Beat Knowledge Rot: The 3 Types of Reading You Need in the Age of AI

AI is fueling a “knowledge rot” panic—claims that people are losing the ability to generate new knowledge because they rely on AI summaries. The core counterpoint is that the real problem isn’t knowledge disappearing; it’s that reading skills haven’t been updated for an era where information is abundant and AI sits beside every search. Instead of treating reading as one activity, the framework breaks it into three distinct modes—each with different goals, costs, and best uses—so learners can decide what to skim, what to retrieve, and what to read deeply.

The first mode is “awareness reading”: quick exposure to what’s happening so someone can keep up with the day’s conversation. This includes skimming news, browsing X, or reading a book just enough to pass a test. The second mode is “information retrieval,” which maps to how people already use search engines: identify a specific fact or concept, locate where it’s written, and pull it into a new output. In this view, Google-style searching is a form of reading, and large language models can function similarly—answering questions as retrieval tools. The third mode is “conneto reading” (also called deep reading), where understanding is formed through sustained processing. It’s the most valuable for long-term learning and memory, but also the most energetically expensive, with research cited that deep reading burns significant glucose as the brain works through complex material.

AI changes the balance among these modes by making it easier to generate and consume more “tokens” than ever—raising the stakes of filtering. The argument is that passive consumption is what weakens reading, not AI itself. When AI is used actively—paired with a real book, targeted questions, and a plan for what to learn—it can speed up retrieval and help learners reach the right gaps faster, leaving more time and attention for deep reading where it matters.

Concrete examples illustrate the split. For AI literacy fundamentals (a beginner book by Ben Jones), the recommended approach is retrieval reading: keep the book, but use AI (with Perplexity described as a useful search layer) to identify missing sections—such as token architecture, next-token prediction, or how chatbots generate responses—then read only those parts. The same retrieval logic is used to assemble a broader, accurate reading list, with a claim that hallucinated book suggestions were rare in testing.

For “co-intelligence living and working with AI” (Ethan Mollick’s bestseller), the advice shifts to conneto reading: read end to end because the narrative builds over time and the writing is described as clear enough to support deep digestion.

For news-style “knowledge rot” coverage, the framework places it in awareness reading. Many articles may be summarized quickly by tools like Perplexity, and the decision to invest in full deep reading should depend on whether the reporting is genuinely new and well researched.

The broader takeaway is a skill update: learn when to read for awareness, when to retrieve, and when to go deep. The argument also pushes back on education systems that rely on writing assignments as proof of understanding—suggesting that re-igniting curiosity and passion is the path back to genuine reading and voice. AI can help learners connect domains and produce new work, such as a personal example of using GPT5 Pro to design a Swedish-learning approach tailored to Indonesian knowledge. The thesis ends with a practical call: filter what to read, use AI as an active learning partner, and then put something back into the world—because curiosity and engagement are what ultimately fight “rot.”