Cognitive Debt - What AI will do to us

Transcript

1. Cognitive Debt What AI will do to us Martin Lippert, Spring Tools Lead, Broadcom - September 2026
2. https://www.scientificamerican.com/article/is-ai-ruining-our-skills-early-results-are-in-and-theyre-not-good/
3. g n i t t e g m ’ h I t i e k w i l r s e l b e e m f u “ t . I d y „ a d d n a y r b e y b a d m , u t d p m o r p y r e v e nt“ e m p t/ elo n v e e d m are elop w v t e f o d s of are e w r t f u t o f-s e fu o h t e r d u t -fu I, an e A h , t t f i „Sw i-and : a s t c f i e öd sw / B r m o o Tib ev.c d t f i w hes t / / : s p htt „I don't really co de anymore. I pr ompt. And somewhere along the way, I lost the part of codin g I actually love d.“ Harsh: „I Used to Love Coding. No https://dev.to/ha w I Just Prompt. rsh2644/i-used-t “ o-love-coding-n ow-i-just-promp t-550l a o t d e t c i d d a e r a e w f t o a l h l t a d m n e a e t s s r d l e u d o n u w t ’ „It n o e d v i e s w s a d n m a s , i g m u r e d h t f w e o n e n o t u B . t e y s “ t . c y e a d y r e its e v e , e u g i fat ff pire“ m a V I A 07163 f e 4 h e T „ 6 : a e d g e pirem a v Steve Yeg i a e om/th .c m u i d e gge.m e y e v e t s / https:/
4. The Research
6. Thinking - Fast, Slow, and Artificial Study Design Task Participants Evaluation • 7 tests adapted from the CRT (Cognitive Reflection Test) • 359 / 485 / 450 • Accuracy • 2 groups • AI chat usage • 3 studies in total • Access to AI (GPT-4o) - with controlled / modified output • No Access to AI • Follow / override rates • Confidence
7. Thinking - Fast, Slow, and Artificial Results Accuracy Brain-Only Baseline: 45.8% With Accurate AI Response: 71% With Wrong AI Response: 31.5%
8. Thinking - Fast, Slow, and Artificial Results High Reliance on AI Following Correct advice: 92.7% Following Wrong advice: 80% High Decision Confidence even when AI was wrong
9. Cognitive Surrender „Cognitive surrender is what happens when you stop constructing the answer at all. The AI’s output becomes your output. There’s nothing to override, because you never formed an independent view to compare it against.“ Addy Osmani: „Cognitive Surrender“ https://addyosmani.com/blog/cognitive-surrender/
11. Your Brain on ChatGPT Study Design Task Participants Evaluation • 3x Write an essay • 54 • Neural Activity (EEG) • Switch Groups • 3 groups • NLP • 1x Write an essay • LLM Group • Interviews • Search Engine • Essay Scoring (humans + AI) • Brain only
12. Your Brain on ChatGPT Results - Brain Activity Brain Only LLM connectivity scaled up connectivity scaled down strongest and widest-ranging networks weakest functional coupling reduced engagement
13. Your Brain on ChatGPT Results - Memory & Retention Brain Only LLM high memory retention + quoting 83.3% failed to recall quotes (0% provided a completely accurate quote in session 1) accurately recalling sentences
14. Your Brain on ChatGPT Results - Ownership Brain Only LLM strongest sense of ownership lowest perceived ownership
15. Your Brain on ChatGPT Session 4 - the Crossover LLM (prior Brain Only) Brain Only (prior LLM)
16. Your Brain on ChatGPT Session 4 - the Crossover LLM (prior Brain Only) network-wide surge in connectivity high memory recall performance high ownership Brain Only (prior LLM)
17. Your Brain on ChatGPT Session 4 - the Crossover LLM (prior Brain Only) Brain Only (prior LLM) network-wide surge in connectivity weaker neutral connectivity high memory recall performance severe quoting deficits high ownership less coordinated lower to medium ownership
18. Your Brain on ChatGPT Session 4 - the Crossover LLM (prior Brain Only) Brain Only (prior LLM) network-wide surge in connectivity weaker neutral connectivity high memory recall performance severe quoting deficits high ownership less coordinated lower to medium ownership This is the part that indicates longer-term effects of using AI
19. Cognitive Debt „When individuals fail to critically engage with a subject, their writing might become biased and superficial. This pattern reflects the accumulation of cognitive debt, a condition in which repeated reliance on external systems like LLMs replaces the effortful cognitive processes required for independent thinking.“ (mentioned as preliminary results, larger participant sample is needed to verify this)
20. The Road to Cognitive Debt
21. Cognitive Offloading Cognitive Surrender Comprehension Debt Cognitive Debt
22. Consequences for us as Software Engineers
23. Writing Code Your knowledge and ability to write code is built on top of your experience writing code. If you let the LLM write all the code, you will loose the ability to write code over time.
24. Reviewing Code Your knowledge and ability to review code is built on top of your experience writing code If you let the LLM write all the code, you will loose the ability to review code over time
25. Escape Path?
27. How AI Impacts Skill Formation Study Design Task Participants Evaluation • 35min coding session (one session only) • 52 (Python) • Questions (point quiz) • 2 groups • Screen recording annotations (manual) • 2 coding tasks each • Learning new library • Access to AI (chat) • No Access to AI
28. How AI Impacts Skill Formation Results - Skill Retention & Speed Skill Retention Speed With AI: 50% With AI: 23min Without AI: 65.5% Without AI: 24.7min
29. How AI Impacts Skill Formation Results - Engage vs. Offload Cognitive Engaged Cognitive Offloaded High Scores Low Scores This is interesting !!!
30. How AI Impacts Skill Formation Interaction Patterns - Cognitively Offloaded AI Delegation Progressive AI Reliance Iterative AI Debugging
31. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Generation-ThenComprehension Hybrid Code-Explanation Conceptual Inquiry
32. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Generation-ThenComprehension Generated code first, then asked AI follow-up questions to understand it
33. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Generation-ThenComprehension Hybrid Code-Explanation Conceptual Inquiry
34. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Hybrid Code-Explanation Asked queries that requested code and explanations together
35. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Generation-ThenComprehension Hybrid Code-Explanation Conceptual Inquiry
36. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Conceptual Inquiry Only asked high-level conceptual questions, wrote and debugged code independently
37. How AI Impacts Skill Formation Interaction Patterns - Cognitively Engaged Generation-ThenComprehension ? t b e D e v i t i n g o ts C n e v Pre Hybrid Code-Explanation Conceptual Inquiry
38. Thank You Martin Lippert @martinlippert
39. References Research Papers: https://www.media.mit.edu/publications/your-brain-on-chatgpt/ https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646 https://www.anthropic.com/research/AI-assistance-coding-skills https://arxiv.org/abs/2601.20245 https://arxiv.org/abs/2510.07435 Blogs: https://margaretstorey.com/blog/2026/02/09/cognitive-debt/ https://theswiftdev.com/swift-ai-and-the-future-of-software-development/ https://addyosmani.com/blog/cognitive-surrender/ https://addyosmani.com/blog/comprehension-debt/ https://steve-yegge.medium.com/the-ai-vampire-eda6e4f07163 https://tomwojcik.com/posts/2026-02-15/finding-the-right-amount-of-ai/ https://dev.to/harsh2644/i-used-to-love-coding-now-i-just-prompt-550l https://www.fast.ai/posts/2026-01-28-dark-flow/ https://www.scientificamerican.com/article/is-ai-ruining-our-skills-early-results-are-in-and-theyre-not-good/ https://danariely.com/where-is-ai-going-and-where-are-we-going-will-we-be-smart-enough-to-make-the-right-choices/ https://techcrunch.com/2026/05/29/coders-are-refusing-to-work-without-ai-and-that-could-come-back-to-bite-them/