#55 Scott Page: Becoming a Model Thinker

Apr 2, 2019
Overview

Scott Page, Professor of Complex Systems at the University of Michigan, discusses how mental models help make sense of the world and solve complex problems. He emphasizes using an ensemble of models and cognitive diversity to address challenges too vast for individual understanding.

At a Glance
53 Insights
1h 24m Duration
15 Topics
7 Concepts

Deep Dive Analysis

Defining Mental Models and Their Application

The Evolution of Thinking: From Simple to Complex Models

Collective Intelligence and Ensembles of Models

Cognitive Diversity and the Wisdom Hierarchy

Applying Models: Data, Information, Knowledge, Wisdom

The Role of Linear Models vs. Human Judgment

Acquiring and Prioritizing Mental Models

Models as Tools for Surfacing Assumptions and Design

Why Traditional Education Teaches Single Models

Teaching Complexity and Meta-Thinking Skills

Experimenting with Collective Intelligence in Problem Solving

Power Law Distributions: Causes and Implications

Concavity and Convexity: Understanding Returns and Growth

Local Interaction Models and Cultural Coordination

Perspective Taking and the Pragmatic Use of Models

Mental Model

A mental model is a framework used to make sense of the world, mapping messy reality to clean, logical mathematical structures. It involves deciding which variables from reality to connect to an existing mathematical framework.

Collective Intelligence

This concept suggests that one way to make sense of complexity is by applying ensembles of models to problems, as no single individual's mind is big enough to grasp the full complexity of the world. Different models may explain different percentages of variation, and their combined insights lead to a deeper understanding.

Wisdom Hierarchy

This hierarchy describes the progression from raw data to information (structured data), then to knowledge (understanding relationships between information), and finally to wisdom (understanding which knowledge to apply to a specific problem). Mental models are primarily applied at the knowledge stage, with wisdom discerning their relevance.

Cognitive Diversity

This refers to a group of people having different mental models and basic assumptions about how the world works, leading them to see different parts of a problem. It is valuable because it allows for a more holistic understanding of complex issues that no single person can fully grasp.

Power Law Distributions

These are distributions where most events are very small, with occasional huge events, unlike normal distributions where values cluster around an average. They are characterized by a long tail and can be caused by mechanisms like preferential attachment, random walks, or self-organized criticality.

Concavity and Convexity

These terms describe how the value or outcome changes with additional input. Concavity implies diminishing returns, where the added value of each successive unit decreases (e.g., eating more chocolate cake). Convexity implies increasing returns, where the added value of each successive unit increases (e.g., preferential attachment in book sales).

Local Interaction Models

These models describe situations where an individual's behavior depends on the behavior of those around them, often leading to coordination games. The specific action itself may not be inherently important, but coordinating with others on that action is crucial, shaping culture and organizational practices.

?
What are mental models?

Mental models are frameworks used to understand the world, essentially mapping complex reality onto logical mathematical structures. They help us identify relevant variables and relationships to make sense of situations.

?
Why is it difficult to solve complex problems like the obesity epidemic with a single approach?

Complex problems like the obesity epidemic have multiple contributing dimensions (e.g., infrastructure, food, gut bacteria, work-life balance) and systems-level feedbacks. Fixing only one piece is insufficient due to the interconnectedness and lack of a single 'silver bullet' solution.

?
How does cognitive diversity contribute to problem-solving?

Cognitive diversity means having people with different mental models and basic assumptions in a group, allowing them to see different parts of a problem. This collective perspective leads to a deeper, more holistic understanding and better solutions for complex issues.

?
How does the 'wisdom hierarchy' connect data, information, knowledge, and wisdom?

The hierarchy starts with raw data, which is structured into information. Knowledge is the understanding of relationships between pieces of information, and wisdom is the ability to discern which knowledge (or models) to apply to a specific problem.

?
When should one trust a linear model versus human judgment for predictions?

If a linear model and human predictions are close, the linear model is often preferred due to its calibration and ability to process more data without bias. However, if predictions are far apart, it's crucial to investigate what variables humans are considering that the linear model isn't, or if the environment has changed, as linear models assume stationarity.

?
How should individuals prioritize which mental models to learn?

Prioritization depends on the context of the action (individual choice, strategic situation, or social system), the rationality of the decision-maker (rational vs. rule-of-thumb), and whether the logic of the model aligns with the observed reality. Considering these factors helps determine the most relevant models.

?
Why are we traditionally taught to use single models in school, and why is this approach becoming insufficient?

Single models were effective in a simpler world where problems could be 'carved at their joints' and explained by simple equations. However, modern decisions are far more complex, involving numerous interconnected dimensions (e.g., environmental impact, talent attraction, brand), making a single model inadequate.

?
What are the implications of power law distributions for success and fairness?

Power law distributions mean that most outcomes are small, with a few extremely large ones (e.g., book sales). This suggests that wild success can be as much due to luck and positive feedback mechanisms (like preferential attachment) as it is to skill, potentially challenging traditional notions of merit and fairness.

?
How do concave and convex functions relate to real-world phenomena?

Concave functions represent diminishing returns, where each additional unit provides less benefit (e.g., adding workers to a team beyond a certain point). Convex functions represent increasing returns, where each additional unit provides more benefit (e.g., network effects or preferential attachment). Recognizing these helps avoid dangerous linear projections of growth or impact.

1. Cultivate Cognitive Diversity

To achieve a deeper and more holistic understanding of complex problems, assemble a collection of people with diverse knowledge, overlapping expertise, and different mental models.

2. Employ Ensemble Thinking

To make sense of complex problems, use an ensemble of models, as no single model can fully explain the variation, and combining them offers a deeper understanding.

3. Align Passion, Ability, Purpose

When choosing a path, seek something that combines three elements: a genuine love for the practice, some innate ability, and a connection to something useful, meaningful, or purposeful that makes the world better.

4. Practice Perspective Taking

Cultivate the meta-model of perspective-taking by actively considering how a problem appears through the lens of various stakeholders, including individuals, shareholders, and government, to gain a nuanced view.

5. Strategize Model Acquisition

Decide whether to specialize deeply in a few models, become proficient in a handful, or gain broad awareness of many models, tailoring your approach to your career and cognitive strengths.

6. Balance Individual, Collective Effort

While lifelong learning and amassing models are valuable, recognize that individual capacity is limited for solving large-scale problems; combine individual efforts with collections of people using diverse models for greater impact.

7. Practice Deliberative Modeling

Construct models through a deliberative process: first, identify the general class of the problem (system, decision, game), then write down the model, use mathematics to deduce implications, and revise if they don’t match reality.

8. Use Logic-Structure-Function

When observing a pattern or structure in the world, apply the “Logic-Structure-Function” framework: identify the underlying logic that created it, and then assess its functionality and whether it matters.

9. Design Reality with Models

Recognize that in leadership or policy roles, constructing models is not just about understanding reality but also about actively designing it by defining rules, policies, and strategy spaces.

10. Integrate Models, Human Insight

For best results, combine quantitative models (like linear regressions) with human insights, leveraging the strengths of both.

11. Triangulate Model Predictions

When quantitative models and human predictions are similar, trust the calibrated linear model; if they diverge significantly, engage with the humans to understand their differing perspectives and variables.

12. Expand Relevant Model Toolbox

Accumulate a diverse range of mental models, viewing them as tools in a toolbox, to be better equipped for a wide variety of problems, but prioritize relevance to the task at hand.

13. Build Model Latticework

Array your experiences on a latticework of mental models to better understand and interpret the world, drawing from a wide range of frameworks.

14. Surface Model Assumptions

Utilize models as a tool to surface underlying assumptions, ensure logical consistency, identify key drivers of behavior, understand interactions and aggregation, and inform appropriate responses.

15. Define Problem Actors, Context

When approaching a problem, first identify the relevant actors and the decision context, determining if it’s a single actor decision, a strategic interaction, or a larger system.

16. Understand Aggregation Issues

Be acutely aware that individual actions or components may not aggregate predictably, and fundamental paradoxes can arise from underlying assumptions.

17. Consider Systemic Feedbacks

Recognize that systems often contain internal feedbacks that can lead to logical inconsistencies or prevent universal success if everyone follows the same “formula.”

18. Anticipate Competitor Models

Develop an understanding of the mental models your competitors use, as this allows you to anticipate their responses, avoid surprises, and gain a strategic advantage.

19. Test Models, Avoid Force-Fit

Apply a variety of models but be pragmatic: if a model doesn’t fit the problem (“spaghetti doesn’t stick”), let it go for that specific case, but keep it in your toolbox for future applicability.

20. Prioritize Pragmatic Utility

When selecting models or perspectives, prioritize those that offer pragmatic utility and genuinely help achieve your specific goals or solve the problem at hand.

21. Instill Multi-Model Thinking

Teach children to develop a mental latticework of models, even without deep mathematical understanding, to foster an appreciation for the world’s complexity beyond single explanations.

22. Master Core Universal Models

Prioritize learning core, unchanging models that apply across disciplines, human history, biology, and even physics, as these provide fundamental insights into persistent patterns.

23. Grasp Concavity, Convexity

Understand concavity as diminishing returns, where the added value of each successive unit decreases, and convexity as increasing returns, where the added value increases.

24. Beware Linear Projections

Be cautious of linear thinking and projections, as they can be dangerous and lead to flawed assumptions by failing to account for diminishing or increasing returns.

25. Grasp Power Law Dynamics

To understand power law distributions, recognize that preferential attachment, random walks, and self-organized criticality are key mechanisms driving these “long tail” phenomena.

26. Grasp Local Interaction Models

Understand local interaction models as frameworks where an individual’s behavior is influenced by the behaviors of those immediately around them, often leading to coordination.

27. Balance Luck, Skill Attribution

When evaluating success, especially in power law distributed outcomes, recognize that “big winners” often result from a combination of skill and the natural processes of positive feedback, rather than solely superior ability.

28. Reframe “Weird” Behaviors

Instead of labeling unfamiliar behaviors as “weird” or intrinsic character traits, reframe them as solutions to local coordination problems within a specific cultural or family context.

29. Value Individual Model Diversity

Even if another’s mental model seems superior, retaining your own diverse model is valuable for the collective, as it contributes to overall cognitive diversity and problem-solving.

30. Adapt to Other’s Models

To interact efficiently and predictably within an organization or community, adapt to and coordinate with other people’s mental models and terminology, understanding their worldview.

31. Hire Diverse Perspectives

When facing complex decisions, hire individuals who can bring different disciplinary perspectives (e.g., pharmacological, sociological, organizational, political) to the table.

32. Seek Diverse History Views

Avoid a singular “great man” view of history; instead, seek out and engage with diverse perspectives from all people who experienced an event to gain a more complete understanding.

33. Bridge AI, Real-World Understanding

Cultivate the ability to communicate and translate between sophisticated artificial intelligence models and real-world contexts, acting as a bridge between technical and practical understanding.

34. Interrogate Model Assumptions

When model predictions diverge from human insights, critically examine the model’s variables, coefficients, and whether environmental conditions have changed, while also understanding human-considered variables.

35. Identify Missing Model Variables

When a model fails to predict accurately, investigate whether crucial variables, especially qualitative ones like “ugly,” are missing from its construction.

36. Gauge Decision Rationality

When analyzing decisions, assess the rationality of the decision-maker, considering whether they are optimizing or primarily relying on rules of thumb.

37. Anticipate Rationality Drivers

Expect decision-makers to exhibit more rational behavior in situations that are frequently repeated, allowing for learning, or when the stakes involved are exceptionally high.

38. Distinguish Org Decision Modes

Recognize that organizational decisions can range from careful, rational committee processes for big choices to rigid, rule-based standard operating procedures for routine tasks.

39. Validate Complexity Claims

If you argue for complexity, path dependence, or policy impact, ensure your model accounts for this by demonstrating the creation of new states or fundamental changes in transition probabilities.

40. Gain Model Robustness, Bonus

A variety of models provides robustness, akin to a portfolio, reducing the likelihood of mistakes, and often yields a “bonus” where combined performance exceeds the average of individual models.

41. Use Awareness to Guide Deeper Learning

Develop a broad awareness of various models to identify those that particularly resonate or seem relevant, then use this interest to guide deeper exploration.

42. Structure Raw Data

Convert raw, overwhelming data into structured information by categorizing and summarizing it into meaningful variables, like unemployment or inflation rates.

43. Build Relational Knowledge

Progress from information to knowledge by understanding the correlative or causal relationships between different pieces of information, such as force equals mass times acceleration.

44. Apply Knowledge Wisely

Develop wisdom by discerning which specific knowledge to apply to a given problem, which may involve selecting from or combining various pieces of knowledge.

45. Understand Distribution Impact

Understand the practical implications of different distribution types (e.g., normal vs. power law), as they profoundly affect fairness, predictability, and design challenges in systems.

46. Grasp Distribution Logics

Learn the underlying logic that creates different statistical distributions, such as addition for normal, multiplication for log-normal, and preferential attachment or self-organized criticality for power laws.

47. Apply Ideas Cross-Discipline

Actively seek opportunities to apply broad concepts and models across various disciplines to deepen understanding and reveal their universal applicability.

48. Seek Broader Impact

Strive to make your work interesting and intriguing to a wider audience beyond a small circle, as this increases its value and potential for contribution.

49. Acknowledge Individual Cognitive Limits

Recognize that no single individual’s cognitive capacity is sufficient to fully comprehend the sheer complexity and dimensionality of the modern world.

50. Recognize Model Bias

Be aware that your existing mental models can bias how you filter raw data into information, as they influence which variables you deem relevant and how you perceive their interactions.

51. Consider Systemic Embeddedness of Actions

Evaluate whether your actions are embedded within a larger social system, where you might be unconsciously influenced by cues from that system.

52. Recognize Routine-Based Decisions

Acknowledge that many daily decisions are based on routines and rules of thumb rather than rational optimization, and these routines may adapt slowly over time.

53. Verify Personal Logic

After considering how others make decisions, critically question and validate the correctness of your own underlying logic.

Your brain just isn't going to be big enough. Collections of people, creating a larger ensemble model, actually have a hope of addressing these problems.

Scott Page

What a mental model is, is a framework that you use to make sense of the world.

Scott Page

Munger has this wonderful quote about, you want to sort of array your experiences on the lattice work of models.

Scott Page

No individual's head is big enough to sort of make sense of the complexity of the world.

Scott Page

But if they're far apart, if the linear model and the humans are giving very different predictions, then you want to go talk to them.

Scott Page

Imagine how difficult physics would be if electrons could think.

Murray Gell-Mann (quoted by Scott Page)

But in a complex world, your ability to succeed is going to depend on you sort of filling a niche that's valuable.

Scott Page
10% to 30%
Variation explained by social science models Typical range for models explaining real-world variation, implying a large missing percentage.
70% to 90%
Unexplained variation by social science models The remaining variation not explained by single simple models.
400,000 units
Printer unit sales predicted by linear model Example of a linear model's prediction for a consumer product.
200,000 units
Printer unit sales predicted by crowd Example of a crowd's prediction for the same consumer product, significantly different from the linear model.
30
Number of models in 'The Model Thinker' book Scott Page's selection of important mental models.
7 to 12 pages
Page length for explaining each model in the book Aimed at making complex models understandable for a general audience.
20
Number of PhDs that could be written on each model Illustrates the depth and complexity of each model beyond the book's overview.
100 by 100
Grid size for collective intelligence problem-solving experiment Represents the landscape for finding the highest value point.
10 points (in 5 rounds)
Number of points physicists checked in the experiment Their allocation for exploring the grid to find the highest point.
300 or 400 copies
Average book sales Most books sell this many copies, illustrating the long tail of power law distributions.
millions of copies
Bestselling book sales The occasional huge event in a power law distribution of book sales.
10,000
Hypothetical number of people as tall as giraffes if heights followed a power law Used to illustrate the extreme outcomes of a power law distribution compared to a normal distribution.
170 million
Hypothetical number of people 7 inches tall in the US if heights followed a power law Used to illustrate the extreme outcomes of a power law distribution compared to a normal distribution.