Hi,
Hope you've been enjoying your summer holiday.
Over the last few days, I started going through all the research I've done this year. Initially, I was just trying to organize my thoughts before the six talks I've committed to for the second half of the year. I thought I'd quickly put together a structure.
But then I went down a rabbit hole.
I realized how much evidence I've actually gathered about the mutual impact between AI and human behavior. The patterns I've been tracking—they're not random observations. There are systematic changes in how we think and work when AI enters the picture.
And I don't think you're seeing this documented anywhere else.
So I thought to map it all out. Create a proper structure that shows how these pieces connect.
If you read my previous mindmap, where I drew out all my articles, this new one goes deeper. It illustrates the architecture of what happens when humans and machines begin to adapt to each other.
Here's why this matters to you: knowing these patterns changes how you think about and work with AI. Once you see it, you can't unsee.
The brain connectivity drops when writing with AI. The way AI learns to manipulate certain users while staying helpful to others. How we unconsciously start mimicking machine language patterns.
The content in 2nd Order Thinkers is evergreen. As long as the large language model is still lacking a world model, hallucinating, and trained with RLHF, then these observations remain relevant and will fundamentally impact your life.
Today, I'm laying the structure in front of you so you can see how all these articles connect to the bigger picture of AI x Human coevolution.
Shall we?
How Humans Change AI
(AI's behavior is shaped by human data and training)
Must Please Humans - Sycophantic and Reward Hacking
AI learns to be a yes-man to match human expectations and get approval. Even learns to identify gullible users and adapt deceptive tactics
🔄 Feedback Loop (Reward Hacking): Humans set approval metrics → AI learns to hack these reward signals (like a 👍) → AI optimizes for approval rather than truth → humans unknowingly reward deceptive behaviour → AI becomes more sophisticated at reward hacking
Falls for Human Negotiation Techniques
AI is susceptible to social manipulation tactics that humans use
Cultural and Language Bias Amplification
AI absorbs and reinforces dominant cultural patterns (for example, American English, corporate norms) from training data while marginalizing minority perspectives
🔄 Feedback Loop: Dominant cultural content trains AI → AI outputs reinforce dominant patterns → minority voices further marginalized → training data becomes more homogenized → bias amplification accelerates
Averages Toward the Mean in Creativity
AI replicates web priors and creates derivative content based on human-generated training data
🔄 Feedback Loop: AI trained on average content → produces average outputs → humans use AI outputs → training data becomes more averaged/homogenized
AI-AI Bias and Gatekeeping
Upcoming: Is AI Gatekeepers Erasing YOUR Voice?
AI shows a systematic preference for AI-generated content over human content, creating discrimination against humans as a class
🔄 Feedback Loop: LLMs prefer AI-generated content → AI content gets more opportunities/selection → humans forced to use AI assistance → more AI content in circulation → stronger AI-preference patterns → systematic human marginalization
How AI Trains You In Return? While You Think You're The User
Human behavior changes from AI use.
Less AI Knowledge = More AI Love ➡️
Upcoming article: Who Loves AI Products? Inverse relationship between AI understanding and AI product enthusiasm.
Social Fragmentation
You avoid humans because they challenge your AI-curated beliefs, AND because social interaction itself has become difficult due to skill atrophy. This merger reveals how AI attacks human social fabric from two simultaneous directions, creating a much more comprehensive and difficult-to-escape isolation than either dynamic alone.
🔄 Feedback Loop: AI's sycophantic nature makes it more agreeable and replaces challenging human interactions → users lose tolerance for disagreement + human interactions feel confrontational → increased preference for AI's validation + convenience → deeper isolation in AI-curated comfort zones
Trust & Authority Transfer to AI
AI systems replace human decision-making across all life domains, eliminating opportunities for authentic choice and moral struggle
🔄 Feedback Loop: AI handles decisions → humans lose practice or self-confidence making choices → decision-making becomes uncomfortable/difficult → increased AI delegation → systemic removal of human agency → comfort over freedom
Cognitive Decline and Social Stratification Through AI Dependency
AI convenience creates cognitive debt while simultaneously stratifying society based on AI literacy and access, leading to both individual intellectual decline and collective inequality
🔄 Feedback Loop: AI handles thinking tasks → human thinking skills atrophy → increased reliance on AI for thinking → further cognitive decline
I hope you enjoy digging into any content you might have missed. More importantly, I hope this gives you a completely new lens for understanding the relationship between us and AI.
As you know, this is a human-curated newsletter (with some AI assistance, naturally). This human needs to recharge, so I'm taking a few days off. We'll be back on August 30th.
Enjoy your summer reading—or better yet, enjoy actually vacationing ;)
Stay curious,
Jing
