eval(charlie-scott, angie)1

2024-10-14 - evaluators: [[charlie24👓_nss🗣️]], [[charlie24🛠️_clockspeed🗣️]], [[🗄️🧠charlie]], [[scott24👓_Bayesian_Entrepreneurship.pdf]], [[scott23🛠️_econ_idea_innov_ent.pdf]], [[🗄️🧠scott]] - which are separated into [[🛝 Slide Deck eval(charlie-scott,angie)1]] [[📝 paper eval(charli-scott, angie)1]] --- plan for today - [[#0. reflect on last meeting|0. reflect on last meeting]] - [[#1. ⚙️Who, When, Why of Machine Partner in Innovative Decision-Making|1. ⚙️Who, When, Why of Machine Partner in Innovative Decision-Making]] - [[#1. ⚙️Who, When, Why of Machine Partner in Innovative Decision-Making#Who|Who]] - [[#1. ⚙️Who, When, Why of Machine Partner in Innovative Decision-Making#When|When]] - [[#1. ⚙️Who, When, Why of Machine Partner in Innovative Decision-Making#Why|Why]] --- plan for today - [[#2.🧠📍Probabilistic Reasoning and Selecting Optimal Action in NSS|2.🧠📍Probabilistic Reasoning and Selecting Optimal Action in NSS]] - [[#2.🧠📍Probabilistic Reasoning and Selecting Optimal Action in NSS#2.1🧠 Probabilistic Reasoning in NSS|2.1🧠 Probabilistic Reasoning in NSS]] - [[#2.🧠📍Probabilistic Reasoning and Selecting Optimal Action in NSS#2.2📍 Selecting Optimal Action in NSS|2.2📍 Selecting Optimal Action in NSS]] - [[#3. Weaving 👁️🧠📍🤝 for 🌳⛰️🌊 on Probabilistic Program|3. Weaving 👁️🧠📍🤝 for 🌳⛰️🌊 on Probabilistic Program]] --- ## 0. reflect on last meeting Reviewing the [recording](https://otter.ai/u/FsRSr6VSqA6QVGtYnEfT6p4Db-c?utm_source=copy_url), I'm glad scott liked the optimal sequencing topic. scott follows Howard Stevenson's definition of entrepreneurship "pursuit of opportunity beyond resources currently controlled". [Tom Eisenmann dissects three words of this definition](https://hbr.org/2013/01/what-is-entrepreneurship). He introduces four types of opportunities that are not mutually exclusive: 1) pioneering a truly innovative product; 2) devising a new business model; 3) creating a better or cheaper version of an existing product; or 4) targeting an existing product to new sets of customers. From operations and innovation and management perspective, each opportunity can be approached three value chain components: solution chain (supply, design, develop, launch), fulfillment chain (supply, produce, distribute, sell), demand chain (design, develop). Focus of 1) is solution, 2) is fulfillment, 3) is fulfillment, 4) is fulfillment and demand. Also, four risks are introduced. _Demand risk_ relates to prospective customers’ willingness to adopt the solution envisioned by the entrepreneur. _Technology risk_ is high when engineering or scientific breakthroughs are required to bring a solution to fruition. _Execution risk_ relates to the entrepreneur’s ability to attract employees and partners who can implement the venture’s plans. _Financing risk_ relates to whether external capital will be available on reasonable terms. Among these woven opportunities and risks, I segment fourth opportunity and financial, demand, execution risks. I have clear idea on motivation behind pursuing "targeting existing products to new sets of customers" opportunity. Moreover, analyzing this opportunity with operations management principles (matching supply with demand) and cognitive science principles (rational belief and action) can impact management practice. Within this opportunity type, technical risk is tamed. I designed three mutually exclusive and extensive contexts in innovation odyssey [^4] : 💰🌳managing financial risk with `capitalize` operations in nail stage , 💠⛰️managing demand risk with `segment-collaborate` operations in scale stage , 🛫 🌊 managing execution risk with `evaluate` operations in sail stage. Within these context, I argue how machine outperforms human for the task of "partnering innovative decision-making". --- ## 1. ⚙️Who, When, Why of Machine Partner in Innovative Decision-Making Table 1: Four Key Functions of Decision-Making | Function | Input | Output | Why Machine Partner is Better | | ------------------------------ | --------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | 👁️Choosing Relevant States | world model at time t $W_t$ | - state of agent model at time t $w^a_tlt;br>- state of environement model at time t $w^e_tlt;/font> | Machines can process diverse, high-dimensional data much faster and more accurately, capturing subtle patterns and relationships | | 🧠Probabilistic Reasoning | - $w^a_t, w^e_tlt;/font> - state of agent's belief at time t-1 $b_{t-1}$ | $b_tlt;/font> | Machines has better inference methods (Markov chain monte carlo, sequential monte carlo) on complex probability distributions, and has higher precision which prevents human biases like probability shifting | | 📍Selecting Optimal Action | - $b_tlt;/font> - utility | action at time t $a_tlt;/font> | Machines can explore vast action spaces and optimize long-term strategies more effectively, considering numerous possibilities | | 🤝Estimating Commitment Effect | - $a_tlt;/font> - $W_t$ | $W_{t+1}$ | Machines can consistently track and update complex state spaces, accurately modeling the long-term impacts of decisions | ![[theory2alg.png|400]] --- ### Who Entrepreneur/Innovator pursuing opportunities of targeting existing products to new sets of customers, who needs to manage financial, demand, execution risk. This has the same effect of anchoring technology among the four key entrepreneurial choices: customer, technology, competition, and organization[^1] --- ### When Machine partners demonstrate superiority in decision-making environments characterized by: 1. Complex Data Landscapes - High-dimensional data environments - Real-time data streams - Situations requiring rapid processing and analysis of vast amounts of information 2. Intricate Decision Spaces - Large action spaces with numerous possible choices - Complex systems with multiple interacting variables - Scenarios where decisions have far-reaching and interconnected consequences 3. Dynamic Temporal Contexts - Long-term planning scenarios requiring foresight and strategy - Environments with frequent state changes and evolving conditions - Situations where historical trends and future projections are crucial for decision-making These conditions often overlap in real-world business scenarios, particularly in rapidly evolving markets, complex supply chains, or when dealing with diverse customer bases. Machine partners excel in navigating these multifaceted environments, offering superior decision-making support across various stages of business growth. --- ### Why Machine partners excel in all four key functions of decision-making, that can be applied to every stage of organization willing to innovate, but for today we focus on the second and thrid function. 🧠 Probabilistic Reasoning: This is where machine partners are ⭐️⭐️⭐️ALWAYS! superior. Machines employ advanced inference methods like Markov Chain Monte Carlo (MCMC) and Sequential Monte Carlo (SMC) to handle complex probability distributions. Unlike humans, who tend to mentally shift probabilities towards 0, 1, or 50/50[^2] (e.g. when presented with 29% chance of an event occurring, many people interpret this as meaning the event simply won't happen), machines maintain high precision in probabilistic calculations. This prevents biases and allows for more accurate updating of beliefs based on new evidence. - Nail Stage: MCMC is used to update beliefs about future company valuation, incorporating new evidence through the `observe` function. This allows for efficient sampling from complex posterior distributions, crucial for accurate predictions in uncertain business environments. - Scale Stage: Updating beliefs about market adoption rates and production efficiency based on new sales data and operational metrics. - Sail Stage: Refining beliefs about potential efficiency gains from AV adoption using new performance data and operational observations. 📍 Selecting Optimal Action: Machines are ⭐️⭐️MOSTLY better at this function. They can explore vast action spaces and optimize long-term strategies more effectively, considering numerous possibilities. This is particularly valuable in scenarios with large action spaces and multi-stage decision processes. - Nail Stage: Determining optimal SAFE terms (investment amount and valuation cap) based on updated valuation beliefs and founder's utility function. - Scale Stage: Selecting the best combination of market-product strategy and sourcing decisions from a complex set of options. - Sail Stage: Deciding whether to adopt AV for baggage loading based on updated efficiency gain beliefs and the airport's utility function. The superiority of machine partners in these functions, particularly in probabilistic reasoning, is fundamental to entrepreneurial decision-making. By leveraging advanced inference methods, machines can handle the complexity and uncertainty inherent in business scenarios, from initial funding decisions to market expansion and technology adoption. This allows entrepreneurs to make more informed decisions based on accurately updated beliefs, considering a wider range of possibilities and long-term consequences than would be feasible through human reasoning alone. In this paper, we assume preference is revealed under SAME utility. If underlying utility changes (e.g. reboot-pivot from [[reid 5lev pivoting]]) we define a new agent. --- ## 2.🧠📍Probabilistic Reasoning and Selecting Optimal Action in NSS we'll focus on the two functions today and the other two in next meeting [[eval(charlie-scott, angie)2]]. | 🧠Probabilistic Reasoning | Input: Prior beliefs on valuation, new market data Output: Updated probability distribution of future valuations | Input: Prior beliefs on market adoption, new sales data Output: Updated probability distribution of market success for each segment-product combination | Input: Prior beliefs on operational efficiency, new AV performance data Output: Updated probability distribution of AV adoption success | | -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------- | | 📍Selecting Optimal Action | Input: Updated valuation beliefs, founder's utility function Output: Optimal SAFE terms (investment amount and valuation cap) | Input: Updated market success beliefs, company's utility function Output: Optimal market-product strategy and sourcing decisions | Input: Updated AV adoption beliefs, airport's utility function Output: Decision on whether to adopt AV for baggage loading | ### 2.1🧠 Probabilistic Reasoning in NSS #### 🌳 financial risk and capitalize in nailing ```javascript // 🧠🌳Probabilistic reasoning var probabilisticReasoning = function(priorBelief, newEvidence) { return Infer({method: 'MCMC', samples: 10000}, function() { var futureValuation = sample(priorBelief); observe(Gaussian({mu: futureValuation, sigma: 1000000}), newEvidence.observedValuation); return futureValuation; }); }; ``` In the Nail stage, MCMC is used to update beliefs about future company valuation. The `priorBelief` represents the initial distribution of possible future valuations. The `observe` function incorporates new evidence (observed valuation) using a Gaussian distribution. The syntax `observe(Gaussian({mu: futureValuation, sigma: 1000000}), newEvidence.observedValuation)` means we're observing a value (`newEvidence.observedValuation`) that we expect to be drawn from a Gaussian distribution centered around our current belief (`futureValuation`) with a standard deviation of 1,000,000. This effectively updates our belief about the future valuation based on the observed evidence. MCMC efficiently samples from this complex posterior distribution, allowing for accurate prediction of future valuations while accounting for uncertainty and new market information. Underlying mechanism of how it adds new observation's log probability of program execution to get the new probabilities is explained in [probabilistic modeling textbook](http://probmods.org/chapters/conditioning.html) where you can play with program and get visualized result of inference. #### ⛰️ demand risk and segment-collaborate in scaling ```javascript // 🧠⛰️Probabilistic reasoning var probabilisticReasoning = function(priorBelief, newEvidence) { return Infer({method: 'MCMC', samples: 10000}, function() { var marketSuccess = sample(priorBelief); observe(Gaussian({mu: marketSuccess, sigma: 0.1}), newEvidence.observedSuccess); return marketSuccess; }); }; ``` In the Scale stage, MCMC is employed to update beliefs about market success. The `priorBelief` here represents the initial distribution of possible market success rates. The `observe` function incorporates new evidence of observed success, again using a Gaussian distribution. Here, `observe(Gaussian({mu: marketSuccess, sigma: 0.1}), newEvidence.observedSuccess)` means we're observing a value (`newEvidence.observedSuccess`) that we expect to be drawn from a Gaussian distribution centered around our current belief about market success (`marketSuccess`) with a standard deviation of 0.1. This updates our belief about market success based on the observed evidence. MCMC allows for efficient sampling from this posterior distribution, enabling accurate predictions of market success while accounting for the complexity of market dynamics and new sales data. #### 🌊 execution risk and evaluate in sailing ```javascript // 🧠🌊Probabilistic reasoning var probabilisticReasoning = function(priorBelief, newEvidence) { return Infer({method: 'MCMC', samples: 10000}, function() { var efficiencyGain = sample(priorBelief); observe(Gaussian({mu: efficiencyGain, sigma: 0.1}), newEvidence.observedGain); return efficiencyGain; }); }; ``` In the Sail stage, MCMC is used to update beliefs about efficiency gains from adopting new technology (AV for baggage loading). The `priorBelief` represents the initial distribution of possible efficiency gains. The `observe` function incorporates new evidence of observed efficiency gains using a Gaussian distribution. The syntax `observe(Gaussian({mu: efficiencyGain, sigma: 0.1}), newEvidence.observedGain)` means we're observing a value (`newEvidence.observedGain`) that we expect to be drawn from a Gaussian distribution centered around our current belief about efficiency gain (`efficiencyGain`) with a standard deviation of 0.1. This updates our belief about the efficiency gain based on the observed evidence. MCMC enables efficient sampling from this posterior distribution, allowing for accurate predictions of efficiency gains while accounting for the complexities of technological adoption and operational data. In all three stages, MCMC proves invaluable for handling the complex, multidimensional probability distributions inherent in entrepreneurial decision-making. It allows for efficient updating of beliefs in light of new evidence, crucial for making informed decisions in rapidly changing business environments. The use of 10,000 samples in each case ensures a good balance between computational efficiency and accuracy of the posterior distribution estimates. The `observe` function plays a key role in these models by allowing us to incorporate observed evidence into our beliefs in a probabilistic manner. This is crucial for entrepreneurial decision-making, where we often need to update our beliefs and strategies based on new market information, sales data, or operational metrics. --- ### 2.2📍 Selecting Optimal Action in NSS Choices at Each Stage: - Nail: Founder's choice in Capitalize operations: investment terms with grow fast OR maintain control - Scale: Manufacturing startup's choice in Segment and Collaborate operations: ((urban OR rurual) & (300 range OR 400 range)) & (in-house OR outsource) & (local OR global manufacturing). - Sail: Airport's choice in Evaluate operations: short term quantity OR long term quality (worse then better dynamics) --- #### 🌳 financial risk and capitalize in nailing ```javascript // 📍🌳Select optimal action var selectOptimalAction = function(belief, utilityFunction) { return argMax(function(terms) { return expectation(Infer({method: 'forward', samples: 1000}, function() { var futureValuation = sample(belief); return utilityFunction(futureValuation, terms); })); }, safeTerms); }; ``` In the Nail stage, the optimal action selection focuses on determining the best SAFE (Simple Agreement for Future Equity) terms. The `selectOptimalAction` function takes the updated belief about future company valuation and a utility function as inputs. It uses the `argMax` function to find the SAFE terms that maximize expected utility. The function generates 1000 samples of possible future valuations based on the current belief. For each sample, it calculates the utility of different SAFE terms and selects the terms that provide the highest expected utility. This approach allows the entrepreneur to make an informed decision about fundraising that balances the amount of investment with potential equity dilution, considering the uncertainty in future company valuation. #### ⛰️ demand risk and segment-collaborate in scaling ```javascript // 📍⛰️Select optimal action var selectOptimalAction = function(marketBelief, sourcingBelief, utilityFunction) { return argMax(function(action) { return expectation(Infer({method: 'forward', samples: 1000}, function() { var marketSuccess = sample(marketBelief); var sourcingEfficiency = sample(sourcingBelief); return utilityFunction(marketSuccess, sourcingEfficiency, action); })); }, Infer({method: 'enumerate'}, function() { return {marketProduct: uniformDraw(marketProducts), sourcing: uniformDraw(sourcingOptions)}; })); }; ``` In the Scale stage, selecting the optimal action involves choosing the best combination of market-product strategy and sourcing decisions. The function takes beliefs about market success and sourcing efficiency, along with a utility function. It uses `argMax` to find the action that maximizes expected utility across all possible combinations of market-product and sourcing options. The function generates 1000 samples each of market success and sourcing efficiency based on current beliefs. For each sample pair, it calculates the utility of different action combinations. The `uniformDraw` function ensures that all possible market-product and sourcing options are considered equally. This approach enables the entrepreneur to make complex, multi-faceted decisions that balance market opportunity with operational efficiency. #### 🌊 execution risk and evaluate in sailing ```javascript // 📍🌊Select optimal action var selectOptimalAction = function(belief, utilityFunction) { return argMax(function(action) { return expectation(Infer({method: 'forward', samples: 1000}, function() { var efficiencyGain = sample(belief); return utilityFunction(efficiencyGain, action); })); }, actions); }; ``` In the Sail stage, the optimal action selection focuses on the decision to adopt or not adopt Autonomous Vehicles (AV) for baggage loading. The function takes the updated belief about potential efficiency gains and a utility function as inputs. It uses `argMax` to determine whether adopting AV will maximize expected utility. The function generates 1000 samples of possible efficiency gains based on the current belief. For each sample, it calculates the utility of adopting or not adopting AV. This approach allows the decision-maker to weigh the potential efficiency improvements against the costs and risks associated with adopting new technology. Across all three stages, the Selecting Optimal Action function plays a crucial role in translating updated beliefs into concrete decisions. By considering a range of possible outcomes and their associated utilities, it enables entrepreneurs and decision-makers to make choices that are robust to uncertainty and aligned with their overall objectives. The use of probabilistic programming techniques like `argMax` and `Infer` allows for sophisticated decision-making that goes beyond simple heuristics, taking full advantage of the machine partner's capability to explore complex decision spaces efficiently. --- ## 3. Weaving 👁️🧠📍🤝 for 🌳⛰️🌊 on Probabilistic Program ```javascript // 🌳Nail Stage: Deciding on SAFE terms (investment amount and valuation cap) // Define possible SAFE terms var safeTerms = [ {investment: 1000000, valuationCap: 5000000}, {investment: 500000, valuationCap: 7000000} ]; // 👁️Choose relevant states var chooseRelevantStates = function(marketData) { return { currentValuation: marketData.valuation, investorInterest: marketData.investorSentiment }; }; // 🧠Probabilistic reasoning var probabilisticReasoning = function(priorBelief, newEvidence) { return Infer({method: 'MCMC', samples: 10000}, function() { var futureValuation = sample(priorBelief); observe(Gaussian({mu: futureValuation, sigma: 1000000}), newEvidence.observedValuation); return futureValuation; }); }; // 📍Select optimal action var selectOptimalAction = function(belief, utilityFunction) { return argMax(function(terms) { return expectation(Infer({method: 'forward', samples: 1000}, function() { var futureValuation = sample(belief); return utilityFunction(futureValuation, terms); })); }, safeTerms); }; // 🤝Estimate commitment effect var estimateCommitmentEffect = function(chosenTerms, currentState) { return Infer({method: 'forward', samples: 1000}, function() { var futureValuation = sample(Gaussian({mu: currentState.currentValuation * 1.5, sigma: 1000000})); var equity = chosenTerms.investment / Math.min(futureValuation, chosenTerms.valuationCap); return {futureValuation: futureValuation, equity: equity}; }); }; // Main decision process var nailStageDecision = function(marketData) { var relevantStates = 👁️chooseRelevantStates(marketData); var priorBelief = Gaussian({mu: relevantStates.currentValuation, sigma: 2000000}); var updatedBelief = 🧠probabilisticReasoning(priorBelief, {observedValuation: relevantStates.currentValuation}); var utilityFunction = function(futureValuation, terms) { var equity = terms.investment / Math.min(futureValuation, terms.valuationCap); return terms.investment - (equity * 10000000); // Assuming $10M exit for simplicity }; var optimalTerms = 📍selectOptimalAction(updatedBelief, utilityFunction); var commitmentEffect = 🤝estimateCommitmentEffect(optimalTerms, relevantStates); return {optimalTerms: optimalTerms, commitmentEffect: commitmentEffect}; }; // Example usage var marketData = {valuation: 3000000, investorSentiment: 0.7}; var decision = nailStageDecision(marketData); print("Optimal SAFE terms:", decision.optimalTerms); print("Estimated future valuation:", expectation(decision.commitmentEffect, function(x) { return x.futureValuation; })); print("Estimated equity dilution:", expectation(decision.commitmentEffect, function(x) { return x.equity; })); ``` ```javascript // ⛰️Scale Stage: Deciding on market-product fit and sourcing strategy // Define possible actions var marketProducts = [{market: 'urban', range: 350}, {market: 'urban', range: 400}, {market: 'rural', range: 350}, {market: 'rural', range: 400}]; var sourcingOptions = [{type: 'inhouse', location: 'local'}, {type: 'inhouse', location: 'global'}, {type: 'outsource', location: 'local'}, {type: 'outsource', location: 'global'}]; // 👁️Choose relevant states var chooseRelevantStates = function(marketData, productionData) { return { marketDemand: marketData.demand, productionCapacity: productionData.capacity, productionCosts: productionData.costs }; }; // 🧠Probabilistic reasoning var probabilisticReasoning = function(priorBelief, newEvidence) { return Infer({method: 'MCMC', samples: 10000}, function() { var marketSuccess = sample(priorBelief); observe(Gaussian({mu: marketSuccess, sigma: 0.1}), newEvidence.observedSuccess); return marketSuccess; }); }; // 📍Select optimal action var selectOptimalAction = function(marketBelief, sourcingBelief, utilityFunction) { return argMax(function(action) { return expectation(Infer({method: 'forward', samples: 1000}, function() { var marketSuccess = sample(marketBelief); var sourcingEfficiency = sample(sourcingBelief); return utilityFunction(marketSuccess, sourcingEfficiency, action); })); }, Infer({method: 'enumerate'}, function() { return {marketProduct: uniformDraw(marketProducts), sourcing: uniformDraw(sourcingOptions)}; })); }; // 🤝Estimate commitment effect var estimateCommitmentEffect = function(chosenAction, currentState) { return Infer({method: 'forward', samples: 1000}, function() { var futureMarketShare = sample(Beta({a: currentState.marketDemand * 10, b: 10})); var futureProductionEfficiency = sample(Beta({a: currentState.productionCapacity * 10, b: 10})); return {marketShare: futureMarketShare, productionEfficiency: futureProductionEfficiency}; }); }; // Main decision process var scaleStageDecision = function(marketData, productionData) { var relevantStates = 👁️chooseRelevantStates(marketData, productionData); var marketPriorBelief = Beta({a: relevantStates.marketDemand * 10, b: 10}); var sourcingPriorBelief = Beta({a: relevantStates.productionCapacity * 10, b: 10}); var updatedMarketBelief = 🧠probabilisticReasoning(marketPriorBelief, {observedSuccess: relevantStates.marketDemand}); var updatedSourcingBelief = 🧠probabilisticReasoning(sourcingPriorBelief, {observedSuccess: relevantStates.productionCapacity}); var utilityFunction = function(marketSuccess, sourcingEfficiency, action) { var revenue = marketSuccess * (action.marketProduct.market === 'urban' ? 1.2 : 1) * (action.marketProduct.range === 400 ? 1.1 : 1); var cost = (1 - sourcingEfficiency) * (action.sourcing.type === 'inhouse' ? 1.2 : 1) * (action.sourcing.location === 'global' ? 0.8 : 1); return revenue - cost; }; var optimalAction = 📍selectOptimalAction(updatedMarketBelief, updatedSourcingBelief, utilityFunction); var commitmentEffect = 🤝estimateCommitmentEffect(optimalAction, relevantStates); return {optimalAction: optimalAction, commitmentEffect: commitmentEffect}; }; // Example usage var marketData = {demand: 0.7}; var productionData = {capacity: 0.6, costs: 0.5}; var decision = scaleStageDecision(marketData, productionData); print("Optimal market-product strategy:", decision.optimalAction.marketProduct); print("Optimal sourcing strategy:", decision.optimalAction.sourcing); print("Estimated future market share:", expectation(decision.commitmentEffect, function(x) { return x.marketShare; })); print("Estimated future production efficiency:", expectation(decision.commitmentEffect, function(x) { return x.productionEfficiency; })); ``` ```javascript // 🌊Sail Stage: Deciding on AV adoption for baggage loading // Define possible actions var actions = [{adopt: true}, {adopt: false}]; // 👁️Choose relevant states var chooseRelevantStates = function(operationsData, avData) { return { currentEfficiency: operationsData.efficiency, avReliability: avData.reliability, avCost: avData.cost }; }; // 🧠Probabilistic reasoning var probabilisticReasoning = function(priorBelief, newEvidence) { return Infer({method: 'MCMC', samples: 10000}, function() { var efficiencyGain = sample(priorBelief); observe(Gaussian({mu: efficiencyGain, sigma: 0.1}), newEvidence.observedGain); return efficiencyGain; }); }; // 📍Select optimal action var selectOptimalAction = function(belief, utilityFunction) { return argMax(function(action) { return expectation(Infer({method: 'forward', samples: 1000}, function() { var efficiencyGain = sample(belief); return utilityFunction(efficiencyGain, action); })); }, actions); }; // 🤝Estimate commitment effect var estimateCommitmentEffect = function(chosenAction, currentState) { return Infer({method: 'forward', samples: 1000}, function() { var futureEfficiency = chosenAction.adopt ? sample(Beta({a: currentState.currentEfficiency * 10 + 2, b: 12 - currentState.currentEfficiency * 10})) : sample(Beta({a: currentState.currentEfficiency * 10, b: 10 - currentState.currentEfficiency * 10})); var stakeholderSatisfaction = chosenAction.adopt ? sample(Beta({a: 8, b: 2})) : sample(Beta({a: 5, b: 5})); return {efficiency: futureEfficiency, satisfaction: stakeholderSatisfaction}; }); }; // Main decision process var sailStageDecision = function(operationsData, avData) { var relevantStates = 👁️chooseRelevantStates(operationsData, avData); var priorBelief = Gaussian({mu: 0.2, sigma: 0.1}); var updatedBelief = 🧠probabilisticReasoning(priorBelief, {observedGain: avData.observedEfficiencyGain}); var utilityFunction = function(efficiencyGain, action) { if (action.adopt) { return (relevantStates.currentEfficiency + efficiencyGain) * 1000000 - relevantStates.avCost; } else { return relevantStates.currentEfficiency * 1000000; } }; var optimalAction = 📍selectOptimalAction(updatedBelief, utilityFunction); var commitmentEffect = 🤝estimateCommitmentEffect(optimalAction, relevantStates); return {optimalAction: optimalAction, commitmentEffect: commitmentEffect}; }; // Example usage var operationsData = {efficiency: 0.7}; var avData = {reliability: 0.9, cost: 500000, observedEfficiencyGain: 0.15}; var decision = sailStageDecision(operationsData, avData); print("Optimal decision:", decision.optimalAction); print("Estimated future efficiency:", expectation(decision.commitmentEffect, function(x) { return x.efficiency; })); print("Estimated stakeholder satisfaction:", expectation(decision.commitmentEffect, function(x) { return x.satisfaction; })); ``` ## Refernces [^1]: entrepreneurship choice and strategy (2025) gans, scott, stern [^2]: [On the edge: the art of risking everything (2024) Silver](https://www.penguinrandomhouse.com/books/529280/on-the-edge-by-nate-silver/) [^3]: Cachon, G., & Terwiesch, C. (2008). _Matching supply with demand_ (Vol. 20012). New York: McGraw-Hill Publishing. [^4]: https://nailitscaleitsailit.com/ using [Optimizing Entrepreneurial Decision-Making with AI cld](https://claude.ai/chat/cf2aabaa-aef2-4a75-adb1-ccb6394441a5) --- Angie's review: - when scott asked - do you mean like a function in math? what i could have said is function with input and output, but i do think function in business context can be interpreted as that - which is all business dynamics is about - representing lots of equations behind to explain behavior then make predictions