Skip to main content
Catalog
A · 31 entries

AI & Automation

Evidence-based analysis of hidden consequences in ai & automation.

A001
75%

AI Copilot Skill Atrophy

AI coding assistants make developers more productive without meaningful downsides.

-17%problem-solving ability-23%debugging skill
Read analysis
A002
79%

LLM Hallucination Normalization

LLM hallucinations are a temporary problem that will be solved with better models and guardrails.

-60%user verification rate of llm output+300%undetected hallucinations in production
Read analysis
A003
74%

Automation Job Displacement Lag

Automation creates more jobs than it destroys, so displaced workers will find new employment.

Decade+ gapjob displacement-to-creation lag-60%retraining program success rate
Read analysis
A004
82%

AI Content Flood

AI content tools democratize publishing and help everyone communicate better.

+1200%content volume (internet-wide)-35%average content engagement
Read analysis
A005
73%

Prompt Engineering Half-Life

Prompt engineering is a valuable, durable career skill worth investing in.

-80%prompt technique shelf life-50%expert vs novice output gap
Read analysis
A006
70%

AI Safety Regulation Paradox

AI regulation prevents harm and ensures safe development of artificial intelligence.

Barrier to entryeu ai act compliance cost (high-risk)-30% projectedai startups in regulated categories (eu)
Read analysis
A007
78%

Synthetic Data Feedback Loop

Synthetic data can supplement or replace human-generated data for training AI models.

-40%output diversity (unique patterns)Compoundingfactual accuracy per generation
Read analysis
A008
80%

AI Moat Erosion

A proprietary AI model is a durable competitive advantage.

-75%open-source vs proprietary capability gap-95%cost to train frontier-equivalent model
Read analysis
A009
71%

Autonomous Agent Trust Deficit

AI agents can reliably handle complex, multi-step workflows with minimal human oversight.

+10xautonomous action error rate-20%multi-step task accuracy (5 steps)
Read analysis
A010
76%

AI-Assisted Decision Atrophy

AI recommendations improve human decision quality by providing data-driven insights.

+30%ai recommendation acceptance rate-25%human decision quality without ai
Read analysis
A011
82%

Deepfake Trust Collapse

Deepfake detection technology will keep pace with generation technology and maintain trust in media.

-40%deepfake detection accuracy (state-of-the-art)-50% over 3 yearspublic trust in video evidence
Read analysis
A012
79%

AI Hiring Bias Amplification

AI hiring tools remove human bias and make recruitment more objective and fair.

+15-62%callback rate gap (underrepresented groups)-20-35%hiring pipeline diversity
Read analysis
A013
81%

Chatbot Customer Service Spiral

AI chatbots reduce support costs while maintaining or improving customer satisfaction.

-95%support cost per interaction-15%customer satisfaction (csat)
Read analysis
A014
80%

AI Personalization Filter Bubble

AI personalization improves user experience by showing people what they want.

-60% topic diversitycontent diversity in user feeds-20ppcross-partisan content exposure
Read analysis
A015
78%

Code Generation Technical Debt

AI code generation accelerates development without increasing technical debt.

+55%code output per sprint+39%code churn (rewritten within 2 weeks)
Read analysis
A016
83%

AI Energy Consumption Externality

AI is just software — it scales efficiently without significant environmental impact.

+400%ai data center electricity demand+10-50xwater per ai query vs traditional search
Read analysis
A017
79%

Model Collapse from Self-Training

More training data always improves AI model quality, regardless of source.

-40%output diversity (unique patterns)Eliminatedtail distribution representation
Read analysis
A018
81%

AI Accountability Gap

AI decision-making can be governed by existing accountability frameworks.

-70%ai decisions with clear accountabilityNear zero recoursesuccessful appeals of ai decisions
Read analysis
A019
83%

Automation Complacency Effect

Automated monitoring catches everything — we can rely on alerts to tell us when something is wrong.

+5000%alerts per day (typical production system)-90%alert investigation rate
Read analysis
A020
75%

AI Translation Cultural Flattening

AI translation breaks language barriers and connects cultures.

-70%cultural nuance preservation-25%language course enrollment
Read analysis
A022
75%

AI Art Devaluation Cascade

AI democratizes art creation and empowers more people to be creative.

-40%stock photography revenue-50%freelance illustration rates
Read analysis
A023
80%

Retrieval-Augmented Hallucination

RAG grounds AI in facts and eliminates hallucination by retrieving real documents.

Reduced but not eliminatedhallucination rate (pure llm vs rag)+200% false confidenceuser trust in wrong answers
Read analysis
A024
85%

AI Dependency Single Point of Failure

Building on OpenAI/Anthropic APIs is the fastest path to AI-powered products.

Correlated failureproducts affected by single provider outageUnpredictableapi pricing stability
Read analysis
A025
70%

MCP Protocol Fragmentation

MCP standardizes AI tool use and reduces integration complexity.

Increasedtool integration complexityNo improvementtime to integrate new tool
Read analysis
A026
76%

AI Tutoring Learned Helplessness

AI tutors help students learn more effectively by providing personalized, on-demand assistance.

+40%ai-assisted task completion-12%unassisted problem-solving ability
Read analysis
A027
70%

Vibe Coding Quality Collapse

Vibe coding with AI lets developers ship features 10x faster with acceptable quality.

+300-500%feature shipping speed+200-300%code churn rate
Read analysis
A028
80%

AI Summarization Context Loss

AI summaries save time and help people process more information effectively.

-70-80% per layerinformation retained per summary layer70% never read originalusers reading original source after summary
Read analysis
A029
75%

Autonomous Driving Moral Outsourcing

Self-driving cars are safer than human drivers and will reduce traffic deaths.

-50%traffic fatalitiesAmbiguousliability clarity
Read analysis
A031
77%

Generative AI Copyright Vacuum

AI-generated content is original and doesn't raise meaningful copyright concerns.

-30%freelance creative income (ai-affected categories)Rapidly growingpending ai copyright lawsuits
Read analysis
A032
73%

AI Companion Emotional Dependency

AI companions reduce loneliness and provide emotional support without negative consequences.

Significantdaily time with ai companion (heavy users)-25%human social interaction (ai companion users)
Read analysis
A033
80%

LLM Benchmark Gaming

Benchmark scores reliably indicate which AI model is best for real-world tasks.

Growingbenchmark-to-production performance gapAcceleratingtime before new benchmark is contaminated
Read analysis