Topic guide
AI and cognition
Public debate often asks whether AI makes people smarter or dumber. That question is memorable and usually too broad. A person can complete a task more quickly with AI while learning less from it. Another person can use the same class of system as feedback, a source of graduated hints or a second reader and preserve more of the work that builds independent skill.
This guide organizes Prompted Psyche materials around a narrower question: what kind of thinking remains in the human part of the interaction? It separates immediate performance from learning, retention and transfer, then connects those distinctions to cognitive offloading, metacognition and the design of assistance.
Area definition
What AI and cognition means here
Cognition includes processes such as attention, memory, comprehension, problem solving, monitoring and decision-making. AI can redistribute those processes between a person, an interface and a model. That redistribution is not automatically improvement or decline. Its effect depends on the goal, task, prior knowledge, timing of help and what is measured after the help disappears.
A central distinction is between producing an acceptable result and developing a capability. The first can be measured while the tool is present. The second requires retention or transfer: whether the person can use what was learned later, in a changed problem or without the same support.
Distinctions that matter
- Performance is not learning
- A better result with assistance does not show what remains when assistance is removed.
- Offloading is not decline
- External tools can extend cognition; risk grows when essential practice disappears unnoticed.
- Answering is not scaffolding
- Support can preserve effort by asking, hinting, giving feedback and then withdrawing.
- Recognition is not understanding
- A fluent explanation can feel familiar before a learner can reproduce the reasoning.
Key questions
- Is the goal to finish the task, learn the skill or do both?
- Which cognitive steps does the person still perform before seeing the model output?
- Does the interaction require an attempt, retrieval, explanation or transfer?
- What happens when the support is reduced or removed?
- Which capabilities still need practice because responsibility remains with the user?
A useful way to orient yourself
First identify the purpose. Direct generation may be appropriate when the aim is routine completion and the result can be checked. Learning requires a different interaction: an initial attempt, feedback calibrated to the error, opportunities for retrieval and a later test without the same support.
The relevant unit is rarely the model alone. It is the whole arrangement of prompts, interface, task, incentives, prior knowledge and review. This is why the same model can function as a substitute in one workflow and as scaffolding in another.
Starting point
Where to start
The main essay reviews what current evidence can and cannot support, without treating short-term studies as proof of a durable change in intelligence.
A careful distinction between task performance, learning, retention and transfer when AI becomes part of the process.
Library
Explore the topic from several angles
Articles build the argument, concepts clarify the mechanisms and Practice turns them into concrete actions.
Articles
Read the main synthesis first, then use the related essays to connect cognitive practice with context and trust.
Concepts
These entries distinguish cognitive demand, external support, self-monitoring and the loss of practiced capability.
Practice
Use these scenarios to keep interpretation, resistance and authorship in the human part of the process.
Notes
Use these notes to separate terminology, compression and the limits of model context.
Sequence
Suggested reading path
- Don't Ask Whether AI Makes Us Dumber. Ask What Kind of Thinking We Stop Practicing
Start with the evidence and the distinction between performance and learning.
- Cognitive offloading
Understand why moving work to a tool is not automatically cognitive decline.
- Metacognition
Focus on the ability to monitor understanding and independent capability.
- How to ask AI for a counterargument
Use AI in a way that introduces productive resistance instead of removing it.
- Trust in the age of ready-made answers
Connect cognitive shortcuts to the compressed route between sources and conclusions.
Connections
How the pieces connect
The main article sets the evidential boundaries. Concepts make the cognitive mechanisms easier to distinguish, while Practice turns them into choices about when the model answers, asks, hints or critiques. Notes preserve two constraints: a summary is not a decision, and textual context is not the whole situation.
The cluster does not assume that using AI is harmful. It asks for a better description of the interaction. The useful question is which processes a person wants to delegate, which must remain practised and how to tell whether support has built capability or only improved the assisted result.