Social Computing Systems

Transcription

1 Social Computing Systems Walter S. Lasecki EECS 498, Winter 2017 (

2 A Quick Note on MVPs Minimum Viable Product Emphasis on the work `product` When prototyping, you won t always HAVE a product In fact, you almost never should in an early prototype Prototypes INFORM the product, they are not the product itself You won t have a market-ready product in mid-april, you ll have an MVP

3 Last Class (review) Methods for analyzing incentives in participation Game theory helps us predict what will happen if our users are rational agents

4 What does a solution look like? A decision or policy. How do we get there? - Compare outcomes P1 - Pick the best one Reveal Reveal Ex: Prisoner s Dilemma Hold Out (5yr, 5yr) (10yr, Free) (Free, 10yr) (1yr, 1yr) P2 Hold out

5 What is the Game Configuration?

6 What is the Game Configuration? Cooperative/Non-Cooperative Simultaneous / Sequential Discrete/Continuous (separable?) Length Perfect/imperfect info Symmetry Zero-sum / non-zero-sum

7 Today Understanding how design can influence participation Mechanism design helps us build reward schemes to encourage the right behavior How else can we include these ideas in Social Computing Systems? Gamification helps us connect rewards to non-monetary settings How do these incentives play out at scale? Markets are an example of how incentives can lead to interesting, useful outcomes

8 Intro to Mechanism Design (Just enough to confuse your friends at parties)

9 Changing the Rules of the Game The other side of game theory: writing the rules Influence behavior by setting up the game

10 Incentive Mechanisms in Practice

11 Incentive Mechanisms in Practice

12 Reputation Systems Honest P1 Honest Cheat (R + R, R +R ) (R, R +R ) Later Later Later Cheat P2 (R +R, R ) Later (R,R )

13 Reputation Systems (in Collective Intelligence systems) Honest P1 Honest Cheat (R + R, R +R ) (R, R +R ) (R (R + R, R +R ) Later Later Later Cheat P2 +R, R ) Later Later Later

14 Reputation Systems (in Collective Intelligence systems) Honest P1 Honest Cheat (R + R, R +R ) (R, R +R ) Later Later Cheat P2 (R +R, R ) Later Later (RN + RL, RN + RL) (RN, RN) (RN, RN) (RN + RL, (RN, RN) RN + RL) (RN, RN) (RN, RN) (RN + RL, RN + RL) (RN, RN)

15 Reputation Systems (in Collective Intelligence systems) Honest P1 Honest Cheat (R + R, R +R ) (R, R +R ) Later Later Cheat P2 (R +R, R ) Later Later

16 Reputation Systems (in Collective Intelligence systems) Honest P1 Honest Cheat (R + R, R +R ) (R, R +R ) Later Later Cheat P2 (R +R, R ) Later Later

17 Reputation Systems in Practice

18 Gamification

19 Challenge Games have to be challenging! Without needing effort, rewards aren t rewarding But not TOO challenging Types of challenge: Mental/sensory (puzzles, strategy games, Tetris) Dexterity (reflex games, first-person games, etc.) Team coordination (Space Team, recent Mario games) Creativity (Minecraft, Mario Builder, etc.)

20 Measures of Success Scores Rankings / leaderboards Levels Achievements / badges

21 Status Sharing

22 Uniqueness Story/setting elements Gameplay elements Team/multiplayer elements In-game items / interactions Atmosphere

23 Uniqueness Story/setting elements Gameplay elements Team/multiplayer elements In-game items / interactions Atmosphere

24 Why Games?

25 Gaming Interactive storytelling (making it a sharing medium) Supports enjoyable (casual) interaction Skills practice (teamwork, dexterity, etc.)

26 Gaming Interactive storytelling (making it a sharing medium) Supports enjoyable (casual) interaction Skills practice (teamwork, dexterity, etc.) Games engage us.

27 Serious Games

28 Techniques for Gamification Leveling Numeric levels Badges Points / scoring Add an underlying story Aesthetic appeal (visual/art) Goal + challenge E.g., puzzle, mission, etc.

29 Techniques for Gamification Leveling Numeric levels Badges Points / scoring d r a rew l l a re story Add ane underlying a s e h TAesthetic appeal (visual/art) Goal + challenge e c n s/i E.g., puzzle, mission, etc.!! s e ntiv

30 Limitations of Game Theory and Mechanism Design (a few of them anyway)

31 Uncertainty of Player Strategies

32 Uncertainty We never really know exactly what people will really do Game theory helps us take a logical guess

33 Reward Strength We don t always know how much our rewards impact people Most of our rewards in Social Computing Systems are pretty minor - Usually: points, stars, karma, etc.

34 Reward Strength We don t always know how much our rewards impact people Most of our rewards in Social Computing Systems are pretty minor - Usually: points, stars, karma, etc. - Hardly ever: - piles of cash - speed boats - large parts of WY - immortality

35 Limits of Collective Intelligence in Games Collective Intelligence methods use agreement to determine correctness This leads to regression to the mean Why? How?

36 Example: ESP Game

37 Example: ESP Game Sony MDRAS40EX Sport Over-the-Ear Headphones!

38 Example: ESP Game Headphones...

39 Example: ESP Game Sony MDRAS40EX Sport Over-the-Ear Headphones!

40 Example: ESP Game Headphones...

41 Example: ESP Game Headphones...

42 Expert Example: ESP Game? ^

43 Plenty of Topics We Didn t Cover! Multiple Equilibria Pare to Op s e r u t c u r t s e k p i Kr timal i ty Incentive Compatibility

44 Markets crowds, prediction, and valuation

45 What is a market? A market is one of the many varieties of systems, institutions, procedures, social relations and infrastructures whereby parties engage in exchange. Wikipedia A place for exchange between people. No surprises here But what do we really mean? Examples: stock markets, futures markets

46 Analysis of an Example Exchange One seller (S) with an item of no use to them any longer S One buyer (A) who makes an offer of $25 Result: S sells to A $25 A

47 Analysis of an Example Exchange New buyer (B) comes in, offers $35 S Result: S sells to B $35 B $25 A

48 Analysis of an Example Exchange A offers more ($50) S Result: S sells to A $35 B $50 A

49 Analysis of an Example Exchange A offers more ($50) S Result: S sells to A -- B $50 A

50 Analysis of an Example Exchange What happened? Seller A offered more Competition resulted in a final market price This price is a valuation on the product The highest price at which it s still worth paying This turns out to be very, very accurate in the long term!

51 Why are markets good at prediction? Collective intelligence many of us are better than any of us Ex. (From Wisdom of Crowds by James Surowieck) In 1906, there was a county fair in Plymouth, England 800 people guessed the weight of an ox oxes are big, people aren t great at this The average of these guesses was within 1lb of the ox s true weight

52 Why are markets good at prediction? Collectively, all of the buyers in a financial market are trying to predict the true value And collectively, they re usually better than any one person

53 Prediction Market Examples Iowa Electronic Markets predict presidential race PredictWise everything from Oscars to Sports Vegas line sports, mainly (through almost anything else too) What about the Lottery? Other examples?

54 Why do we care in Social Computing? Predict value Collective prediction can yield better data for a service Yelp restaurant reviews / Uber driving ratings ebay seller ratings any other user-generated reviews, at least over time Collect information ebay/amazon/etc. product prices Pintrest Facebook Other examples?

55 Limitations of Markets It takes time for these collective decisions to stabilize naturally We ll see how we may be able to overcome this later Missing knowledge may still exist in the system/collective Popular opinion is not always well-informed (again, regression to the mean ) Subject to bias, incorrect info, etc.

56 Generalizing: Collective Intelligence Emergent behaviors Not constrained to humans swarms of animals also exemplify this (e.g., flocks) Evolutionary algorithms Wikis / info. curation (libraries)

57