📝 docs: import Mollick's Choosing to Stay Human and four cited papers

Adds Mollick's One Useful Thing post on AI pedagogy and cognitive
surrender/offloading, plus entity pages for all four Readwise-linked
papers (Bastani et al. Turkish/Taipei studies, BCG jagged frontier,
Deslauriers active learning). Journal reflects the full cluster with
Slack-style synthesis block.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: codekiln

journals/2026_06_16.md

@@ -1,8 +1,23 @@
+# Lessons Learned from Mollick's *Choosing to Stay Human*
+id:: 6a316c16-3f53-4015-8466-97688bcb8aac
+	- I recently read Ethan Mollick's [Choosing to Stay Human](https://readwise.io/reader/shared/01ksjxvecbcf3m8dbs31bfw71b) (One Useful Thing) and I was reminded about a distinction between
+	- ***Cognitive offloading*** - using tools to extend what you can do (notes, calculators, search). Historically benign, often beneficial.
+	- ***Cognitive surrender*** - when the tool does your thinking and you stop trying. Mollick mentions [a BCG consultant study](https://readwise.io/reader/shared/01kv8h9zgcdq6jrffx1832xspr) where professional consultants abandoned critical thinking when AI gave confident-sounding wrong answers. This implies that the "default" interaction with AI seems to be one of cognitive surrender rather than cognitive offloading
+	- He talks about two separate papers from the same Wharton team (Bastani et al.) reach apparently opposite conclusions w.r.t AI's ability to help learning:
+	- ***AI is detrimental to learning*** - [Turkish high school math](https://readwise.io/reader/shared/01kv8gk5bpgtdk5d1gkbcjs69z) - ChatGPT for homework → better homework scores, worse unassisted test performance. Shortcutting the struggle shortcutted the learning.
+	- ***AI can help learning if it produces optimal friction*** - [Taipei Python course](https://readwise.io/reader/shared/01kv8fncv79d7aab45vrs1xwjx) - have AI sequence practice problems matched to student knowledge state) → +0.15 standard deviation on final exams, ≈ 6–9 months of schooling. Same researchers, different outcome.
+	- Mollick mentions that people regularly report that they learn more from entertaining lectures, even if they actually learn more from doing hard problems and quotes a paper [Measuring actual learning vs feeling of learning in response to being actively engaged in the classroom](https://readwise.io/reader/shared/01kv8hp8j38ws6jnwwr4n1bvk3).
 - # [[Filed]]
+	- [[Person/Ethan Mollick/Blog/26/05/Choosing to Stay Human]]
+	- [[AI/Paper/25/Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning]]
+	- [[AI/Paper/25/Generative AI Without Guardrails Can Harm Learning]]
+	- [[AI/Paper/25/03/Navigating the Jagged Technological Frontier]]
+	- [[AI/Paper/19/09/Measuring actual learning versus feeling of learning]]
 	- [[tmux/Keyshort/Session/Choose a Session from a List]]
 	- [[tmux/Keyshort/Session/Rename Current Session]]
 	- [[tmux/Session/Q/What is the shortcut to choose a session from a list?]]
 	- [[tmux/Session/Q/What is the shortcut to rename the current session?]]
 	- [[tmux/Session/Q/What is the idiomatic way to create a new tmux session from inside an existing session?]]
-- [[TIL]]
+- # [[TIL]]
 	- [[LangSmith]] Agent Builder has been rebranded to [[LangSmith/Fleet]] - https://docs.langchain.com/langsmith/fleet/index
+	- [[LangSmith/Fleet]] depends on the [[OpenAI]] chat completions spec or [[Anthropic]] chat spec for custom models - https://docs.langchain.com/langsmith/fleet/essentials#custom-models

pages/AI___Paper___19___09___Measuring actual learning versus feeling of learning.md

@@ -0,0 +1,15 @@
+logseq-entity:: [[Logseq/Entity/Article]]
+source-link:: https://www.pnas.org/doi/10.1073/pnas.1821936116
+readwise-link:: https://readwise.io/reader/shared/01kv8hp8j38ws6jnwwr4n1bvk3
+date-created:: 2019-09-04
+- # [Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom](https://www.pnas.org/doi/10.1073/pnas.1821936116)
+	- *PNAS*, September 2019
+	- ## Key Findings
+		- Controlled study comparing active vs passive instruction in introductory physics (Harvard)
+		- Students in active classrooms learned more, but *felt* they learned less than peers in passive (lecture) settings
+		- Increased cognitive effort during active learning causes students to misinterpret productive struggle as poor learning
+		- Perception gap risks undermining motivation early in courses — instructors should explicitly frame struggle as evidence of learning, not failure
+	- ## Authors
+		- Louis Deslauriers, Logan S. McCarty, Kelly Miller, Kristina Callaghan, Greg Kestin — Harvard University
+	- ## Links
+		- Cited in [[Person/Ethan Mollick/Blog/26/05/Choosing to Stay Human]]

pages/AI___Paper___25___03___Navigating the Jagged Technological Frontier.md

@@ -0,0 +1,15 @@
+logseq-entity:: [[Logseq/Entity/Article]]
+source-link:: https://pubsonline.informs.org/doi/10.1287/orsc.2025.21838
+readwise-link:: https://readwise.io/reader/shared/01kv8h9zgcdq6jrffx1832xspr
+date-created:: 2025-03-11
+- # [Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality](https://pubsonline.informs.org/doi/10.1287/orsc.2025.21838)
+	- *Organization Science* (INFORMS), March 2025
+	- ## Key Findings
+		- Preregistered field experiment with 758 BCG consultants using GPT-4
+		- **Within AI's capability frontier**: +12.2% tasks completed, 25.1% faster, higher quality — benefits largest for lower-skilled workers
+		- **Beyond AI's capability frontier**: workers with AI access were 19% *less* likely to produce correct solutions — they failed to recognize AI's limitations and became overly reliant on its output
+		- Introduces the "jagged technological frontier" concept: seemingly similar tasks fall on opposite sides of where AI helps vs. harms performance
+	- ## Authors
+		- Fabrizio Dell'Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz-Assaf, Katherine Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, Karim R. Lakhani — Harvard, MIT, Wharton, University of Warwick, BCG
+	- ## Links
+		- Cited in [[Person/Ethan Mollick/Blog/26/05/Choosing to Stay Human]]

pages/AI___Paper___25___Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning.md

@@ -0,0 +1,16 @@
+logseq-entity:: [[Logseq/Entity/Article]]
+source-link:: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6423358
+readwise-link:: https://readwise.io/reader/shared/01kv8fncv79d7aab45vrs1xwjx
+- # [Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6423358)
+	- [PDF preprint](https://hamsabastani.github.io/llmRL_doe.pdf) — Working Paper
+	- > Generative AI (GenAI) is rapidly reshaping education by unlocking the potential for personalized tutoring. Yet, emerging platforms largely focus on GenAI chatbot tutors that reactively answer student questions. We design a novel tutoring platform that tightly integrates a carefully-designed GenAI chatbot with a reinforcement learning algorithm for sequencing practice problems. This algorithm leverages rich signals from student-chatbot interactions to adaptively select practice problems of an appropriate difficulty level.
+	- ## Key Findings
+		- Randomized controlled trial across 10 high schools in Taipei (n=770 students) — 5-month "AI for Python Learning" course with Taipei City Government and American Institute in Taiwan
+		- Adaptive problem sequencing (RL-guided) vs fixed sequence
+		- Adaptive sequencing increased unassisted final exam performance by **0.15 standard deviations** (≈ 6–9 months of schooling)
+		- Mediation analysis: gains driven by increased engagement
+		- LLMs used both to measure quality of student-chatbot conversations and to analyze student solution attempts, enabling richer knowledge-state estimation than binary correct/incorrect signals
+	- ## Authors
+		- Angel Tsai-Hsuan Chung, Botong Zhang, Ling-Chieh Kung (National Taiwan University), Hamsa Bastani, Osbert Bastani — Wharton / UPenn
+	- ## Links
+		- Cited in [[Person/Ethan Mollick/Blog/26/05/Choosing to Stay Human]]

pages/AI___Paper___25___Generative AI Without Guardrails Can Harm Learning.md

@@ -0,0 +1,15 @@
+logseq-entity:: [[Logseq/Entity/Article]]
+source-link:: https://www.pnas.org/doi/10.1073/pnas.2422633122
+readwise-link:: https://readwise.io/reader/shared/01kv8gk5bpgtdk5d1gkbcjs69z
+date-created:: 2025
+- # [Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics](https://www.pnas.org/doi/10.1073/pnas.2422633122)
+	- [PDF preprint](https://hamsabastani.github.io/education_llm.pdf)
+	- *PNAS* (2025) — cited in the [2025 Economic Report of the President](https://bidenwhitehouse.archives.gov/wp-content/uploads/2025/01/ERP-2025.pdf)
+	- ## Key Findings
+		- Turkish high school math students randomized to ChatGPT access for homework had better homework completion but **worse performance on unassisted tests**
+		- Self-regulated AI use short-circuits the productive struggle required for learning
+	- ## Authors
+		- Hamsa Bastani*, Osbert Bastani*, Alp Sungu*, Haosen Ge, Özge Kabakçı, Raphael Mariman (* equal first authorship) — Wharton / UPenn
+	- ## Links
+		- Cited in [[Person/Ethan Mollick/Blog/26/05/Choosing to Stay Human]]
+		- Contrast with [[AI/Paper/25/Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning]] — same team, adaptive tutoring improves outcomes

pages/Person___Ethan Mollick___Blog___26___05___Choosing to Stay Human.md

@@ -0,0 +1,19 @@
+logseq-entity:: [[Logseq/Entity/Article]]
+created-by:: [[Person/Ethan Mollick]]
+date-created:: 2026-05-26
+source-link:: https://www.oneusefulthing.org/p/choosing-to-stay-human
+readwise-link:: https://readwise.io/reader/shared/01ksjxvecbcf3m8dbs31bfw71b
+- # [Choosing to Stay Human](https://www.oneusefulthing.org/p/choosing-to-stay-human)
+	- ## Summary
+		- Examines how AI tool usage can either enhance or undermine human cognition depending on implementation, advocating for intentional adoption over reflexive dependence.
+	- ## Highlights
+		- "By short-circuiting effort, you short-circuit learning"
+		- **Shortcut approach (harm)**: [[AI/Paper/25/Generative AI Without Guardrails Can Harm Learning]] — Turkish high school math, ChatGPT users had better homework completion but worse test performance
+		- **Enhancement approach (benefit)**: Taipei Python course — adaptive AI tutoring (LLM + RL) improved unassisted final exam scores by 0.15 standard deviations (≈ 6–9 months of schooling) — [[AI/Paper/25/Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning]]
+		- **Cognitive surrender**: BCG consultant study — even elite workers abandoned critical thinking when AI gave confident but wrong answers
+		- Defaults established now around AI use may become difficult to reverse
+	- ## Links
+		- [[AI/Paper/25/Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning]]
+		- [[AI/Paper/25/Generative AI Without Guardrails Can Harm Learning]]
+		- [[AI/Paper/25/03/Navigating the Jagged Technological Frontier]]
+		- [[AI/Paper/19/09/Measuring actual learning versus feeling of learning]]