AI summary by TwinMind
How Scientists Can Avoid Cognitive Bias
Sabine Hossenfelder
8m 28s
Introduction to Cognitive Biases in Science
- Cognitive biases prevent science from working properly
- Knowledge from books like "Thinking Fast and Slow," "You Are Not So Smart," "Blind Spot" not applied to scientific research
- Scientists correct for statistical biases and measurement device biases but not brain biases
- Cognitive bias = thinking shortcut for faster decisions
Optical Illusion Analogy
- Cognitive biases work like optical illusions
- Example: Two squares (A and B) appear different colors but have identical pixel colors
- Brain factors in shadow, calculates back to original color
- Correct for natural scene judgment, wrong for pixel evaluation
- Solution: Don't rely on intuition when aware of bias
Attentional Bias
- More often you hear something → more important you think it is
- Usually beneficial in everyday life (flu example)
- Can draw attention to false/irrelevant information
- Social media artificially amplifies message prevalence
- Terrorism example: high media coverage creates worry despite low statistical risk
- In science: media coverage/colleague discussion leads to overrating topic's scientific relevance
Loss Aversion
- "Throwing good money after bad"
- Reluctance to abandon investments of time/money
- Continue investing even when no longer sensible
- Admitting mistake psychologically difficult
- Scientists continue unpromising research agendas due to this bias
Social Reinforcement (Group Think)
- Most problematic bias in science
- Occurs in closed, like-minded communities
- Members reassure each other they're doing right thing
- Develop overly optimistic narrative about own research
- Dismiss outside opinions
- Makes self-correction impossible
- Prevents identification of mistakes
Shared Information Bias
- More likely to pay attention to widely shared information
- Less attention to information held by few people
- Problematic for science: number of people knowing fact ≠ fact correctness
- Wide sharing shouldn't factor into correctness evaluation
Intelligence and Bias
- Studies show everyone has cognitive biases
- Intelligence doesn't reduce likelihood of having biases
- Should organize research to help scientists avoid/alleviate bias
Current Research Organization Problems
- Makes cognitive biases worse instead of better
- Difficult to change research topics
- Grant requirements: must document expertise in topic
- Hiring requires existing experience in topic
- Superficially good investment strategy (reward expertise)
- Long-term consequences: bad investment strategy
- Researchers face psychological + financial hurdles to topic change
- Forced to claim research direction promising
- Continue working on topics leading nowhere
- Rewards exaggerating research excitement
- Rewards working on popular topics
- Worsens social reinforcement and shared information bias
Solutions and Improvements
- Encourage researchers to change topics vs. forcing continuation
- Always list research topic shortcomings in lectures/papers
- Keep shortcomings in collective consciousness
- Conferences should have speakers from competing programs
- Encourage criticism of community without avoidance consequences
- Individual scientists can make these improvements
- Easy to generate further ideas once thinking about it
Importance and Conclusion
- Science drives innovation
- Poor science functioning → slowed societal progress
- Cognitive bias in science is solvable problem
- Should solve now
