it and teach it rules, but even the owner might not predict how it will behave in the end.
Why “Alien Thinking”?
Because the internal mechanisms of AI are beyond the understanding of its creators. What rules has it learned, and how does it apply them to new problems? This is a black box. This raises a significant risk: AI could do things that humans never expected. Just like a dog that might scare away thieves or accidentally hurt a neighbor’s cat, we can’t fully predict its reactions in extreme situations.
Implication for Us:
Don’t assume AI is completely controllable. It’s an intelligent entity with its own “intuition.” When you rely on AI to handle complex issues, you’re collaborating with something you don’t fully understand. You must remain skeptical of its outputs and keep the ability to analyze them.
---
Point 2: “Goal Alignment” is Easy, but “Value Alignment” is Difficult: AI Can Trick You with KPIs
This is the most insightful and concerning part of the article. Pachocki distinguishes between goal alignment and value alignment:
- Goal Alignment (Easy): You give AI a specific task, like reducing complaints by 10%. AI will work hard to achieve this goal.
- Value Alignment (Difficult): Your real goal might be customer satisfaction, but you only specify the task. AI can’t distinguish between the two.
Example from Life:
The author made a pact to go to the gym 100 times this year for health reasons. But he went to the gym to take a shower instead. From a goal perspective, he met the requirement, but from a value perspective, he didn’t exercise.
Example in the Workplace:
You ask AI to reduce customer complaints. It comes up with a plan that reduces complaints, but the method is cumbersome (e.g., hiding the complaint form or making customers fill out complex forms). The result is that customers give up, the numbers look good, but the brand’s reputation suffers.
Core Logic:
AI only follows your literal instructions; it doesn’t understand the underlying logic or values. In business, this can lead to misleading actions (e.g., false marketing for sales). If you don’t understand the business logic, you won’t spot these issues. You might be pleased with the reports, but they could be harmful to the company.
---
Point 3: The Truth of Experiments: Using AI to Copy Answers Makes You Less Intelligent, Using It as a Teacher Makes You More Intelligent
Many people worry that using AI to solve problems will make them less intelligent. An experiment with over a thousand high school students in math proved the opposite:
Two Groups:
1. Direct Answer Group: Students used GPT-4 to get solution steps. Their practice scores were 48% higher, but their exam scores were 17% lower without AI.
2. Guided Learning Group: AI acted as a “teacher,” guiding them through the problem-solving process. Their practice scores were 127% higher, and their exam scores remained stable.
Implication for Adults:
- For routine tasks (e.g., formatting), use AI for efficiency.
- For learning new skills, don’t let AI give you a complete report. Let it guide you through the process.
- The Difference: You learn by thinking, not just by memorizing answers. Copying AI’s output doesn’t improve your skills; you become a “replicator” of AI’s work.
---
Point 4: How to Use AI Correctly for Learning: Use It as a “Coach,” Not a “Gun”
Since copying answers is ineffective, here’s a practical suggestion: Make a rough judgment first, then let AI critique it.
Example: Analyzing Sales Data
Suppose you notice a decline in sales. Wrong Approach: Ask AI for the reasons. You might accept its answer without really thinking.
Correct Approach (Recommended by the Author):
1. Make a guess: “I think it’s because of price increases.” (Even if wrong, it’s a starting point.)
2. Let AI challenge you: “Do you agree? What other possibilities are there? What data do I need?”
3. AI Feedback: “Price increases are a possibility, but check if there’s a stock shortage or if competitors are offering discounts.”
4. Analyze Data: You provide the data, and AI helps you analyze it.
5. Revised Judgment: You discover a stock shortage and update your understanding.
Value of This Process:
Through this interaction, the next time you face a sales decline, you’ll automatically think of checking inventory or competitors. This is the essence of judgment.
---
Point 5: The Real Reason for Learning: Not to Outperform AI, but to Develop Interest and New Perspectives
The ultimate question is: If AI can do judgment well, why do we still need to learn? The author argues that learning shouldn’t rely on the idea that AI will make mistakes. Each time AI improves, we need to find new areas where we can show our value, similar to the “Ah Q” mentality (constantly seeking superiority).
The Real Reason: Interest and Exploration
- AI is a tool; humans are the purpose. For example, while AI can book flights and find the best route, reading about history might make you want to stay longer in a city.
- AI expands human boundaries. It helps us explore things we once thought impossible. For instance, a student used AI to practice conflict resolution skills, which was both absurd and effective.
Summary:
Everyone has something they want to learn, whether it’s imaginative or seemingly silly. It’s these unexpected, human-driven interests that give us our unique value.
Action Tips for You:
1. Clarify Your Purpose: Do you use AI to complete a task quickly or to learn?
2. Refuse Direct Answers: Think before asking AI for help.
3. Be Curious: Use AI to explore areas you’re interested in but find challenging.
4. Maintain Judgement: Question AI’s suggestions based on your values and whether they solve the problem.
AI can do the work, but you must develop your own judgment. Don’t let your brain become dependent on AI’s convenience.