Every major technological shift produces a version of the same argument: this new technology will make the old skill obsolete. Calculators made arithmetic obsolete. Search engines made memorisation obsolete. Now AI can read for you — summarise any document, extract key points from any book, answer questions about any text.
The argument is wrong in the same way it was wrong before. Not because technology does not change what skills are needed — it does — but because the skills that matter most are not the surface-level operations that technology is replacing. They are the deeper capacities that technology augments but cannot build.
What AI actually does to reading
AI language models can:
- Summarise text accurately at scale
- Retrieve specific information from documents on demand
- Synthesise across multiple sources at a surface level
- Generate readable prose on almost any topic
- Answer questions about content they have been trained on or provided
These are genuinely valuable capabilities. They change the economics of information access fundamentally — you can now get a reasonable summary of any book in seconds, or find the relevant passage in any document without reading it.
What they cannot do:
- Evaluate arguments critically — AI can identify logical form but is unreliable on whether specific premises are well-supported by evidence in the world
- Develop judgment about a domain — that requires sustained engagement with primary sources, not summaries
- Form genuine analogies between distant domains — the kind of insight that comes from reading widely and deeply across fields
- Notice what is absent — recognising that an argument is missing a key consideration requires knowing what the full consideration space looks like
The skills that deep reading builds — critical evaluation, analogical reasoning, recognising argument quality, domain intuition — are precisely the skills needed to direct, evaluate, and make use of AI outputs. You cannot use AI effectively without the judgment to know when it is wrong.
The cognitive development argument
Maryanne Wolf (2018) argues that deep reading does not just transfer information — it changes the brain's cognitive architecture. The slow, inference-rich processing of complex text develops:
- Critical analysis circuits — evaluating claims against evidence
- Empathic cognition — inhabiting perspectives sufficiently different from your own to understand them
- Analogical reasoning — finding structural similarities between apparently different domains
- Integration across long timespans — maintaining and using context from many pages earlier
These develop through practice over years. They are not acquired through reading summaries or interacting with AI interfaces, because the development requires the sustained cognitive engagement that deep reading provides. An AI-generated summary activates recognition; a read book activates construction.
This is not technophobia. It is a mechanistic claim about what kinds of cognitive engagement produce what kinds of cognitive development. AI tools are excellent at providing inputs (information, summaries, examples); they cannot substitute for the processing that builds capacity.
AI literacy requires reading literacy
Here is the underappreciated inversion: AI makes deep reading skills more valuable, not less.
As AI-generated text becomes ubiquitous, the ability to evaluate it becomes critical. AI language models produce fluent, confident text that is sometimes accurate and sometimes not. They hallucinate citations, misrepresent study findings, and build coherent-sounding arguments on shaky foundations.
Evaluating AI output requires exactly the critical reading skills described in our reading comprehension guide: inference monitoring (noticing when a claim does not hold), argument tracking (identifying whether evidence actually supports the stated conclusion), and domain knowledge (recognising when a claim contradicts what you know from primary sources).
A reader who has worked through primary literature in a domain — who has read the actual papers, followed the actual arguments, formed their own understanding from the ground up — can evaluate AI summaries of that domain reliably. A reader who has only encountered that domain through AI summaries cannot.
Reading as a competitive advantage
In professional contexts, reading widely and deeply is increasingly a differentiator — not despite AI, but because of it.
Everyone with access to the same AI tools gets access to the same summaries, the same retrieved facts, the same surface-level synthesis. What AI cannot level-equalise is the judgment, domain depth, and cross-domain pattern recognition that comes from years of sustained reading.
The business executive who has read widely in economics, psychology, history, and strategy does not have better access to information than competitors with equal AI tools. They have better judgment about what that information means and what to do with it. That judgment is built through reading that cannot be shortcut.
This is consistent with what CEOs and high-performers say about reading: reading widely is not a credential or a hobby. It is an ongoing practice of judgment development.
How to read in an AI world
Use AI for triage, not reading: AI summaries are excellent for deciding whether a document deserves full reading — use them to identify the 20% of papers or books worth reading in full.
Read the source for what matters: For books and arguments you want to genuinely understand, read the source. AI summaries strip the argument structure, the nuance, and the cognitive development that full reading provides.
Practise critical evaluation: Apply the comprehension monitoring techniques to AI-generated content as deliberately as you apply them to human-written text. Actively question: Is this claim actually supported? Is this inference valid? What is this missing?
Read in domains where AI is weakest: Philosophy, literature, history, political theory — domains where the value is in the argument and the prose rather than the fact content — are where human reading is least substitutable. These are also the domains where reading produces the most durable cognitive development.
Use RSVP to read more primary sources: The ability to read primary sources faster — using warpread.app at a comfortable WPM — means you can engage with more primary sources in the time available. This is a direct investment in the judgment that AI cannot build for you.
The readers who thrive in an AI world will be those who use AI for information access and their own reading for intellectual development. These are not competing activities — they are complementary. AI handles the retrieval; reading builds the capacity to direct and evaluate what AI retrieves.
Build your reading practice on warpread.app — free RSVP reader
References
- Wolf, M. (2018). Reader, Come Home: The Reading Brain in a Digital World. Harper.
- Carr, N. (2010). The Shallows: What the Internet Is Doing to Our Brains. W.W. Norton.
- Dweck, C.S. (2006). Mindset: The New Psychology of Success. Random House.
- Harari, Y.N. (2018). 21 Lessons for the 21st Century. Spiegel & Grau.
- Tomas, T., et al. (2023). Critical AI literacy: what it means to think critically about AI-generated content. Computers & Education, 203, 104820.
Optimise your reading environment
Benchmark your current reading speed, then take the Focus & Deep Work course to build the conditions for sustained concentration and reading flow.