![[session5 - frontiers of bayesian decision making 2025-04-13-8_0.svg]] %%[[session5 - frontiers of bayesian decision making 2025-04-13-8_0|🖋 Edit in Excalidraw]]%% ent persuing idea, engage in some level idea, experiment, strategy, ent choose the opportunity they persue impact - pursue my view on their perception is correct - group of experienced investor about the idea - group of experts disagree with each other - demand and supply curve -> everyone agree on same answer _**Shark Tank**_[[b]](https://en.wikipedia.org/wiki/Shark_Tank#cite_note-3) is an American business [reality television series](https://en.wikipedia.org/wiki/Reality_television#Investments "Reality television") that premiered on August 9, 2009, on [ABC](https://en.wikipedia.org/wiki/American_Broadcasting_Company "American Broadcasting Company").[[2]](https://en.wikipedia.org/wiki/Shark_Tank#cite_note-Insider_TV-4) The show is the American franchise of the international format _[Dragons' Den](https://en.wikipedia.org/wiki/Dragons%27_Den "Dragons' Den")_, a British TV series, which itself is a remake of the Japanese TV show The Tigers of Money.[[3]](https://en.wikipedia.org/wiki/Shark_Tank#cite_note-5) It shows [entrepreneurs](https://en.wikipedia.org/wiki/Entrepreneurship "Entrepreneurship") making business presentations to a panel of five [angel investors](https://en.wikipedia.org/wiki/Angel_investors "Angel investors") (providers of [venture capital](https://en.wikipedia.org/wiki/Venture_capital "Venture capital") to early stage [start-ups](https://en.wikipedia.org/wiki/Start-up "Start-up")) called "Sharks" on the program, who decide whether to invest in their companies. purposeful exploration breaking core econ assumption (subjective priors) - person to be believer that update belief -> proactively update human beings are walking around () perceptions of the same data - new situations are not overconfidence can be rational - as long as there's those other ppl said no (s) ⭐️demand for experimentation - implementation vs giving up no biased learning - signal type1, type2 error - lucky to have ; proportion btw type1,2 is built in - currently optimistic eliminate false negative; pessimistic eliminate false positive crossing the chasm ; incumbants hate ⭐️planning depth - expertise develop; entrepreneurs don't start as experts (novice to expert) 1. how dow e decde whether worth play 2. decide what action to take 3. create new games expertise literature has its role - abstracting away (how the value of expertise; ) 🙋‍♀️scott's point: penalty shot on soccer and tennis (marginal return; reward sholud be equalized; game theoretic expertise ) adaptation decision in new game - early stage NOT FUN CUZ who goes first win building computation model for cognitive - how do we get started playing new games, novel game experiment, 121 new games opponents = reasonable player 1. how likely is the game to end in a draw 2. if the game does not end in a draw, how likely is it that the first player will win? 3. how fun would this game be to play? how far ahead do we search? how do we choose moves to search over? depth limited search with general game feature (you vs opponent progress) cognitive constraints (having a bias or belief - ) computationally bounded way (active bias or constraint) theory of computation curves shaped by what's the right thing (payoff) ability of the model and human's to predict relative to soh ; true distribution is more flat calibrating to the tradeoff - what are the circumstance under which does the machine know that vs humans know that - three objects (every outcome as triplet) 🙋‍♀️depend on expected outcome experts might do entropy = (variation in distribution; winning draws losing) - fun = challenge of overcoming - how likely you're to win against random agent, no advantage over random agent - theory of i know sth but what you don't have (information; cognitive computational) micro processor - homebrew (same computer kits, intel; take things interface, wazniak vs jobs - new ) tolerance to risk, ENT don't think it's risky knowledge, theories, building computation models (individual - scientific theories) - tested intuitive theory building all come to similar theory - (멘델 유전설; finches - 진화설) game - select into what game you play - endow (spot you powered) some games you gie advanatge while the other don't (private information; fundametnal info) flagship - proactive about nailing about (what do ppl believe - hard tech - ) tell ppl (advisory board - constraint doesn't exist; implication of - reasonably good - use that to bioengineering around those problem) relax if you give million dollar how to ai (how smart they are) what if you relax the axiom - iterate to intuitive theory (dimension), approximately bayesian (converge on the same thing - similar data converge to different things) hierarchical model - interpret (data is biased) - resource rational bayesian process (structured interesting hypothesis, ) bayesian inference for intuitive theory COMPUTATION REALISTIC VIEW OF COGNITION - interesting parts of thought don't seem like computation accout of mind --- intention of inviting [[tan_zhixuan]] is to give talk for [[session7_social planner]] on group level prior 2025-04-03 | Paper | Katie's Curation | | ----------------------------------------- | ---------------- | | [[📜zhang24_intuitive_game]] | | | [[📜zhu24_capturing_complexity_game]] | | | [[📜allen24_using_games_understand_mind]] | | # Model Answers from an EV Startup Operations Expert ## Questions from Zhang et al. - Intuitive Game Theory ### Q1: Your research shows humans evaluate novel situations using fast simulation with limited lookahead. How might this explain how EV startup founders like Elon Musk assessed the luxury EV market with minimal prior data before committing significant resources? **Expert Answer**: What we see in successful EV startups mirrors your research findings perfectly. When Tesla approached the luxury EV market, they couldn't afford comprehensive market studies across all segments. Instead, they used resource-rational decision-making with limited sampling – testing key hypotheses about performance and sustainability preferences among luxury buyers. Musk's approach was brilliant in this context – he used what I call "strategic sampling" by testing critical assumptions (like acceleration and design appeal) that had asymmetric information value. Rather than running 50 different market tests, he focused on 5-10 high-leverage experiments that would validate or invalidate their mental model about luxury performance buyers. This aligns perfectly with your finding that humans make remarkably good decisions with k=20 samples rather than exhaustive search – Tesla's early prototype feedback loops functioned exactly this way. ### Q2: You found that participants make reliable judgments about game fairness with minimal experience. How might this connect to Tesla's early decision to begin with high-end vehicles despite limited market validation for mass-market EVs? **Expert Answer**: Tesla's premium-first strategy demonstrates exactly what your research uncovered – entrepreneurs can make remarkably accurate judgments about complex market dynamics with minimal direct experience. Tesla effectively ran a "fairness" assessment of different market entry strategies. By sampling the luxury segment first, they could evaluate both the technological and market "rules" with lower stakes before attempting mass-market entry. Their mental simulation correctly identified that starting with the Roadster would yield crucial learning while generating revenue and brand equity. This approach validated their "game theory" that high-performance vehicles would create sufficient margin to fund technology development for mass-market vehicles. What's fascinating is that Tesla's strategy wasn't based on comprehensive market studies – it came from rapid simulation of market dynamics with limited information, just as your research shows humans evaluate novel games after minimal exposure. ### Q3: Your model with simple heuristics and limited samples (k=20) outperformed more complex models. What implications might this have for how EV startups should balance intuitive market assessment versus exhaustive market research? **Expert Answer**: Your findings explain why we see successful EV entrepreneurs using simplified decision models rather than exhaustive analysis. In the EV market's early days, complex market models were actually counterproductive because they created false precision about fundamentally uncertain parameters. Smart EV founders employ what I call "heuristic-driven experimentation" – using simple decision rules with limited sampling to drive rapid learning. For example, rather than commissioning a $500K market study on 50 vehicle attributes, an effective founder might identify 3-5 key parameters (range, performance, design) and run quick experiments to validate them. This approach creates what I call "iterative precision" – starting with rough estimates that become more refined through targeted learning. Your research explains why this works: when facing novel, complex situations, simple models with limited samples often outperform more sophisticated approaches because they avoid overfitting to noise in limited data. ## Questions from Zhu et al. - Capturing Complexity in Strategic Decision-Making ### Q1: Your research identified factors determining strategic complexity, including Nash equilibrium payoff dominance. How might this explain Tesla's decision to develop proprietary battery technology rather than adopting the industry-standard approach Better Place was pursuing? **Expert Answer**: The Nash equilibrium payoff dominance framework perfectly explains Tesla's battery strategy versus Better Place's approach. Tesla identified a strategy with clear payoff dominance across multiple dimensions: vertical integration of battery development yielded compounding advantages in performance, cost, and IP protection. In contrast, Better Place pursued a battery-swapping model with unclear payoff dominance – their strategy created complex interdependencies between vehicle design, infrastructure deployment, and consumer behavior. While potentially superior in a perfect coordination scenario, the equilibrium was significantly harder to achieve. This exemplifies what our operations team calls "convergent versus divergent strategy spaces." Tesla's approach converged toward a stable equilibrium where technological improvements reinforced their competitive position. Better Place faced a divergent strategy space where successful execution required simultaneous coordination across multiple stakeholders – a classic example of higher strategic complexity due to unclear payoff dominance. ### Q2: You found that context-dependent decision parameters work better than fixed ones. How might this inform how EV startups should adjust their decision criteria when transitioning from premium markets (like Tesla's Roadster) to mass-market vehicles (like Model 3)? **Expert Answer**: Your finding about context-dependent parameters is precisely what we observe in successful EV scaling strategies. Tesla's transition from Roadster to Model S to Model 3 demonstrated brilliant application of this principle. In the premium phase, Tesla weighted performance and brand parameters heavily, accepting high production costs and limited scalability. But when transitioning to mass market, they dynamically reweighted their decision parameters to emphasize manufacturability, cost structure, and charging infrastructure. This matches what we call "phase-appropriate decision frameworks" in operations. The decision parameters that drive success in early phases (emphasizing differentiation and proof-of-concept) must evolve in later phases (emphasizing scalability and cost efficiency). Your research explains why many EV startups fail during transition points – they apply fixed decision parameters across fundamentally different strategic contexts, resulting in poor decisions. The most successful companies actively recalibrate their decision weights as market complexity changes. ### Q3: Your game complexity index predicted response times and cognitive uncertainty. Could a similar framework help EV entrepreneurs identify which strategic decisions require more deliberation? **Expert Answer**: Absolutely – a complexity index would be transformative for EV startup decision allocation. We already see intuitive versions of this in successful companies, but formalizing it would be tremendously valuable. Currently, EV startups often misallocate cognitive resources – spending excessive time on simple decisions while rushing complex ones. For example, many founders obsess over logo design (low complexity) while making snap judgments about battery chemistry (high complexity). A formal complexity framework similar to yours would help founders quantify which decisions deserve deeper deliberation based on: 1. Equilibrium stability (decisions with multiple viable options) 2. Outcome variance (decisions with highly dispersed potential outcomes) 3. Iterative dependencies (decisions requiring multi-level strategic thinking) This would create what I call "cognitive resource optimization" – applying appropriate deliberation to each decision based on its complexity. Your research provides the theoretical foundation for building such frameworks. ## Questions from Allen et al. - Using Games to Understand the Mind ### Q1: Your paper argues that games allow study of complex behavior in ecological settings. How might EV startups design resource-efficient market experiments to test consumer reactions to novel features before full-scale production? **Expert Answer**: Your research on games as ecological settings provides the perfect template for what I call "compressed market simulations" in EV development. Instead of fully developing features before testing, forward-thinking EV startups create game-like experiments that compress the essence of new experiences into testable interfaces. For example, rather than building a complete self-parking system, Tesla might create a simple simulation that tests user trust and interaction with autonomous features. The key insight from your work is that well-designed "games" can elicit authentic responses that predict real-world behavior. In our EV work, we've found that simplified but authentic interactions yield remarkably accurate predictions of market reception. For instance, testing user interfaces with wireframes often reveals 80% of usability issues at 5% of the development cost. Your research explains why these approaches work – they tap into the same cognitive systems that would engage with the full product, while dramatically reducing development resources. ### Q2: You discussed how games enable research on intrinsic motivation. How might this inform Tesla's approach to creating compelling EV experiences that motivated early adoption despite range limitations and charging infrastructure challenges? **Expert Answer**: Tesla's brilliance lies precisely in applying the principles of intrinsic motivation you've studied in games. They recognized that early EV adoption couldn't rely solely on extrinsic motivators (like saving money on gas) given the practical limitations. Instead, Tesla created what we call "multi-layered engagement loops" that tapped into intrinsic motivations: - Mastery motivation through acceleration performance and handling - Exploration motivation through Easter eggs and software updates - Social identity motivation through environmental pioneering status This exactly parallels your research on why games sustain engagement – they create intrinsically rewarding experiences that outweigh friction points. Tesla didn't just sell transportation; they created an intrinsically motivating experience ecosystem that made early adopters willing to overlook range limitations. Your work explains why this approach succeeded where competitors failed – Tesla understood that adoption of radically new technologies requires tapping into deeper motivational systems, not just offering rational benefits. ### Q3: You outlined benefits and pitfalls of game-based research. What analogous benefits and pitfalls might EV entrepreneurs face when using limited market tests to inform large-scale strategic decisions? **Expert Answer**: The parallels between your game-based research challenges and EV market testing are striking. EV startups face precisely the experimental control versus ecological validity tradeoff you describe. The primary benefit of limited market tests (like Tesla's Roadster) is revealing authentic user behavior in natural contexts. We discovered that controlled lab tests of EV interfaces completely missed critical adoption factors that only emerged in real-world use. This mirrors your finding that game environments uncover behavioral patterns missed by artificial laboratory tasks. However, EV entrepreneurs face the same pitfalls you identify: selection bias (early adopters don't represent mainstream users), experimental control limitations (difficult to isolate variables), and generalizability challenges (behavior in one market segment doesn't predict others). Successful EV companies address these through what I call "triangulation testing" – using multiple test methodologies with complementary strengths and weaknesses, just as your research recommends combining game-based and controlled experimental approaches to overcome limitations of either method alone. using [crafting personalized mentor outreach cld](https://claude.ai/chat/14f75f71-3510-4664-8515-f49637f18a3c) given [[Angie's proposal for Apr7 session.pdf]], claude's strategic approach to connect with each mentor by highlighting the relevance of intuitive game theory to their research interests: |Aspect|Hazhir Rahmandad|Abdullah Almatouq|Jinhua Zhao| |---|---|---|---| |**Research Vision**|Integrating rigorous estimation methods into system dynamics; modeling organizational learning and capability development under uncertainty|Knowledge production frameworks in academic/research contexts; program theory and temporal validity in meta-science|M3S initiative applying AI to practical problems; creating inclusive, resilient technology solutions with global impact through SMART program| |**Connection to Intuitive Game Theory**|Frame intuitive game theory as a way to enhance system dynamics models with realistic decision-making processes; show how limited lookahead and heuristics in Katie's presentation mirror organizational learning under delay that Hazhir studies|Position the event as a concrete example of knowledge production across disciplines; show how the "cognitive science of pivots" relates to his interest in theory development and dissemination|Highlight how intuitive game theory models human adaptation to complex AI systems; connect to M3S goals of addressing technology design and human skill development in AI ecosystems| |**Understanding Mind**|Subject line: "System Dynamics Meets Theory of Mind: Invitation to MIT Bayesian Decision-Making Event"<br><br>Emphasize: How Josh's Bayesian models could enhance parameter estimation for human behavior in system dynamics; potential for collaboration on capturing realistic decision-making in organizational systems|Subject line: "Knowledge Integration in Action: MIT's Bayesian Decision-Making Frontier"<br><br>Emphasize: How the event demonstrates effective knowledge routing between cognitive science and entrepreneurship; practical applications of meta-science theory|Subject line: "AI Innovation Ecosystems & Human Decision Models: MIT Frontiers Event"<br><br>Emphasize: How the event explores human adaptation to technology through Bayesian cognitive models; relevance to creating resilient AI systems that account for human decision processes| **For Hazhir:** - Mention specifically how Katie's presentation on 1-step lookahead and limited sampling relates to his work on organizational learning under delay - Connect Josh's inverse planning models to Hazhir's interest in capability development tradeoffs - Emphasize the opportunity to bring system dynamics' rigor to entrepreneurial decision-making **For Abdullah:** - Focus on how the event demonstrates knowledge production frameworks in action - Highlight how this bridge between cognitive science and entrepreneurship exemplifies effective knowledge transfer - Position the May 5th session on "misaligned stakeholder beliefs" as directly relevant to his interest in improving academic knowledge dissemination **For Jinhua:** - Emphasize the AI applications throughout the three-session series - Connect intuitive game theory to creating more adaptive mobility systems that account for human decision processes - Position Bayesian models as tools for designing more inclusive technological innovations aligned with M3S goals josh said he can't prepare the talk and recommended to reach out to his collaborator (katie collins) to give a talk. he seems to be ok with leading the session (let's go with proposing certain format for gentle acceptance). angie: based on the spirit of "frontier" and information angie has been sharing since last fall on bayesian entrepreneurship movement, josh and angie have converged to the following three topics: - conceptual development (child as hacker) as hierarchical program induction () - intuitive game theory (katie collins 30min talk - can ask for longer) - agent’s reasoning about reasoning [[📜zhang24_intuitive_game]] papers on Theory of Mind approaches in AI/ML, along with Xuan's curation notes for each paper.