Understanding becomes a new bottleneck: Cognitive inflation in the AI era
AI did not eliminate bottlenecks, but only moved it from the "generating end" to the "understanding end", creating cognitive inflation.
production costs zero, but digestion is becoming extremely expensive
In May 2024, an HN hot post asked: "Why do I read 100 AI-generated analyses every day, but feel like I haven't learned anything?" The most favorable reply below is: "Because your understanding is being diluted." This is not a joke. In the past year, the amount of text generated every day around the world has increased from 1.2 exabytes in 2022 to 3.5 exabytes-almost entirely driven by AI. In the same period, the average human reading speed still stayed at 200-300 words per minute, and the depth of understanding even dropped due to information overload. We are experiencing a quiet period of cognitive inflation: AI has pushed the cost of generation to near zero, but the value of understanding skyrocketed, and the human brain has not kept up with this currency devaluation.
1. The nature of inflation: information money overflows
In economics, the root cause of inflation is currency overload. The same is true for the information ecosystem in the AI era-when the marginal cost of the generator approaches zero, information currency is supplied in an unlimited amount, but understanding (i.e.,"information purchasing power") does not increase simultaneously. This directly leads to three consequences:
1.1 Dilution of the actual value of information In 2023, I participated in the product review of a knowledge-based payment platform. Data shows that the average user completion rate has dropped from 42% in 2021 to 18% in 2023, but the number of courses has increased by 3.7 times. More importantly, users '"depth of understanding rating" for the course (measured through post-test questionnaires) dropped from 7.1 to 5.3. Platform PM concluded: "What we sell is not knowledge, but the illusion of knowledge. "This illusion comes from the massive amount of content generated by AI-when everyone can draw" professional-level "illustrations with Midjourney and write" in-depth "analysis with Claude," professionalism "and" depth "themselves depreciate.
1.2 Relative scarcity of understanding In October 2023, a survey of Silicon Valley engineers showed that with the assistance of AI, code generation speed increased 2.3 times, but the code review pass rate dropped by 31%. The reason is simple-AI can generate code, but there is no guarantee that developers truly understand the design intentions and potential risks of the code. Understanding has become a new bottleneck. This scarcity is directly reflected in salaries: FAANG will pay 28-42% more annually in 2024 for engineers who "can independently understand and optimize ai-generated code," according to Levels.fyi Understanding is becoming the new hard currency.
1.3 Accumulation of cognitive liabilities The most insidious inflationary effect is perceived debt . When you quickly consume AI-generated content, your brain develops a "false understanding"-you think you understand, but in fact you just remember the conclusions, not the logical chain. This kind of debt can explode at critical moments. For example: An investor in - told me that he had read more than 300 AI-generated industry reports in 2023 and stumbled on a key investment because "the logic in the report seemed self-consistent, but completely ignored the real constraints of the supply chain." A doctor friend at - found that interns are increasingly relying on AI diagnostic tools, but "when I asked them why they chose this treatment plan, they would only say 'AI recommended' and could not tell the pathological mechanism behind it."
The essence of cognitive debt is: AI helps you save generation time, but puts you in debt of understanding. And the interest on this debt will come to you when you need to think most deeply.
2. Why can't understanding keep up with the speed of generation?
2.1 Hardware limitations of the brain The information processing speed of the human brain has changed little in the past 100,000 years. According to cognitive science research, human working memory capacity is approximately 4±1 "chunks", and understanding complex information requires maintaining and manipulating multiple chunks in the brain at the same time. Content generated by AI is often extremely information-dense, but the human brain cannot process so many chunks simultaneously. For example:
- An AI-generated financial analysis may contain 15 key data points and 3 logical chains, but the human brain can only remember 3-4 of them at the same time when reading.
- A piece of AI-generated code may have 5 layers of nested logic, but the human brain can only understand it layer by layer during debugging and cannot process it in parallel.
It's like using a 56K dial-up Internet to download 4K videos-the bandwidth doesn't match.
2.2 The "Matthew Effect" of Understanding Understanding does not increase linearly, but shows a clear Matthew effect: People with strong - understanding can more quickly identify the quality of AI-generated content, thereby screening out truly valuable information and further improving understanding. People with weak understanding of - will be overwhelmed by low-quality AI content, forming a vicious cycle of "bad money driving out good money."
This effect is particularly evident in the field of education. A study of American universities found that after the popularization of AI-assisted learning tools, the scores of students in the top 20% of GPA increased by 12%, and the scores of students in the bottom 20% dropped by 8%. The reason is that the former can effectively use AI to assist understanding, while the latter is misled by the "pseudo-knowledge" generated by AI.
2.3 The "Time Tax" of Understanding Understanding complex information takes time-a cost that AI cannot compress. For example:
- Understanding a technical paper requires 3-5 hours of in-depth reading and thinking.
- understands that a business model requires repeated scrutiny of its cash flow and competitive barriers.
- Understanding the design intent of a piece of code requires line-by-line debugging and scene simulation.
AI can generate a "seemingly in-depth" paper analysis, business plan or code comment in 30 seconds. This speed difference leads to a paradox: The more we rely on AI to generate content, the less time there is to truly understand the content .
3. Steelman: Opposing views and responses
Opposition 1:"AI can assist understanding, such as generating summaries and drawing mind maps" It is true that AI can help generate summaries, keywords or mind maps, but this is only a shallow understanding aid, not a deep understanding. The essence of understanding is to establish connections between knowledge, and this connection requires the active thinking of the human brain. Summaries generated by AI may help you remember conclusions, but they cannot help you understand "why." For example:
- AI can tell you that "this code implements the X function" but cannot explain "why this algorithm was chosen over another."
- AI can summarize the conclusions of a paper, but it cannot help you evaluate "under what circumstances this conclusion does not hold true."
Counterpoint 2:"Understanding can be improved through training, and AI only accelerates the process" Understanding can indeed be trained, but training requires high-quality input and feedback . The content generated by AI often lacks depth and diversity and is prone to the "echoing wall effect"-AI continues to strengthen your existing cognitive prejudices rather than challenge your boundaries of understanding. For example:
- If you ask AI "Why is capitalism good", it will give you a bunch of arguments in support of capitalism, and will not proactively provide counterexamples.
- If you ask AI to explain a technical concept, it will give a standard answer without telling you the limitations or controversial points of the concept.
Counter-view 3:"Cognitive inflation is temporary and humans will adapt" There have been similar adaptation processes in history, such as the gradual increase in literacy rates after the invention of printing. But what's different this time is:
- Speed difference : It took 300 years for printing to become popular, while the popularity of AI-generated content only took 3 years.
- Complexity Differences : Printing only reduces the cost of information dissemination, while AI also reduces the cost of generation and dissemination, and the content generated is much more complex than printed matter.
- Missing feedback mechanism : In the printing era, the spread of misinformation requires manual correction (such as church censorship), while in the AI era, misinformation can be automatically amplified by AI (such as fake news generated by AI being forwarded by other AI).
4. Cross-border analogy: Understanding inflation versus housing bubble
Understand the striking similarities between inflation and the housing bubble:
| dimension | real estate bubble | Understand inflation |
|---|---|---|
| Currency overissued | Easy monetary policy causes funds to flood into the housing market | AI reduces generation costs, resulting in information overload |
| Asset mismatch | Houses are regarded as investment rather than residential goods | Information is viewed as a consumer good rather than an object of understanding |
| Bubble indicator | The price-to-income ratio is much higher than a reasonable level | Information consumption far exceeds understanding ability |
| Rupture trigger point | Interest rate increases or policy controls | Critical decision-making mistakes (e.g. investment, medical, legal) |
| Consequences | Assets shrink, economic recession | Cognitive debt explodes and decision-making quality declines |
The most critical similarity lies in asset mismatch: in the real estate bubble, the residential function of a house was concealed by financial attributes; in understanding inflation, the "understanding value" of information was concealed by "consumption pleasure." Both lead to a systemic mismatch of resources-the former is a mismatch of capital, the latter is a mismatch of attention.
5. Understanding the social consequences of inflation
5.1 Universalization of knowledge anxiety In 2023, a survey of knowledge workers aged 18-35 showed that 68% said they "feel overwhelmed by information every day" and 42% said "the more they learn, the more anxious they become because they always feel that they don't know enough." This anxiety is not due to lack of knowledge, but due to the relative devaluation of understanding. When AI can instantly generate an "expert-level" analysis, the sense of value of human experts is diluted.
5.2 Systematic decline in decision-making quality Understand that the most dangerous consequence of inflation is a decline in the quality of decision-making . When decision makers cannot deeply understand what AI-generated content, they will:
- relies too much on AI recommendations (such as investment, medical, legal decisions).
- cannot identify AI errors (such as fake data generated by AI, logic vulnerabilities).
- falls into the trap of "pseudo-understanding"(thinking you understand it, but actually you don't).
In 2023, a large hedge fund lost $1.2 billion in a quarter because of its excessive reliance on AI-generated trading strategies. A post-event review found that the AI-generated strategies performed well in backtesting, but ignored the real constraints on market liquidity during actual trading.
5.3 Solidification of Levels of Understanding Understand that inflation will intensify the solidification of social classes: QKPFX25 People with strong QK understanding can use AI to improve efficiency and further widen the gap with ordinary people. QKPFX26 People with weak QK understanding will be overwhelmed by low-quality content generated by AI and fall into a "cognitive trap."
This can create a vicious cycle: People with strong understanding are getting stronger, and people with weak understanding are getting weaker. Instead of narrowing the gap, AI has become an amplifier of the gap.
6. How to deal with cognitive inflation?
6.1 Redefinition of "understanding" In the AI era, understanding is no longer about "remembering conclusions" but:
- can identify logical vulnerabilities in AI-generated content.
- is able to assess the reliability and limitations of information.
- can combine information with actual scenarios to make correct decisions.
This means we need to develop a new skill: critical understanding. For example:
- When AI generates a business plan, you need to ask: "Under what circumstances will this model fail? "
- When AI generates a piece of code, you need to ask: "What are the boundary conditions for this piece of code? "
6.2 Limit information intake and strengthen understanding and output Instead of consuming 100 AI-generated articles every day, it is better to deeply understand 1 classic document. The improvement of understanding does not depend on the amount of information, but on the depth of thinking. For example: After - finished reading a paper, it tried to repeat the core logic in its own words. After - looks at a piece of code, it attempts to manually simulate its execution process. After - listens to a point, it tries to find potential counterexamples.
6.3 Establishing a personal "filter of understanding" In an era of information overload, the core of understanding is the ability to screen. You need to establish a personal filtering mechanism, such as:
- For technical content: preference is given to articles supported by actual code or data.
- For business content: Preference is given to reports with cash flow models or case studies.
- For academic content: Preference is given to peer-reviewed papers.
The core of this filter is: Reject shallow "pseudo-understanding" and pursue deep true understanding .
7. Epilogue: A reassessment of the value of understanding
The biggest cognitive illusion in the AI era is that we think that an improvement in generation efficiency is equivalent to an improvement in understanding efficiency. But in fact, AI just moves the bottleneck from the generation end to the understanding end . Understanding is becoming a new scarce resource, and this scarcity will redefine the value of knowledge.
In the future, the most valuable thing is not "what you know" but "what you understand." Those who can deeply understand AI-generated content will become new intellectual aristocrats; while those who can only consume AI content will become cognitive proletarians. The endgame of this cognitive inflation depends on how we redefine the value of understanding.
Golden sentence :
- "AI makes generation cheap, but understanding priceless. "
- "In an era of information overload, understanding is not the knowledge you have, but the depth of thinking you can control. "
The motif echoes : This cognitive inflation is essentially the price of efficiencyism . We pursue the ultimate in generating efficiency, but ignore the bottleneck of understanding efficiency. When the cost of the generation end approaches zero, the value of the understanding end will be infinitely amplified. Humanity is standing at this crossroads of value reassessment.