behavior and computational thinking; - every organization (ac) exists to change someone's behavior (QQ. what's the difference between organization and someone - active+producer (policy makers and tool builder) vs passive+consumer (modeler and public)) - business decision -> behavior science P2: data, algorithm, computing power -> representation (construct mental model), prediction (), explanation (travel, environmental, consumption behavior), control (broader - actuating, intervention), creation (ai impact notion of creativity) - one example of predict and control - if we can predict vs if we cannot predict - if today we can predict one person's travel behavior tomorrow perfectly, how may we offer service differently - desirability (individual), feasibility (community) order than deliver vs deliver than order - amazon last mile - can we predict mobility in individual level? (travel time, origin, destination); do you want to go? if so what time? - ![[Pasted image 20230911095420.png|200]] subway and railway (600), when get in, get out, activity ranked by the duration of activity (prediction is on community level) telecommunication and physical movement) - policy plays a bigger role in transpo. to - origin and destination is predicted independently - prediction accuracy spatial side is more granualar than time side (more regularity in space side than temporal side) - ![[Pasted image 20230911100842.png]] - • P2A: at the start of a day, predict when and where the first trip will be, given that the user will make a trip within that day (i.e., result of P1A); P2B: based on the last trip observed, predict when and where the next trip will be, given that the user will make another trip within that day (i.e., result of P1B). - what does (TFL: public vs private): travel information, marketing, travel demand management, pricing, journey planning definition of efficiency and equity is different (computation) - importance of agile communication between private and public sector as it's forms a loop. frontier evolves based on the action of participating agents: eg of loop is gpt bans using lobby data, monopoly providing api ; leveling the prediction accuracy of different region (this is trade-off between efficiency and equity, but public sector can enhance efficiency by providing api so that wheels don't get reinvented) by the way, ISO standard decomposes: (efficiency, effectiveness, satisfaction (equity need distributional - observe non-user's behavior (fallacy in transportation policy) knowledge)) - recommendation level (chatgpt); individualize communicaton QQ. Frontier evolves, which evolves faster between equity (behavior) and efficiency (technological development + behavior from adoption)? RO-based is much longer six month refresh schedule (longer~STR_STATE), RL-based is day to day operation (shorter, OP_STATE), demand prediction? (database); analogy for transportation mode (pivot path?); how subway, commuter rail, shuttle bus, new mobility service are embedded? ![[Pasted image 20230911103449.png|500]] QQ. dispatch rule for rule-based vs rl-based? (filter actions); hate - ![[Pasted image 20230911104856.png]] - long relationship of chief administration officer, - certain uncertainty that bus driver has but ai doesn't (local knowledge, common sense judgement) - labor shortage and abstentism, we can't use just control, RL-engine for decision - compliant rate (with ai guidance - 0 is recommneded adjustment relative to original schedule, add one min, move for; further - less compliant; aggressiveness of the recommended control actions)