AI summary by TwinMind
Science is in trouble and it worries me.
Sabine Hossenfelder
24m 32s
Scientific Growth Metrics vs. Productivity Decline
- Number of scientists worldwide increased rapidly, continuing to grow
- Both absolute numbers and share of population rising
- Past 30 years: steep increases in EU, North America, parts of Asia
- Scientists per 1 million people data shows dramatic growth
- Scientific publications increasing exponentially across fields
- Physics literature doubles every ~18 years
- Electrical engineering literature doubles every ~9 years
- Despite growth, fewer results from increased effort
Economic Impact Analysis
- Total Factor Productivity measures inputs (labor, capital, technology/R&D) vs outputs (goods/services)
- 2016 economists paper "Are Ideas Getting Harder to Find?"
- US factor productivity growth remained constant despite steep researcher increase
- More effort in research (money + people) barely maintains stable growth
- Specific sector analysis shows declining returns:
- Agricultural yields: researcher numbers up, results flat
- New FDA drug approvals declining vs researcher numbers
Patent and Innovation Studies
- US patent novelty analysis: quantified by new connections between technology sectors
- Patent novelty declining since 1960s
- Invention narrowness has increased
- 2022 study: research productivity by top researchers per community
- Steep decrease since 1960s
- 2023 Nature study: 45 million papers analyzed worldwide
- Disruptiveness measured by papers made redundant
- Disruptive ideas noticeably decreased across physical sciences, chemistry, computer science, biomedicine
- Even top journals (Nature, Science) showing decline
Historical Context and Trends
- Jonathan Hubner 2005: innovation rate declining since industrialization
- Rate = technological events ÷ number of people
- Extrapolation suggests no progress or infinite population by century end
- Multiple studies using different metrics consistently show same pattern
- Technologies like AI, quantum computing, fusion, autonomous vehicles based on decades-old breakthroughs
- Slow pace + enormous investments illustrate rather than contradict the problem
Potential Causes Analysis
- R&D Funding: Not the cause
- R&D as share of GDP roughly constant past decades
- Absolute terms increased, federal contribution decreased
- Not exclusively American phenomenon
- Scaling Issues: Hiring more researchers has worked to some extent
- Largely compensated for decreased productivity per researcher
- Companies wouldn't invest if no returns
- Doesn't explain qualitative change in knowledge discovery
- Low-Hanging Fruit Theory: Implausible as primary cause
- Effect happened simultaneously across disciplines, countries, sectors
- Agriculture, drug development, chemistry, biomedicine, physics
- Systemic problem starting 1960s-70s globally
Academic Research Organization Issues
- Most R&D privately funded (US + EU)
- Most basic research (long-term breakthrough potential) happens in publicly funded academia
- World War II/Apollo missions had shared purpose + enormous pressure
- Academia has growing fraud/misconduct problem but not major factor yet
Bureaucracy Problems
- Elon Musk examples: SpaceX shark/whale impact studies, underwater explosion whale hearing damage analysis
- Peter Thiel: Large bureaucratic institutions from moon program slowed innovation
- Patrick Collison Fast Grants during COVID
- No scientific grantmaking designed for decisions in hours/days
- NIH proposals: specific font requirements, narrow criteria, months to prepare
- Fast grants: most researchers complete form within 1 hour
- Academic bureaucracy: grant applications → accounting → reports → reports on reports
Publish or Perish Culture
- Problem of valuing quantity over quality
- Rewarding by paper numbers, journal prestige, citations
- Known for decades but persists
- Not in anyone's interest to fund useless research
Risk Aversion in Academic Research
- High competition for funding/jobs creates incentives for reliable results
- Encourages low-risk mainstream research vs novel/disruptive ideas
- Funding agencies explicitly want "transformative" or "high-risk high-payoff" research
- Yet consistently fund "same old boring stuff"
Public Perception and Trust Impact
- Scientists declare much research "boring" to excuse lack of tangible benefits
- Author calls it "bullshit research": underpowered trials, irreproducible psychology studies, crappy meta-analyses
- Examples: airport noise affects sleep, better photovoltaics claims, unmeasured particles
- Most bullshit research invisible to average person
- Fox News examples of questionable federal spending on research
- People will question why taxes fund academic jobs with no societal benefit
Researcher Satisfaction Data
- Fast Grants survey: 78% of biomedical researchers would change research agenda significantly if given flexibility
- Author's unpublished 20-year-old physics survey: majority would change topics if possible
- Substantial number of scientists feel forced to work on research they don't consider worthwhile
- Researchers know they're doing bullshit research
Existential Concerns
- Scientific progress feeds technological progress needed for species protection
- Future threats: super volcano eruptions, asteroid impacts, solar mega flares
- Could kill billions without technological protection
- Technological progress stalling due to academic problems
- First step: acknowledging the problem exists
