Can we choose who we are? (with Gavin Leech)

Aug 16, 2023 Episode Page ↗
Overview

Spencer Greenberg speaks with Gavin Leach about the malleability of personality and preferences, the ongoing replication crisis in scientific research, the challenges of integrating prediction markets into policy, and the emerging implications of AI voice simulation on privacy and identity.

At a Glance
42 Insights
1h 7m Duration
11 Topics
7 Concepts

Deep Dive Analysis

Can We Choose Who We Are? Fixed vs. Changeable Identity

Interventions for Radical Personal Change

Personality Stability and Lifespan Changes

The Value of Generalism and 'Methods Fields'

Gavin Leech's List of Failed Replications and Its Impact

Debating the Interpretation of Replication Failures

Critiques and Proposed Reforms for Academic Peer Review

The Underutilization of Forecasting in Policy and Business

Barriers to Adopting Forecasts and Prediction Markets

The Threat of AI Voice Simulation and Deepfakes

Implications of AI Simulating Personal Identity

Adaptive Preferences

This concept suggests that what a person likes or prefers can change and adapt based on their actions and behaviors. By choosing what activities to engage in, individuals can, in a sense, choose parts of who they are by cultivating new preferences.

Methods Fields

These are academic disciplines like statistics, mathematics, philosophy, and computer science that provide foundational tools and principles applicable across a wide range of other fields. Proficiency in these areas allows individuals to contribute to and understand diverse research domains, acting as a 'generalist' advantage.

Replication Crisis

A significant concern in science, particularly social sciences, where many published research findings have been found difficult or impossible to reproduce when independent researchers attempt to conduct the same experiments. This raises questions about the reliability and robustness of some scientific literature.

Null Hypothesis Significance Testing (NHST)

A statistical framework often used in research to determine if an observed effect is likely due to chance. A common misuse is interpreting a non-significant result (failure to reject the null hypothesis) as proof that no effect exists, rather than simply a lack of sufficient evidence for an effect.

Post-Publication Review

A proposed alternative or supplement to traditional peer review where research papers are made publicly available (e.g., on preprint servers) before or immediately after formal publication, allowing for open, community-wide scrutiny and feedback after the work is released.

Judgmental Forecasting

A method of predicting future events by systematically eliciting and aggregating the probabilistic judgments of human forecasters, often associated with 'superforecasters' and prediction markets. It aims to produce more accurate and well-calibrated predictions than traditional expert opinions.

AI Simulation of Identity

The use of advanced artificial intelligence, such as large language models or speech generators, to mimic a specific individual's voice, writing style, or even their likely responses and beliefs, based on their existing digital footprint. This raises profound questions about privacy, control over one's persona, and the nature of identity in the digital age.

?
Can people truly choose or change who they are, including their personality and preferences?

While many adults perceive their identity as fixed, dramatic interventions (like intentionally depriving oneself of a habit), purposefully cultivating new hobbies, or immersing oneself in new social groups can lead to significant changes in behavior, preferences, and even aspects of personality.

?
How stable is personality in adulthood?

Personality traits, such as those measured by the Big Five, show relative stability in adulthood with year-to-year correlations typically between 0.6 and 0.7. However, this doesn't mean change is impossible, and some traits, like narcissism, can show significant decreases over the lifespan.

?
What is the 'replication crisis' in social sciences?

The replication crisis refers to the widespread observation that many published findings in fields like psychology have failed to be reproduced when independent researchers attempt to replicate the original studies, casting doubt on the robustness of some established scientific claims.

?
How effective are 'growth mindset' interventions?

While the concept of a growth mindset (believing one can improve through effort) is intuitively appealing, large-scale intervention studies designed to teach it have shown a very small effect size on academic performance, with a Cohen's D of 0.08.

?
What are common misconceptions about academic peer review?

Many laypeople misunderstand peer review, often believing it involves rigorous, line-by-line checking or blinded review. In reality, in many fields, reviewers often know the authors, rarely run code or check data, and the process primarily screens for facial absurdity rather than guaranteeing correctness or significance.

?
Why haven't forecasting and prediction markets seen wider adoption in policy and business, despite their demonstrated accuracy?

Uptake is slow because the accuracy advantage over experts might be smaller than sometimes claimed (3-10%), better probabilities don't always translate directly into clear actions, and there's resistance from existing expert systems or political motivations to avoid information that complicates policy advocacy.

?
How worried should we be about deepfakes and AI voice simulation?

While photo manipulation has existed for decades without a widespread crisis, AI-generated voice and video are more visceral and can craft more complete, convincing narratives. This could lead to spear phishing, international incidents, and a general downgrading of confidence in digital media evidence, forcing us to be more skeptical.

?
What are the implications of AI simulating a person's identity from their digital footprint?

AI models can be trained on a person's online text to simulate their writing style and even predict their responses to questions, raising deep privacy concerns about losing control over one's self-concept and digital representation, even after death, as a 'model' of oneself could interact with others without consent.

1. Embrace Activities Outside Identity

Purposely engage in activities that are contrary to your current self-perception or identity, as these are often the areas where you can experience the most significant personal growth and benefit.

2. Intentional Deprivation for Change

Intentionally deprive yourself of a habit or comfort (e.g., leaving computer at home to avoid video games) to force new behaviors and intellectual exploration, as this can lead to radical personal change.

3. Choose Influential Social Groups Wisely

Select your social groups carefully, as being part of a group that deeply values a certain behavior or interest can strongly influence you to adopt those values and change your own behavior accordingly.

4. Choose Your Preferences

Understand that your preferences are adaptive, meaning if you intentionally engage in activities, you can grow to like them, thereby choosing aspects of who you are.

5. Master Foundational ‘Methods Fields’

Gain a strong grasp of foundational ‘methods fields’ like statistics, mathematics, philosophy, and computer science, as these provide underlying principles and tools that grant you the license to work effectively across a wide variety of disciplines.

6. Cultivate Cross-Disciplinary Thinking

Embrace a generalist or polymathic approach by mixing ideas and methods from different fields, as this cross-pollination can provide a significant creative advantage for invention and discovery.

7. Undertake Challenging Projects

Engage in incredibly challenging projects, such as starting a business, as they force you to perform unprecedented tasks, fostering greater capability and agency.

8. Utilize Therapy for Radical Improvement

Consider therapy, with the right therapist and focus, as a potentially life-changing intervention to overcome issues like addiction, anxiety, or depression and radically improve your well-being.

9. Strategic Personality Change Possible

Recognize that while average personality traits may be stable, strategic and intentional interventions can lead to significant personal change beyond typical life drift.

10. Use Pretexts, Social Support

To change a self-concept (e.g., from intellectual to physically active), find a ‘pretext’ (e.g., scientific evidence for benefits) and seek out social support (e.g., a like-minded friend) to ease into new behaviors and identity.

11. Seek Life-Changing Interventions

Actively seek out dramatic, potentially life-changing interventions like a 10-day meditation retreat or Burning Man, as these can occasionally lead to profound personal transformation.

12. Immerse in Desired Social Groups

To understand or adopt new behaviors and perspectives, actively seek out and spend time with social groups that embody those traits, observing and trying their activities.

13. Challenge ‘Never Done It’ Barriers

Avoid not doing something simply because you’ve never done it before; challenge this as a ‘stupid reason’ to limit your experiences and personal growth.

14. Strategically Avoid Undesirable Preferences

Consciously decide not to cultivate preferences or hobbies that you deem expensive, risky, or otherwise undesirable, even if you know you could grow to like them.

15. Develop Broadly Applicable Thinking Strategies

Focus on developing powerful, cross-applicable thinking strategies, such as critical thinking, as these serve as broad tools that enhance your ability to analyze and solve problems in diverse contexts.

16. Sharpen Critical Thinking Through Philosophy

Engage with philosophical training or principles to sensitize yourself to sloppy arguments, ambiguities, and overloaded words, thereby improving the rigor and clarity of your own thinking and communication.

17. Utilize Forecasting for Policy Decisions

Recognize that judgmental forecasting (e.g., superforecasting) provides demonstrably better information than traditional expert opinion, making it incredibly valuable for improving policy decisions, even if the advantage seems small.

18. Supplement Probabilistic Forecasts with Causal Models

Understand that while probabilistic forecasts provide better information, they don’t mechanistically explain problems or dictate actions; supplement them with causal models (e.g., SEIR models for disease) to better guide policy and decision-making.

19. Sustain Prediction Markets with Incentives

To maintain the relevance and accuracy of internal prediction markets, ensure there are strong incentives for participation, sufficient volume of users, and consistent updating of forecasts.

20. Explore Paid Prediction Market Services

For businesses and nonprofits, seek out or advocate for prediction market services that offer guaranteed predictions on specific topics for a fee, providing a clear way to exchange money for valuable, crowd-sourced information.

21. Present Forecasts Appropriately for Audience

When communicating forecasts to decision-makers, be aware of their preference for clear, definitive answers; sometimes simplifying the presentation (e.g., removing error bars) is done to ensure the information is used, even if it’s a compromise.

22. Prioritize Arguments Over Authority

While respecting expertise, prioritize the strength of arguments and reasoning over mere authority; be willing to update your beliefs based on compelling evidence, even if it contradicts an expert’s initial claim.

23. Practice Deference to Experts

When engaging with experts, particularly when challenging their views, demonstrate that you have listened carefully and understood their arguments; this proper deference can make them more receptive to your perspective.

24. Proactively Unlearn Discredited Information

Actively identify and unlearn information that has been discredited or found to be less robust than initially thought (e.g., from replication failures), to avoid spreading misinformation and maintain an accurate understanding of the world.

25. Understand Field’s Methodological Limitations

As a scientist or researcher, critically understand the average methodological standards and their insufficiencies within your field to improve your own practice and contribute to better science.

26. Avoid Misinterpreting Replication Failures

When evaluating scientific findings, be cautious not to misinterpret a failed replication (e.g., not reaching statistical significance) as definitive proof that an effect doesn’t exist, as it might simply indicate a small effect size or insufficient power.

27. Understand True Scope of Peer Review

Recognize that peer review in most fields primarily ensures a paper is not facially absurd or below average for its field, rather than guaranteeing its absolute correctness or rigorous mathematical verification.

28. Design Tiered Credentialing for Research

Advocate for or design a system where all research can be published, but with tiered credentialing (from highly stringent expert review and data checks to basic publication) to indicate the level of scrutiny and reliability.

29. Implement Identity-Linked Comment Threads

Support or implement online comment threads for scientific papers where commenters’ identities are linked (e.g., showing their academic credentials), with anonymous comments potentially down-ranked, to foster accountable and high-quality critique.

30. Leverage Online Platforms for Critique

Utilize platforms like Twitter for rapid and public scientific criticism and discussion, and consider ways to incentivize high-quality, constructive ’trolling’ to improve the scientific discourse.

31. Advocate for Post-Publication Review

Consider advocating for or participating in systems of post-publication review, where research is made public first and then filtered and critiqued, as a potential improvement over traditional pre-publication peer review.

32. Recognize Journals’ Role in Attention

Understand that journals serve a valuable role in allocating attention to meaningful or noteworthy research, beyond just peer review, by acting as a filter for what is considered worth reading by a broader audience.

33. Be Wary of AI Voice Spoofing

Be highly cautious of unexpected or distressed voice calls, particularly from family members, as AI speech generators can spoof voices with minimal audio, making spear phishing and identity theft a significant risk.

34. Exercise Skepticism Towards AI Content

Develop a strong skepticism towards audio and video content, especially if it’s unusual or controversial, as AI can now generate highly convincing deepfakes for misinformation or malicious purposes.

35. Downgrade Confidence in Unverified Media

Proactively reduce your default confidence in unverified audio and video evidence, recognizing that advanced AI tools make it increasingly easy to create convincing fakes.

36. Be Aware of Deepfake Spread

Understand that AI-generated deepfake videos can craft highly specific and convincing narratives that spread rapidly online, potentially misleading millions before their inauthenticity is revealed.

37. Utilize Digital Signatures for Authenticity

For critical visual evidence, seek out or advocate for the use of cameras with digital signing modules that cryptographically certify image authenticity, making it possible to detect manipulation.

38. Be Mindful of AI Simulation

Be aware that AI can create highly accurate simulations of individuals based on their online text, potentially allowing others to interact with a ‘model’ of you without your involvement, raising new privacy and identity concerns.

39. Respect Autonomy Over AI Simulation

When considering creating an AI simulation of another person, even with disclaimers, respect their explicit wishes regarding such simulations, as individuals may not want their identity or beliefs represented by an AI model.

40. Establish Early Digital Identity

Consider establishing a substantial digital footprint of your original work (e.g., blogging) before advanced AI simulation becomes ubiquitous, as this could serve as a ‘cryptographically signed proof’ of your human authorship and thought.

41. Address Innumeracy in Decision-Makers

Recognize that a significant lack of basic probabilistic understanding (innumeracy) among policymakers can hinder the effective use of forecasts, suggesting a need for improved quantitative literacy.

42. Be Aware of Justification Bias

Understand that policymakers may seek forecasts not for objective decision-making, but for retroactive justification of already chosen policies, which can limit the impact of accurate forecasting.

Be a statistician. It lets you play in everyone else's backyard.

John Tukey

I've never had a problem publishing in mathematics. People observe my argument and test it extremely rigorously and then publish the linguist. But none of my work in political philosophy has ever gone unmolested, has ever made it through something like that.

Noam Chomsky

My brain is full of stuff, which I keep on telling people like excitedly at parties. I do not want to do that anymore.

Gavin Leech

This user has deleted his comment.

GPT-3 (simulating Wei Dai)

It's up to us to ensure its benefits to humanity while respecting privacy and individuality.

GPT-3 (simulating Gavin Leech's podcast wrap-up)

Cold Turkey Video Game Cessation

Gavin Leech
  1. Intentionally leave your computer at home (or remove access to video games).
  2. Be forced to engage in alternative activities, such as going to the library.
  3. Spend time on intellectual exploration and reading random subjects.

Identity Shift for Weightlifting

Gavin Leech
  1. Find a pretext or justification, such as researching the psychological and health benefits of weightlifting (e.g., for bones and joints).
  2. Find a friend who shares a similar 'nerdy' interest to engage with the activity together.
  3. Ease into the activity gradually, allowing your self-concept to expand to include both intellectual and physical aspects.
0.6 to 0.7
Correlation of Big Five personality traits (1 year apart) Indicating relative stability in adulthood.
1 standard deviation
Decrease in narcissism between ages 20 and 65 Observed in a 63-year longitudinal study.
60
Initial number of replication failures on Gavin's list Expanded to 400 effects by a volunteer group of psychologists.
0.08
Effect size (Cohen's D) of growth mindset interventions Observed in a large-scale study with tens to hundreds of thousands of participants, indicating a very small effect on academic performance.
57%
British MPs who incorrectly answered the probability of two coin flips resulting in heads-heads From a 2012 study, highlighting innumeracy among policymakers.
3% to 10%
Accuracy advantage of superforecasters over experts Based on a review of evidence, suggesting a smaller advantage than sometimes claimed.
1 minute or so
Minimum audio duration needed to spoof a voice using machine learning Making individuals vulnerable to deepfake audio attacks.