Summary of Key Points
This article highlights the impact of the rapid development of AI on the human system of knowledge production and sharing: AI-generated content has already surpassed that of humans, with the potential to be 1000 times more extensive in the future. Public knowledge platforms like Stack Overflow are declining due to the rise of AI; knowledge sharing is facing a “tragedy of the commons,” as experts are becoming less willing to contribute. Training AI using its own generated content can lead to performance degradation, similar to the phenomenon of inbreeding. In the long run, humans may lose the ability to oversee AI, potentially threatening the vitality of collective wisdom. The article emphasizes that AI cannot replace humans in actively creating new knowledge, and it is essential to maintain human involvement in the process of knowledge generation.
1. The “Silence” of Stack Overflow: AI Has Taken Its Users and Its “Brain”
Stack Overflow once served as a collective brain for programmers—during late-night debugging sessions, searching for error messages would often yield answers from five years ago, accompanied by dozens of votes. However, with the emergence of ChatGPT, the number of monthly questions on Stack Overflow has dropped from 200,000 to less than 50,000 (returning to the levels seen when it was first launched in 2009). The reason is simple: people now prefer to ask AI. Even worse, the most skilled and reputable users have left the platform, leading to an increase in the proportion of unanswered questions from 19.9% to 25.7%. Nowadays, some people even use Stack Overflow to fix mistakes made by AI—AI-generated answers may seem correct but are not entirely accurate, necessitating further verification on the platform. Without new questions, old answers become outdated (for example, solutions from five years ago may not work for modern software frameworks), and AI cannot predict new issues, turning the platform into a repository of outdated knowledge.
2. The “Tragedy of the Commons” in Knowledge Sharing: Why Are Experts Unwilling to Share?
The economic concept of the “tragedy of the commons” describes a situation where individuals act selfishly (seeking maximum benefit while minimizing costs for others), leading to the degradation of a shared resource. Knowledge platforms were designed to operate on a “positive cycle”: by contributing knowledge, users gained reputation (through votes and rankings), which in turn encouraged more contributions from the community. However, AI has disrupted this cycle by providing nearly perfect answers at zero cost. As a result, experts’ motivations for sharing—reputation and feedback—are diminishing, and many have given up. Valuable knowledge is now confined to paid communities or corporate intranets, reducing human-generated content and leading to a decline in AI’s training data quality, resulting in a downward spiral.
3. AI’s “Inbreeding”: Reusing the Same Content Leads to Degradation
A 2024 paper published in Nature suggests that repeatedly training AI using its own generated content can lead to performance degradation. Rare early-stage knowledge is lost, and subsequent outputs become more monotonous and error-prone. This phenomenon is akin to “inbreeding” in computer science: errors in AI-generated content are compounded over time, causing it to deviate further from reality. With the increasing amount of AI-generated content on the internet, these errors could become training data for future AI models, potentially leading to AI that fails to recognize new technologies or even spreads misconceptions.
4. Future Concerns: Can Humans Still Oversee AI?
If AI becomes expert-level in 15 years, who will be able to verify the accuracy of its answers? Currently, people rely on AI for learning and working, and future generations may depend even more on it. However, AI is not like a calculator—its decision-making processes are complex and beyond human comprehension. As AI’s capabilities grow, our trust in it could shift from rational verification to blind faith. The World Economic Forum warns that as AI becomes more powerful, our ability to oversee it will weaken (due to the reduction of cognitive tasks performed by humans). This is similar to relying on navigation systems too much, which can lead to a loss of independent thinking skills. Some suggest that we should require humans to perform challenging cognitive tasks regularly, similar to pilots needing to fly manually occasionally, to maintain their judgment. More importantly, while AI can distill existing knowledge, it cannot create individuals who are willing to take responsibility for the future. The “knowledge pasture” will not thrive on its own; human participation in discovering and sharing new knowledge is essential to prevent AI from becoming stagnant and declining in quality.
Conclusion
No matter how fast AI develops, it cannot replace humans’ ability to actively generate new knowledge. To prevent collective wisdom from becoming a mere echo of existing information, it is crucial to reward those who create knowledge and provide opportunities for the next generation to learn and make mistakes. After all, the source of human civilization lies in living, thinking individuals.