2025-04-04 by [Özgecan Koçak](https://scholar.google.com/citations?user=Spn57BcAAAAJ&hl=en) [[ozge_kocak]] ## 🧠🗣️transcripts # SD seminar Coordination and Compliance Analysis ## Transcript [https://otter.ai/u/BjBZUiq6tB0KC7umgDur3IsAK6M?view=summary](https://otter.ai/u/BjBZUiq6tB0KC7umgDur3IsAK6M?view=summary) The discussion centered on the impact of coordination (C) and interdependent payoffs (B) on decision-making accuracy and performance. When C = 0, individual actions are independent, but as C increases, coordination becomes crucial. The model shows that high C leads to inaccurate beliefs and reduced performance, even with increased exploration (tau). Rotating dictatorships achieve the highest accuracy and performance, while plurality voting is least accurate. Compliance with collective choices is modeled, with frequency-based compliance improving accuracy. The conversation also touched on the challenges of balancing exploration and exploitation in interdependent scenarios. ## Action Items - [ ] Explore incorporating the importance of ordering of payoffs, rather than just the exact values, into the analysis. - [ ] Consider how to model situations where there are lags and different timelines between decisions and payoff information. - [ ] Investigate the impact of incorporating resource rationality and the decoupling of beliefs and actions into the model. ## Outline ### Coordination and Value in Decision-Making - Speaker 1 explains the concept of coordination (C) and its impact on payoffs, emphasizing that coordination matters when everyone chooses the same action. - The parameter C is introduced to represent the need for coordination, with C = 0 indicating parallel search and C > 0 indicating division of labor or collaborative decisions. - Speaker 1 discusses the importance of coordination in real-life scenarios, such as division of labor, candidate evaluations, and collaborative research. - The variable B is introduced to represent the intrinsic value of actions, with B = 1 indicating equal payoffs for all actions. ### Impact of Coordination on Payoffs - Speaker 2 questions the confounding effect of C on the penalty of not coordinating or having a large capital N. - Speaker 1 clarifies that C represents the penalty of not coordinating, using the example of traffic congestion affecting travel time. - Speaker 3 raises the issue of small values of C and their impact on payoffs, noting that high payoff options are more likely to be sampled. - Speaker 5 discusses the Keynes description of the stock market and the beauty contest, highlighting the challenge of learning intrinsic value while coordinating actions. ### Learning and Coordination in Decision-Making - Speaker 1 explains that models do not assume expectations about others' actions, focusing on individual reinforcement learning. - Speaker 5 provides an example of vaccination decisions, where individual payoffs depend on collective actions. - Speaker 3 discusses the tension between assuming agents are blind to others' actions and providing information about payoff interdependence. - Speaker 1 presents performance data showing the shift in the individual frontier due to interdependence, leading to a conditional accuracy penalty. ### Exploration and Accuracy in Coordinated Actions - Speaker 1 explains that increased exploration does not improve accuracy in interdependent payoff scenarios. - Speaker 3 discusses the trade-off between steady-state performance and accuracy as C increases. - Speaker 1 presents a simulation with seven agents and seven alternatives, showing the impact of noise on payoffs. - Speaker 3 highlights the lack of an efficient frontier in high C regimes, where coordination effects dominate. ### Inaccuracy and Interdependent Payoffs - Speaker 1 introduces the concept of interdependent bias, explaining that inaccuracy increases as C > 0. - Speaker 5 discusses the example of Bitcoin and other cryptocurrencies, where individual exploration is challenging due to interdependent payoffs. - Speaker 1 explains that inaccurate beliefs cannot be corrected by individual exploration in interdependent payoff scenarios. - Speaker 3 suggests incorporating the importance of ordering payoffs in accuracy measurements. ### Collective Decision-Making and Compliance - Speaker 1 discusses the role of collective decision-making in achieving accuracy and performance, using the example of rotating dictatorships. - Speaker 1 presents data showing that rotating dictatorships achieve the highest accuracy and performance. - Speaker 3 questions the effectiveness of plurality voting and autocracy in achieving accuracy and performance. - Speaker 1 explains the trade-off between exploration and exploitation in different decision-making regimes. ### Frequency-Based Compliance and Accuracy - Speaker 1 introduces the concept of frequency-based compliance, where agents' compliance depends on the support for collective choices. - Speaker 3 discusses the importance of tracking others' actions and updating beliefs accordingly. - Speaker 1 presents data showing that morality voting achieves the highest accuracy in frequency-based compliance scenarios. - Speaker 3 highlights the trade-off between accuracy and performance in different compliance regimes. ### Balancing Exploration and Exploitation - Speaker 1 discusses the exploration-exploitation dilemma, where higher exploration leads to lower performance. - Speaker 3 suggests using upper confidence bound methods to balance exploration and exploitation more effectively. - Speaker 1 acknowledges the need to simplify the model by fixing tau and considering different learning styles. - Speaker 3 emphasizes the importance of understanding the underlying assumptions and their impact on decision-making. ### Separating Beliefs and Actions - Speaker 1 discusses the challenge of separating beliefs from actions, noting that agents' beliefs are private and actions are simultaneous. - Speaker 3 suggests defining prior distributions separately from beliefs to better understand decision-making. - Speaker 1 acknowledges the need to consider how beliefs and actions are decoupled in interdependent payoff scenarios. - Speaker 3 highlights the importance of understanding the impact of different learning styles on decision-making accuracy and performance. - # 1:1 Meeting with Özgecan Koçak ## Transcript [https://otter.ai/u/wJXTum57R0kCchHpGyRcisl1_-Q?view=summary](https://otter.ai/u/wJXTum57R0kCchHpGyRcisl1_-Q?view=summary) Angie Moon and Özgecan Koçak discussed the interdependence between founders and investors, emphasizing the challenges founders face in securing investments. Angie highlighted her research on probabilistic programming tools for entrepreneurs, noting the optimism founders often possess compared to investors. They explored the concept of "C," which represents the impact of others' actions on one's own payoff, and how it influences decision-making. Özgecan explained his model of interdependent decision-making, where individual exploration is challenging due to the uncertainty of others' actions. They also discussed the relevance of bandit models in understanding these dynamics. ## Action Items - [ ] Angie to send the slides from the seminar she referenced to the speaker. ## Outline ### Introduction and Initial Conversations - Angie Moon and Speaker 2 discuss their backgrounds, with Angie mentioning her interest in Turkey and Speaker 2 sharing his curiosity about South Korea. - Angie and Speaker 2 exchange gifts, with Angie offering a Korean item and Speaker 2 offering sour cherries. - Angie expresses her interest in Bayesian entrepreneurship and the concept of contrarian behavior in entrepreneurs like Elon Musk and Peter Thiel. - Speaker 2 explains that contrarian behavior is not necessarily more exploratory but involves different beliefs and willingness to try unconventional ideas. ### Entrepreneurial Risk Appetite and Optimism - Angie Moon discusses her PhD research on probabilistic programming tools and the misconception that entrepreneurship equals high risk appetite. - Angie explains her optimistic view on the viability of her PhD goal, despite differing opinions with her advisors. - Speaker 2 clarifies that optimism in this context means confidence in one's own belief despite others' skepticism. - Angie elaborates on the importance of probabilistic reasoning in entrepreneurial activities, such as capitalization and social interactions. ### Founder-Investor Collaboration and Optimism - Angie Moon discusses the interdependence between founders and investors and the challenges of optimism in investment decisions. - Speaker 2 suggests that founders need to accurately estimate which investors are likely to rank them highly. - Angie explains the cost of reaching out to different VCs and the importance of providing information to change investors' beliefs. - Speaker 2 emphasizes the importance of understanding investors' previous investments and how similar or complementary the founder's business is to those investments. ### Challenges in Investment Decisions - Angie Moon discusses the difficulty of reconstructing investors' ranking systems and the lack of data on what they didn't invest in. - Speaker 2 explains that founders can infer investors' preferences based on their previous investments. - Angie mentions the need for automation and segmentation in capitalization to make bottleneck-breaking decisions. - Speaker 2 introduces the concept of interdependence bias and the challenges of learning from others' actions in investment decisions. ### Exploration and Interdependence in Investment - Angie Moon and Speaker 2 discuss the concept of exploration and interdependence in investment decisions. - Speaker 2 explains that individual exploration does not help learning when others' actions are unknown. - Angie provides examples of how founders can learn from others' actions, such as traffic congestion on different routes. - Speaker 2 clarifies that the payoff from actions depends on others' actions, making it difficult to learn from individual choices. ### Vicarious Learning and Revealed Preference - Angie Moon and Speaker 2 discuss vicarious learning and the difference between revealed preference and stated preference. - Speaker 2 explains that learning from others' actions can be vicarious, either from their beliefs or observed actions. - Angie mentions a demand model in class that combines revealed preference and stated preference. - Speaker 2 emphasizes the importance of understanding others' actions to learn and make better decisions. ### Interdependent Decision-Making and C as Decision Variable - Angie Moon and Speaker 2 discuss the concept of C as a decision variable in interdependent decision-making. - Speaker 2 explains that C varies for different domains of choice, such as vaccination, masking, and investment decisions. - Angie asks if any models frame C as a decision variable for founders in the electric vehicle domain. - Speaker 2 clarifies that C is given by the environment, and founders choose their actions based on the payoffs. ### Bandit Models and Learning from Actions - Angie Moon inquires about the relevance of bandit models in the discussion of interdependent decision-making. - Speaker 2 explains that their models are similar to bandit models, with deterministic payoffs from different actions. - Angie asks about the number of episodes in their models, and Speaker 2 mentions having around 500 periods. - Angie finds the framework useful and asks to share it with others for further discussion. ### Final Remarks and Knowledge Sharing - Angie Moon thanks Speaker 2 for the insightful discussion and offers to share the framework with others. - Speaker 2 agrees to share the slides and further discuss the topic. - Angie expresses her appreciation for the opportunity to learn and share knowledge. - The meeting concludes with mutual thanks and well-wishes. ## 📝 angie's note only ordering matters log(mu_k^eq/v_k) = interdepence bias - exploration corection + log(S) rank order of (indepence) - not changing the rank order but only absolute value options of lower v value (graviat twd high) - inaccuracy same absolute value of inaccuracy - synthetic population? - pi_ik^t = v_ik (1-c(1-F_kt)) - hyper coordinated exploration (Every should ) action to belief (collectively chosen action - individual frontiers; c=0) john's Q: if every body choose the same decision (not the same as everyone doing - herd immunity - ); intrinsic value?? actual value being fixed -> depend on coordination - connectivity (socially producing ; by not adopting) compliance being low world everyone getting game stock (ngeative feedabck llop kick in) - "not to disagree with you too much" - adverse reaction no cost of taking any action learn by practicing - (how should coollective ) - explatory as a group hazhir's question on every prior is calibrated (have information is not the same as knowing - having prior) ⭐️giving voice to contrarians⭐️ = high prediction get the most votes - out of seven individuals (four choose option a, ) pluarity vs autocracy (dictator stays the same) vs with full compliance, centralized decsion making achieves highest accuracy at any given tau group is making choice (how many is thinking for suboptimal - highest in autocracy and loto but low in pluarirty) collective choice at time t is differen from collective choice at time t -1 these group doesn't expllore too much - lottocracy: Democracy Without Elections (rotating dictatorship - random; not picking most atypical action) group is making choice (how many is thinking for suboptimal - highest in autocracy and loto but low in pluarirty) ## hyper coordinate - what if executing is not perfect (if agents do not comply with collective choice) conformancy and compliance (what )