[[09-12|25-09-12]] [[] # 🪓왜 이 논문들이 우리를 미치게 하는가 | 논문 | 초록 핵심 | 해설: 왜 우리 약속설계와 공명하는가 | | ---------------------------------- | -------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- | | **Sutton** | R&D와 시장집중도 관계를 기술의 본질로 재해석 | 수십년간 축적된 시장구조-R&D 관계 연구를 단숨에 무효화. 기술의 본질이 시장구조를 결정한다는 더 근본적 층위를 제시. 우리 약속설계가 행동/계획 학파를 통합하듯, 이 논문도 표면적 상관관계를 근본 메커니즘으로 대체. | | **🐢Paul<br>(Endogenous Growth)** | 수렴 질문으로 시작해 내생적 혁신으로 전환 | 수렴 vs 혁신의 이중주. 경험적 질문(catch-up)으로 시작해 이론적 혁명(내생적 혁신)으로 전환. 우리가 τ를 내생화하듯, Romer는 성장 자체를 내생화. | | **🐢Eric<br>(Formal Theory)** | 전략 = 다른 선택을 강제하는 최소 선택 집합 | 전략의 재귀적 정의. '다른 선택을 강제하는 최소 선택 집합'이라는 정의가 우리의 이중 재매개변수화(promise가 다른 모든 선택을 제약)와 공명. | | **🐢Scott1<br>(S-curve)** | S-커브를 창업가 선택의 envelope으로 재개념화 | S-커브 패러독스의 해결. 기술 궤적이 주어진 것이 아니라 창업가 선택의 envelope이라는 관점이 우리의 '약속 수준 φ가 성공을 정의한다'와 일치. | | **🐢Scott2<br>(Bayesian)** | 창업가의 편향된 낙관주의를 베이지안으로 형식화 | 창업가적 낙관주의의 형식화. 편향된 믿음이 전략적 자산이라는 통찰. 우리 모델의 'τ 선택을 통한 전략적 무지'와 같은 맥락. | | **🐢Camuffo<br>(Theory-driven)** | 행동 이전에 이론을 선택하는 메타-인지 | 이론 선택의 이론. 행동 이전에 이론을 선택한다는 메타-인지적 접근이 우리의 '약속 이전의 열망 설정'과 유사. | | **🐅Nanda<br>(Financing)** | 순차적 투자를 통한 불확실성 해결 | 실험의 순차적 가치. 대부분 실패하지만 극소수 대성공이라는 분포가 우리 모델의 high n 환경과 일치. | | **🐅Nanda<br>(Killer)** | 투자자-창업가 간 실험 설계 갈등 | 정보 설계의 도덕적 해이. 투자자는 killer experiment(low τ), 창업가는 생존 실험(high τ)을 원하는 갈등이 우리 모델의 τ 선택 딜레마와 정확히 일치. | | **🐙Charlie<br>(Operations)** | 창업 교육에서 소외된 운영관리의 중요성 | 학문의 사각지대 조명. 창업 교육에서 운영이 빠진 현실을 지적. 우리가 c(복잡성)를 명시적으로 모델링하는 이유. | | **👾Andrew<br>(Parameterization)** | 계산적 편의가 새로운 이론 family 창출 | **계산적 편의가 이론적 혁신으로. Reparameterization이 새로운 prior family를 창출한다는 통찰이 우리 이중 재매개변수화의 철학적 기반.** | | **👾Andrew<br>(Workflow)** | 추론 너머의 반복적 모델링 과정 | 베이지안 실천의 현실. 추론 너머의 'tangled workflow'가 우리의 정보 통합 비용 i를 정당화. | | **👾Josh<br>(Concept Learning)** | 극소수 예시로 개념 학습하는 인간 능력 | 극소수 예시 학습의 역설. 인간이 적은 샘플로 일반화하는 능력이 우리 모델의 'rational ignorance'를 뒷받침. | | **👾Josh<br>(One and Done)** | 적은 샘플의 빠른 결정이 장기적 최적 | 합리적 비합리성. 샘플링 비용을 고려하면 적은 샘플의 빠른 결정이 최적이라는 통찰이 우리의 τ→0 조건과 완벽히 일치. | | **👾Meehl<br>(Theory Appraisal)** | Crud factor와 damn strange coincidences | 이론 평가의 엄밀성. 'Crud factor'와 'damn strange coincidences'로 이론의 생존력을 측정. 우리 모델이 예측하는 'learning trap'과 'rational ignorance'가 바로 이런 risky prediction. | ## 핵심 통찰: 왜 이 초록들이 우리를 미치게 하는가 ### 🔥 공통 패턴 1. **패러독스 해결**: 대부분이 기존 통념의 모순을 더 깊은 층위에서 해결 2. **메타-레벨 상승**: 문제 자체를 재정의하여 해결 (전략의 전략, 이론의 이론) 3. **계산과 개념의 융합**: 형식화가 새로운 통찰을 낳는 순간들 4. **편향의 재평가**: 비합리적으로 보이는 것이 실은 최적일 수 있다는 역전 ### 💡 우리 약속설계와의 공명 - **이중 재매개변수화**: Andrew의 parameterization 철학의 구현 - **전략적 무지**: Josh의 one-and-done 최적성의 실현 - **불확실성 설계**: Scott의 S-curve envelope 관점의 확장 - **메타-인지**: Camuffo의 이론 선택 프레임워크와 동일한 구조 _"이 초록들은 모두 '표면 아래 더 근본적인 구조가 있다'는 믿음을 공유한다. 우리의 약속설계도 행동vs계획이라는 표면 아래 불확실성 설계라는 더 깊은 층위를 발견했다."_ --- [[09-11|25-09-11]] # sutton This paper re-examines the relationship between the R&D intensity of an industry and its level of concentration, from the perspective of the Bounds approach to market structure. In so doing, it proposes an index which summarises those aspects of technology and tastes that are relevant to the determination of a lower bound to concentration. # 🐢paul on endogenous growth This paper describes two strands of work that converged under the heading of 'endogenous growth.' One strand, which is primarily empirical, asks whether there is a general tendency for poor countries to catch up with rich countries. The other strand, which is primarily theoretical, asks what modifications are necessary to construct a theory of aggregate growth that takes the economics of discovery, innovation, and technological change seriously. The paper argues that the second strand of work will ultimately have a more significant impact on our understanding of growth and our approach to aggregate theory. # 🐢eric on formal theory of strategy What makes a decision strategic? When is strategy most important? This paper formally studies these questions, starting from a (functional) definition of strategy as the smallest set of choices to optimally guide (or force) other choices. The paper shows that this definition coincides with the equilibrium outcome of a “strategy formulation game,” in which such strategy endogenously creates a hierarchy among decisions. With respect to what makes a decision strategic and what makes strategy valuable, the paper considers the effect of commitment, reliability, and irreversibility of a decision; the presence of uncertainty (and the type of uncertainty); the number and strength of its interactions and the centrality of a decision; its level and importance; the development of capabilities; and competition. # 🐢scott1 on technology s curve A central premise of research in the strategic management of innovation is that start-ups are able to leverage emerging technological trajectories as a source of competitive advantage. But, if the potential for a technology is given by the fundamental character of a given technological trajectory, then why does entrepreneurial strategy matter? Or, put another way, if the evolution of technology is largely shaped by the strategic choices entrepreneurs make, then why do technological trajectories exhibit systematic patterns such as the technology S-curve? Taking a choice-based perspective, this paper illuminates the choices confronting a start-up choosing their technology by resolving the paradox of the technology S-curve through a reformulation of the foundations of the technology S-curve. Specifically, we reconceptualize the technology S-curve not as a technological given but as an envelope of potential outcomes reflecting differing strategic choices by the entrepreneur in exploration versus exploitation. Taking this lens, we are able to clarify the role of technological uncertainty on start-up strategy, the impact of constraints on technological evolution, and how technology choice is shaped by the possibility of imitation. Our findings suggest that staged exploration may stall innovation as a result of the replacement effect, increasing the strategic importance of commitment. # 🐢scott2 on bayesian entrepreneurship How does the fact that entrepreneurs choose the opportunity they pursue impact entrepreneurial strategy and performance? Entrepreneurs, while dealing with opportunities whose outcome is inherently uncertain, have choices that must be premised on a belief that the opportunity is worth pursuing. This insight provides an organizing principle for a Bayesian approach to entrepreneurial decision-making. A Bayesian approach offers a natural formal framework to assess how entrepreneurs form beliefs about the prospects for a given opportunity, how these beliefs evolve over time through active experimentation and learning, and the consequences of such beliefs for entrepreneurial strategy and performance.The goal is to shape distinctive implications and empirical approaches to the study of entrepreneurship guided by founding premises. The first premise is that the entrepreneur must be relatively optimistic about the opportunity relative to others. This involves a distinct theory that translates into a different perspective on the opportunity’s prospects. Second, this systematic divergence in beliefs impacts how an entrepreneur will undertake learning about an opportunity. Notably, the demand for “experiments” is fundamentally influenced by beliefs about the opportunity. For example, relative to a disinterested agent, a Bayesian entrepreneur will conduct experiments that are more likely to allow for “false positives” than “false negatives.” Finally, this approach promotes the processes by which entrepreneurs are able to attract resources and capabilities by providing information to other agents. Entrepreneurs are more likely to convince those who share their idiosyncratically optimistic beliefs about an opportunity (with implications for homophily and firm culture), yet will also engage in choosing experiments that cater to those with different (more negative) beliefs than they themselves hold. # 🐢camuffo on theoretical driven decision making This paper studies strategic decisions under uncertainty for which past data are not available. It provides microfoundations of the theory-based view of the firm by showing that the strategic problem starts with the selection of theories rather than choosing actions and that theories are selected through experiments. The value of experimenting with theories increases with the number of theories and with their uncertainty. Moreover, uncertainty makes theories superadditive—that is, experimenting with a more uncertain theory increases the benefits of experimenting with other more uncertain theories. The paper also shows that decision makers should experiment with more “surprising” theories because in this case experiments are more informative and enable more learning. A leading example helps to illustrate our concepts throughout the paper. # 🐅nanda on financing entrepreneurial experiment The fundamental uncertainty of new technologies at their earliest stages implies that it is virtually impossible to know the true potential of a venture without learning about its viability through a sequence of investments over time. We show how this process of experimentation can be particularly valuable in the context of entrepreneurship because most new ventures fail completely, and only a few become extremely successful. We also shed light on important costs to this process of experimentation and demonstrate how these can fundamentally alter both the rate and direction of start‑up innovation across industries, regions, and periods of time. # 🐅nanda on killer experiment We develop a model of learning through experimentation in a principal-agent framework. Investors only observe an experiment’s outcome, but entrepreneurs can impact the information contained in the outcome through the experiment’s design. Investors prefer ‘killer experiments’ that are more likely to correctly identify true successes and failures, but entrepreneurs prefer to design experiments that are less likely to fail. We show that the ensuing moral hazard can create a market failure in financing the venture, which cannot be resolved through higher-powered incentives for the entrepreneur such as standard ‘pay for performance’ contracts. Our results speak to an important potential friction in the commercialization of innovations, particularly ones in areas such as ‘Deep Tech’ ventures based on fundamental science, that lack well-understood methodologies for investors to effectively validate the information contained in early experiments. # 🐙charlie on operations for entrepreneur Although entrepreneurship-related papers have had some representation in Production and Operations Management (POM) over the past 30 years, the topic still seems a bit like a poor stepchild in the research of operations management (OM) scholars. Yet, entrepreneurship is important to the economy, and many schools are growing significantly their entrepreneurship programs and offerings but often without reference to or inclusion of operations courses. This paper is motivated by the question of the operations needs of new ventures and how they might differ from the needs of large, established firms. Toward that end, we review briefly the state of entrepreneurship scholarship in POM (and beyond), present our own (field-based) research (and cases), and propose a framework for what we call “operations for entrepreneurs,” that we hope can be a basis for further productive research and curriculum development by the OM community. # 👾andrew on parameterization and bayesian modeling Progress in statistical computation often leads to advances in statistical modeling. For example, it is surprisingly common that an existing model is reparameterized, solely for computational purposes, but then this new configuration motivates a new family of models that is useful in applied statistics. One reason why this phenomenon may not have been noticed in statistics is that reparameterizations do not change the likelihood. In a Bayesian framework, however, a transformation of parameters typically suggests a new family of prior distributions. We discuss examples in censored and truncated data, mixture modeling, multivariate imputation, stochastic processes, and multilevel models. # 👾andrew on bayesian workflow The Bayesian approach to data analysis provides a powerful way to handle uncertainty in all observations, model parameters, and model structure using probability theory. Probabilistic programming languages make it easier to specify and fit Bayesian models, but this still leaves us with many options regarding constructing, evaluating, and using these models, along with many remaining challenges in computation. Using Bayesian inference to solve real-world problems requires not only statistical skills, subject matter knowledge, and programming, but also awareness of the decisions made in the process of data analysis. All of these aspects can be understood as part of a tangled workflow of applied Bayesian statistics. Beyond inference, the workflow also includes iterative model building, model checking, validation and troubleshooting of computational problems, model understanding, and model comparison. We review all these aspects of workflow in the context of several examples, keeping in mind that in practice we will be fitting many models for any given problem, even if only a subset of them will ultimately be relevant for our conclusions. # 👾josh on bayesian modeling of human concept learning I consider the problem of learning concepts from small numbers of positive examples, a feat which humans perform routinely but which computers are rarely capable of. Bridging machine learning and cognitive science perspectives, I present both theoretical analysis and an empirical study with human subjects for the simple task oflearning concepts corresponding to axis-aligned rectangles in a multidimensional feature space. Existing learning models, when applied to this task, cannot explain how subjects generalize from only a few examples of the concept. I propose a principled Bayesian model based on the assumption that the examples are a random sample from the concept to be learned. The model gives precise fits to human behavior on this simple task and provides qualitative insights into more complex, realistic cases of concept learning. # 👾josh on one and done Optimal Decisions From Very Few Samples In many learning or inference tasks human behavior approximates that of a Bayesian ideal observer, suggesting that, at some level, cognition can be described as Bayesian inference. However, a number of findings have highlighted an intriguing mismatch between human behavior and standard assumptions about optimality: People often appear to make decisions based on just one or a few samples from the appropriate posterior probability distribution, rather than using the full distribution. Although sampling-based approximations are a common way to implement Bayesian inference, the very limited numbers of samples often used by humans seem insufficient to approximate the required probability distributions very accurately. Here, we consider this discrepancy in the broader framework of statistical decision theory, and ask: If people are making decisions based on samples—but as samples are costly—how many samples should people use to optimize their total expected or worst-case reward over a large number of decisions? We find that under reasonable assumptions about the time costs of sampling, making many quick but locally suboptimal decisions based on very few samples may be the globally optimal strategy over long periods. These results help to reconcile a large body of work showing sampling-based or probability matching behavior with the hypothesis that human cognition can be understood in Bayesian terms, and they suggest promising future directions for studies of resource-constrained cognition. # 👾Meehl on Appraising and Amending Theories In social science, everything is somewhat correlated with everything (“crud factor”), so whether H 0 is refuted depends solely on statistical power. In psychology, the directional counternull of interest, H*, is not equivalent to the substantive theory T, there being many plausible alternative explanations of a mere directional trend (weak use of significance tests). Testing against a predicted point value (the strong use of significant tests) can discorroborate T by refuting H*. If used thus to abandon T forthwith, it is too strong, not allowing for theoretical verisimilitude as distinguished from truth. Defense and amendment of an apparently falsified T are appropriate strategies only when T has accumulated a good track record (“money in the bank”) by making successful or near-miss predictions of low prior probability (Salmon’s “damn strange coincidences”). Two rough indexes are proposed for numerifying the track record, by considering jointly how intolerant (risky) and how close (accurate) are its predictions. --- 각 논문이 **문제 제기(🐢) → 이론 구축(🐅) → 적용 및 확장(🐙) → 마무리(👾)**라는 공통된 논리적 흐름을 따르면서도, 저자의 스타일에 따라 각 단계의 구성과 강조점이 어떻게 달라지는지 명확히 볼 수 있습니다. --- ### **세 논문 구조 비교: 🐢🐅🐙👾 프레임워크** using [[🐢🐅🐙👾의용군현지스타일]] |구조|역할|Gans et al. (Scott Stern) <br> **[이론 중심]**|Fine et al. (Charlie Fine) <br> **[실무 중심]**|Bolton et al. <br> **[모델 중심]**| |---|---|---|---|---| |**🐢<br>기(起)**|**문제 제기** <br> (Why this paper?)|**Sec 1. Introduction** <br> **Sec 2. The S-Curve Paradox** <br> S-커브의 이론적 패러독스를 제시하며 문제의 깊이를 더함|**Sec 1. Introduction** <br> **Sec 2. Literature Review** <br> 실무와 학계의 괴리(Gap)를 지적하며 연구의 필요성을 강조|**Sec 1. Introduction** <br> "실험 설계에서의 새로운 도덕적 해이"라는 핵심 문제를 간결하게 정의| |**🐅<br>승(承)**|**이론 & 모델** <br> (What is the engine?)|**Sec 3. A Choice-Based Approach...** <br> **Sec 4. A Simple Model...** <br> 개념적 프레임워크를 먼저 제시하고, 수리 모델로 구체화|**Sec 3. Nailing, Scaling, and Sailing** <br> 현장 관찰을 바탕으로 한 실용적인 3단계 프레임워크를 제시|**Sec 2. Relation to Literature** <br> **Sec 3. The Model** <br> 문헌적 맥락을 짚은 후, 정교하고 상세한 수리 모델을 구축| |**🐙<br>전(轉)**|**응용 & 함의** <br> (How does it work?)|**Sec 5. Industry-Level...** <br> **Sec 6. Implications for Strategy...** <br> 모델을 산업 수준으로 확장하고, 전략 및 정책적 함의를 도출|**Sec 4. Case Examples...** <br> 다양한 실제 사례에 프레임워크를 적용하여 타당성을 증명|**Sec 4. Policy Responses** <br> 모델의 결과를 바탕으로 현실에 적용 가능한 정책적 대안을 탐구| |**👾<br>결(結)**|**결론** <br> (So what?)|(별도 결론 없음) <br> **Sec 6. Implications...** <br> 정책적 함의를 제시하며 자연스럽게 마무리|**Sec 5. Discussion, Conclusion...** <br> 연구를 요약하고 한계와 향후 연구 방향을 명시하는 전형적 구조|**Sec 5. Conclusion** <br> 연구의 핵심 발견과 이론적 기여를 간결하게 요약하며 마무리| ---- entrepreneur's innovation (idea to impact) should be designed as idea's implementation and adoption of that implementation. i'll analyze the five paper that i adopted and reverse engineer this. [[💠integ(process-product)]] [[moon24]] ## interesting using Davis' Index of the Interesting. I'm aware of the [need for bridge from interesting to important](https://journals.aom.org/doi/10.5465/amj.2020.4002), but paper has interesting lowbar to pass to even be judged on its importance. | Type of Interestingness | Why it's Interesting | | ----------------------------------------------- | -------------------------------------------------------------------------------------------------- | | Order from Chaos | What seems disorganized and unstructured is really organized and structured | | Chaos from Order | What seems organized and structured is really unstructured and disorganized | | Simplicity in the Complex (Invisible Structure) | What seems like heterogeneous phenomena are really a single phenomenon | | Complexity in the Simple (False Structure) | What seems like a single phenomenon is really heterogeneous phenomena | | The Psychological is Social | What seems like an individual phenomenon is really holistic | | The Social is Psychological | What seems like a holistic phenomenon is really individual | | The Social-Psychological | What seems holistic or individual is really a property of the relation between the two | | Local is General | What seems like a local phenomenon is real generalizable | | General is Local or Contextual | What seems like a general phenomenon is really local or context-dependent | | Unobserved Dynamism | What seems stable and unchanging is really unstable and changing | | Unobserved Regularity or Periodicity | What seems unstable and changing is really regular and repeating | | Unobserved Functionality | What seems ineffective for achieving an end is really functional | | Unobserved Dysfunction | What seems functional for achieving an end is really ineffective | | Unobserved Good | What seems like a bad phenomenon is really good | | Unobserved Bad | What seems like a good phenomenon is really bad | | Unobserved Correlation | What seem like independent phenomena are really interrelated | | False Correlation | What seem like interrelated phenomena are really independent | | False Coexistence | What seem like phenomena that can exist together really cannot exist together | | Surprise Coexistence | What seem like phenomena that cannot exist together can really exist together | | False Positive | What seems like positive covariation is really negative covariation | | False Negative | What seems like negative covariation is really positive covariation | | Header (General) | Incremental is continuous, continuous is incremental, curvilinear is linear, linear is curvilinear | | False Similarity | What seem like nearly similar phenomena are really opposite phenomena | | False Difference | What seem like different phenomena are really the same | | Dependent Variable is Independent Variable | What seems like the predictor is really the outcome | | Independent Variable is Dependent Variable | What seems like the outcome is really the predictor | | One-Way relationship is Complex | What seems like a direct relationship is really a mutual or non-recursive relationship | ## implementable it should have a tool to implement what the audience learn. - how to resolve "resistance to change" - if supplying the research for the audience with economics background, i should camouflage the language to what they can (and don't have uncomfortableness in) digesting. | Application of Research method (Bayesian Computation and Simulation | Bible (2020+) | Relevance to Angie's Problem | Seed paper for | Limitation | outside school but insightful | Current Frontier | interacting | honorable mention | | ---------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- | ------------------------------ | --------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------- | | Application1 <br>Modeling workflow<br> | Bayesian Workflow (statistics)<br><br>World to Word Model (cognition)<br><br>Analytical Methods for Dynamic Modelers (dynamics) | experiment choice (which data to collect)<br><br>rational meaning construction | 1️⃣Parameterization & Modeling<br>[[_ref/Gelman04_parameterizationBayes.pdf]] (Andrew Gelman's favorite)<br>[[📜gelman04]] | Holes in Bayesian Stats<br><br>data collection is not cast as resource allocation | Computational Rationality | posterior SBC (statistics)<br><br>ADEV (computer science) | Andrew (statistics)<br><br>Vikash (computer science)<br><br>Tom (dynamics) | [[📜gans23_choose(ent, exp)]] | | Application2<br>Operations and Innovation Management using Bayesian computation and simulation | Bayesian Entrepreneurship<br><br>Product development and opportunity tournament | <br><br><br>what to be uncertain about | 4️⃣empirical approach recipe [[📜MackeyBarneyDotson15_CorpDiv]] [[_ref/MackeyBarneyDotson15_CorporateDiv.pdf]] (Jay Barney's favorite)<br><br> | | 3️⃣Operations for entrepreneur | Need Analysis capturing psychological inventory (Moshe) | Moshe (choice analysis, demand modeling)<br><br>Scott (economics of idea, innovation, entrepreneurship)<br><br>Charlie (decision science, operations management) | | | Counterfactual - Operations and Innovation Management without Bayesian Computation | | | 2️⃣[[📜Meehl90_appraising_amend]] | 2️⃣One and Done (Efficiency) | | | | |