1. Introduction
    1. Machine intelligence
      1. AI - not really intelligence - but critical component of intelligence
        1. AI as prediction technology
          1. central input to decision making
          2. economic has framework to understand decision making
          3. new and poor understand implications of advances in prediction technology
          4. old an well understood logic of decision theory
          5. way to understand AI
          6. tradeoffs
          7. more speed
          8. less accuracy
          9. more autonomy
          10. less controll
          11. more data
          12. less privacy
          13. Book: method
          14. evaluate pro and cons of applied AI to particular business
          15. optimal ai strategy
          16. optimal ai tools
  2. Cheap Changes Everything
    1. economic understand business implications of cheaper prediction
      1. prediction machines will be used for traditional "prediction" tasks (inventory, demand forecasting, and new problems - like navigation)
        1. drop of cost for prediction, will reflect on other costs
          1. diminished values of substitutes
          2. human prediction
          3. increase on value of complements
          4. data
          5. judgment
          6. action
        2. organization - exploit prediction machines
          1. AI insight - reframing problems as prediction problems
          2. adapting ai tools
          3. assisting in adapting current strategies
          4. changing strategies
          5. opportunities to do what before was not possible
          6. if Amazon know what customers want - change from shop/ship to ship/shop strategy
          7. new set of tradeoffs (AI impact society) - based on needs and preferences - different - cultures and countries
          8. 1. Prediction
          9. 2. Decision making
          10. 3. Tools
          11. 4. Strategy
          12. 5. Society
  3. 1. Prediction
    1. prediction machine magic
      1. filling in missing information
        1. taking information you have (data)
          1. generate information you do not have
          2. about future
          3. about present
          4. about past
        2. improvement in accuracy
          1. deceptive (not intuitive)
          2. can be transformational
          3. seems magical
      2. usage
        1. traditional usage of predictions
          1. fraud
          2. credit worthiness
          3. health insurance
          4. inventory management
        2. emerged
          1. object identification
          2. language translation
          3. drug discovery
          4. New usage
    2. Why it is called intelligence
      1. machine learning science has different goals from statistics
        1. statistics - being correct on average
          1. articulated of hypothesis / human intuition in model specification
        2. machine learning - operational efficiency
          1. could have biases
          2. experiment and improvements
          3. advantage of rich data
          4. fast computesr
          5. no need for specification in advance
          6. much more complex models with anticipation of interactions
          7. advanced in machine learning
          8. advances in AI
          9. 1. system predicated on this technique learn and improve over time
          10. 2. produces more accurate predictions then other approaches
          11. 3. enhanced accuracy - performing new tasks (translation, navigation) - previously exclusive - human intelligence
          12. prediction is central to intelligence
    3. data is new oil
      1. types of data
        1. 1. training data
        2. 2. input data for predicting
        3. 3. feedback data
          1. improving prediction accuracy
      2. data collection is costly
        1. investmet
          1. how much is enough
          2. how intrusive collection process is
          3. balance data acquisition / enhanced prediction accuracy
          4. ROI for each type of data
          5. how much cost to aquire
          6. how valuable prediction will be
          7. approaches
          8. statistical point of view
          9. data diminished value
          10. more data - more value
          11. economic point view
          12. ambiguous
          13. more data can reach threshold of new insight, but when threshold is reached - new data is not very useful
          14. must understand relation between adding more data and accuracy of prediction
    4. new division of labor
      1. humans - poor prediction in certain conditions
        1. overweight salient information
        2. do not account for statistical properties
        3. machines and humans distinct strength and weaknesses in context of prediction
          1. adjusting division of labor
          2. prediction machines - better in complex interaction among different indicators / rich data
          3. human accuracy diminishes / relative to machines
          4. humans is better when understanding data is advantage - especially: tin data
      2. Taxonomy of prediction settings
        1. known/knowns
        2. known/unknowns
          1. machines - do not known
          2. people - much better
          3. understand inaccuracy
        3. unknown/knowns
          1. machines makes predict, but wrong
          2. data form decision
          3. data comes from decision
          4. machines do not understand how data is prepared
          5. the counterfactual problem form machines
          6. reverse causality
        4. unknown/unknowns
          1. machines can't predict
          2. people can't predict
      3. Scales
        1. prediction machines scales
          1. unit cost per prediction fall as frequency increase
          2. most predictions predicated on routine, regular data
          3. rare event - need help from human
        2. Human machines do not scale the same way
          1. use cognitive model how world works
          2. prediction on small amount of data
          3. human prediction by exception
  4. 2. Decision making
    1. unpacking decisions
      1. Reasons of prediction machines usefulness: prediction
        1. Cheaper, faster, better predictions
        2. Key ingredient in uncertainty
        3. ubiquitous - in economic and social life
      2. prediction is not a decision
        1. components
          1. prediction
          2. impact humans / assets
          3. value of substitutes (human prediction) - decline
          4. value of compliments (data collections, judgment, actions) - more valuable
          5. judgment
          6. knowledge
          7. judgment value increase, as prediction become cheaper
          8. action
          9. outcome
          10. 3 types of data
          11. input
          12. training
          13. training - relations between data
          14. feedback
    2. value of judgment
      1. prediction machines increase return to judgment
        1. lowering cost of prediction - increase value of understanding reward associate with action
          1. judgment is costly
          2. uncertanity
          3. payoff for different actions
          4. time
          5. effort
          6. experimentation
    3. predicting judgment
      1. humans - 3 types of data machines don't
        1. human senses
        2. ultimate arbiters
        3. privacy
      2. machines can learn to predict human judgment
        1. learning limits
          1. lack of data
          2. little data - humans uses, machines don't
          3. analogy
          4. models
          5. individual preferences
          6. machines - bed in prediction rear events
    4. taming complexity
      1. compromising - complexity to predicting
        1. human satisfying
          1. good enough
        2. better prediction
          1. less satisfying
          2. many if - then
    5. full automated decision making
      1. machine perform all components
        1. task is fully automated
        2. humans removed from the loop
        3. automation first - when / where higher return
          1. 1. other elements are already automated, except prediction
          2. 2. return of speed of action is high
          3. autonomous vehicle
          4. 3. return of reducing predicting time is high
          5. space exploration
          6. Risks
          7. externalities
          8. regulation - barrier
          9. law - regulatory issues
  5. 3. Tools
    1. deconstructing workflow
      1. ai tools - point solutions
        1. related to specific task
      2. braking down - in set of tasks
        1. ROI - building / buying ai
          1. working top / down in building ai solutions
          2. immediately test benefit
          3. increasing productivity
          4. rethinking
          5. reingeneering whole workflow
    2. decomposing decisions
      1. decompose tasks
        1. estimate benefits / cost of implementation
        2. rank ROI - highest/lowest
        3. implementing AI tools - until ROI makes sence
          1. use ai canvas for decomposition process
          2. every task / decision
          3. data (3 types)
          4. what have to be predicted
          5. core prediction
          6. objective
          7. action possibilities
          8. outcome possibilities
    3. job redesign
      1. job
        1. collection of tasks
          1. some tasks automated
          2. other tasks change
          3. new task created
          4. AI tool
          5. implications
          6. augment jobs
          7. contract jobs
          8. reconstitution of jobs
          9. eliminate tasks
          10. add new tasks
          11. shift emphasis on skill required
  6. 4. Strategy
    1. AI in see sweet
      1. not delegate AI strategy to IT department
        1. beyond enhanced priductivity
        2. changing strategy
          1. core tradeoff in business model
          2. shop/ship vs ship/shop
          3. trade of - uncertanity
          4. decrease expenses but increase uncertainty in choice
          5. AI help in handling of the uncertainty
          6. tool
          7. implementation in one part - affect other parts of business
          8. redesign workflow
        3. increase value of complements
          1. judgment
          2. change in organizational hierarchy
          3. actions
          4. competitive advantage
          5. data
    2. AI transforming business
      1. human judgment
        1. more important
      2. subjective performance evaluation
      3. key strategic points
        1. where business end
        2. where another business begin
        3. boundary
          1. uncertainty influences choice
          2. prediction machines reduce uncertainity
          3. increase ability to write contracts
          4. increase incentive - contract capital equipment and labour
          5. data
          6. prediction
          7. action
          8. decrease incentives to write contract
          9. labour
          10. judgment
          11. if well specified - programmed
          12. do not need humans
          13. in house enployment arise
          14. contracting out labour - fall
          15. increase incentives to own data
          16. decide
          17. purchase predictions
          18. purchase data
    3. learning strategy
      1. shifting to AI first
        1. downgrading previous priorities
          1. real trade off
          2. maximizing AI goals
          3. compromising other goals
        2. disruptive
          1. startups - focused on AI, not based on old priorities
        3. products inferior at first
          1. time to train
          2. learning and improving
    4. Managing AI risk
      1. many types of risk
        1. predictions lead to discrimination
          1. liability
        2. ineffective if data is sparse
          1. quality risk
          2. unknown/known type
          3. prediction with confidence, but wrong
        3. incorrect input data
          1. fool prediction machines
        4. diversity of prediction machines
          1. individual / system level outcomes trade off
          2. less diversity
          3. benefit individual level performance
          4. increase the risk of massive failure
        5. interrogated
          1. intellectual property thief
        6. feedback can be manipulated
          1. destructive behavior
  7. 5. Society
    1. beyond business
      1. many choices
        1. productivity / distribution
          1. enhances productivity
          2. not evenly distributed welth
          3. increase competition between humans for the remaining tasks
          4. lowering vages
          5. increase value for owner of capital
          6. skill biased
          7. benefit highly skilled worker
        2. innovation / competition
          1. scale economies
          2. increase returns
          3. better prediction - more users
          4. more users - more data
          5. more data - better prediction
          6. monopolization
        3. performance / privacy
          1. access to more personal data
          2. reduced privacy
  8. Finite type of actions
    1. Judgment - transfer to prediction machine
      1. reward function ingeneering
        1. automation of decision
    2. to many options
      1. judgment - transfer to humans