CS278 Notes PDF

Preview

Summary

Notes I took as a student in Michael Bernstein's class at Stanford University, CS278: Social Computing....

Description

Lecture 1 4/1/19: - Social computing systems are computational systems that mediate social interactions - Can help with productivity, fun, decision making, etc - Social computing design asks how to fashion social environments online in ways that support participants in achieving goals - Design challenges: make people feel safe / post memes / want to engage - Serious responsibility because of potential for adverse effects - We’ll learn about how to design these systems, the fundamental principles by which they operate, and the challenge/responsibility of designing effectively and ethically - Lesson 1: Every social system is designed - If you don’t design, you default and the default is often worse - Lesson 2: Don’t Kludge (paste together bits of other apps) - Higher order social dynamics matter a lot more than specific UI elements - Surface features of a meme: shareable URL, simple message, low friction to share, catchy hashtag - But these alone aren’t enough and only this means you aren’t trying - Cultural innovation sometimes come most from people who are intermediaries between core (mainstream group of people) and periphery (outside, smaller group) - Maybe intermediary takes things that only they see from periphery and help spread it and make it mainstream in the core - Experiment tried seeing effect of social influence on song popularity - popular songs become more popular when people can see that it was already more popular - Social proof - when people copy each others’ behavior - We assume others know what they’re doing so we do the same - Viral truth - fake news was more novel, humans diffuse false news more quickly than real news Lecture 2 4/3/19: - Jane Jacobs wrote book with Eyes on the Street theory that neighborhoods/places with more people around will keep people on good behavior - How can we design bustling environments with eyes on the street and not ghost towns? - Sociotechnical system - system defined by social interactions backed by technological infrastructure - Term captures how technical elements aren’t enough to determine its behavior - Interplay of tech and users are what makes things tick - Individual factors - Intrinsic motivation - derived from my own desires to complete a goal (pleasures, hobbies, developing/demonstrating a skill) - Extrinsic motivation - not derived from my relationship with the goal (money, graduation, points, badges) - Motivation crowding - If you take something intrinsically motivated and add extrinsic motivators, it may reduce overall motivation level (israeli preschool study)

-

-

-

-

-

-

-

Big problem with gamification - adding game aspects / rewards lessens people’s intrinsic motivation to use that platform Gamification can sometimes help though, especially for things that are weaker intrinsic motivators and have more autonomy - badges can help measure intrinsic goal of progress Start out with intrinsic motivation, then transition to add extrinsic motivators to keep people invested Channel factors - Minor upstream changes in a decision process that produces larger changes downstream - Students more likely to get shot if they got map/hours of health center, even if they already knew those things Social loafing - If there are more people around, each individual puts in less effort - Blindfolded tug of war with telling people different numbers of teammates - Can combat by calling out people’s uniqueness to draw attention to how group may not be as big as we think Reciprocity - More willing to give back when someone does a favor for us even if we didn’t ask for the favor Contribution pyramid - about a 10x drop-off between levels often - Mods - Contributors - Co mm en ters - L I K E R S - L U R K E R S How to design - Support both kinds of motivations - ID channel factors and manage carefully - Combat social loafing - Support autonomy

Lecture 3 4/8/19: - Norms: what makes similar platforms have such different results? - How does design influence norms? - How do norms in turn influence how people use the design? - We are different people when we are in different spaces (Goffman “Presentation of Self”) - How is design of social computing system influencing what side of people comes out - Norms: informal rules that govern behavior in groups and societies - Descriptive norms: norms that describe common behavior - Social proof! When we’re not sure what the right behavior is, we take cues from what other people are doing in that environment

-

-

-

-

Injunctive norms: norms that describe what you should or should not do (rules) Can make norms salient by showing that comments were removed by mods - Thus encourages behavior in line with norms - Could also make it more salient that people are breaking norms so it’s ok for you to break that norm too, even if it gets removed Stack Overflow started with 500 people in closed beta test for 3 months, helped curate a sense of community with enthusiastic members and was full of positive examples of helpful answers, set examples of good norms for future users to follow once public - New users also can only ask and answer, can’t upvote/downvote/edit to make you work for full privileges; makes new users absorb norms of what makes a good question/answer Defaults influence norms: Instagram default is public, long process to change to private - Sets descriptive norms because most people don’t change defaults Defaults can be a huge channel factor that designers have to choose that will greatly influence norms on a platform Identity influence norms - anonymity/pseudonymity can foster more negative environments, but also possibly tighter community - But forcing using real name can make people uncomfortable Culture influences norms - Certain cultures can set different ideas of what’s ok to do on a platform Aesthetic choices can delineate status and attract certain kinds of people

Lecture 4 4/10/19: - Waste of time/energy/resources to get idea and just build it - Go through prototyping process to find issues with it early on - Prototyping - Gas state: low-fidelity / paper prototype - Liquid state: get richer - Solid state: more detailed - Paper prototypes aren’t the best for social computing systems - Can’t simulate interactions between users - Can’t simulate evolving dynamics over time - Focus instead on what the big risky unanswered question is about the idea - Use social bricolage: use existing systems to simulate features that your app may have - Texts for push messaging, GDocs for collab, Slack for chat, etc - Prototype simple social formulas - Identify simplest core set of social interactions and test those first - Don’t prototype sparsely, instead get a tightly connected group and have them try it out - Sometimes need to bootstrap content to make it feel more bustling, less ghost town - Diary study of structured notes about when/how they used the system each day can be good way to understand how people use it / how it makes them feel

-

-

Cold start problem where there’s none on the platform yet so there’s no reason for anyone to join it - Hard to overcome this and hit critical mass when no one has a motivation to do it - Starting broad and trying to get all your users at once typically fails - Better to start by targeting a small, focused, engaged core, you can get things going well there and then broaden out - Can have a good sense of users and their issues and expected norms Have to ask who is your prototypical user? - Who isn’t a prototypical user and do we need to change that?

Lecture 5 4/15/19: - Wikipedia grew fast but peaked and dropped off to a fairly constant number of monthly active editors - Possible reasons: social loafing, nothing left for general user to add, novelty effects, regulations - But different languages of wikipedia reached this point at different times - More wikipedia action has moved to discussion pages rather than articles - Facebook started narrow and broadened out to larger and larger groups of users - New members = new norms, new cultures, etc - Broader participation exposes cultural rifts - tinder auto-banning trans women after cis men reported them - Newcomers challenge norms - New members more energetic and participate more - Interested in a broader range of discussion - Haven’t been enculturated, don’t know norms, more likely to breach them - Puts strain on existing members - Results in Eternal September: permanent destruction of a community’s norms due to influx of newcomers (September is when new college freshmen start) - To survive an Eternal September, pages tend to need strong moderation and increased provision of attention - Invisible labor and moderation - To survive massive growth, mods need to shepherd behavior toward desired norms (remove bad content, etc) - Invisible labor - term draws from women in the home doing lots of work to cook/clean/take care of kids, but don’t get recognized for that hard work - Examples in social computing: moderation, paid data annotation, server administration - Most people just see the results of the curation, not the curation itself - Can put serious strain on emotional wellbeing of mods because it feels thankless - But moderation works! Has shown proven results - One strategy is community moderation (privately flag things, upvote/downvote - Mod bots also help moderators ID potentially problematic content

-

-

-

-

Just-in-time reminders make norms salient just before someone is about to post something - Hellbanning: put people who you want to “ban” in a circle of hell that only other trolls can see so they don’t realize they’re banned and they also don’t bother majority of users Information overload and the economics of attention - Information consumes the attention of recipients and lots of platforms are competing for our attention - Humans as information processors model says more info = higher performance - Other model represents information overload as concave-down parabola curve - With more and more content on a site, users tend to get overloaded by more information and sometimes even leave - Most content that makes it to the top of reddit has been posted multiple times, because not everyone sees everything that is posted - Ranking information presented to users is an unintuitive mental model, but the front page makes more sense - Chronological info presentation is a simple mental model but can be vulnerable to domination from spammy accounts Designing for global usage - How do you A/B test when you can’t cleanly divide between different groups - Most common thing is to do country comparisons where you deploy different versions to different countries - How do you build empathy for users other than you yourself - Show engineers videos produced by users - Bring in stakeholders to have conversations - Don’t always assume that you can create empathy - Focus first on building tools to empower users to manage overwhelming amounts of content Back to wikipedia - Started small, little moderation needed - Formula works and wikipedia grows - Volume of low-quality contributions grows and puts more strain on core group - To deal with strain, start rejecting more posts and implementing bots - Newcomers less likely to stay and keep contributing after content gets rejected

Lecture 6 4/17/19: - Do social computing systems make us lonely? - Not all of our relationships are the same - Strong ties - trusted friends and family, in social networks that we are deeply embedded in, provide social/emotional support, improve mental health, communicate with us through multiple channels - Design goals are often to maintain or deepen strong ties - typically communication is more in private channels (narrow-casting), broader

-

-

-

-

-

-

norms allow more wacky interactions, require or encourage some previous offline connection, don’t advertise clout - Weak ties - rough acquaintances, thin context, sparse interactions, fewer different kinds of interactions, connections to parts of social network that we don’t inhabit, can help people come up with better ideas because they have more diverse views, Facebook good at it, LinkedIn too, valuable for new perspectives and new job opportunities - Typically have more weak ties, and majority of content that you see is from weak ties - How can you make this weak tie content feel alive and make people feel comfortable / at home Easy in social situations to throw quick and easy queues that you care about someone - Other signals are more costly to produce and so are more honest - spending time on something for someone shows honest caring / commitment No tie systems - is the point to build up ties? - Maybe sometimes, but a lot target more finding new content that you like or having fun with people with similar interests/identity - weak/strong tie platforms are bond-based, these are more identity-based Designing for identity-based groups - Highlighting group’s unique identity increases people’s commitment to the group - Let people express that shared identity Tie strength varies over time and social media use can change it, but strong communication (DM, wall posts, etc) increases tie strength much more than just seeing and liking public posts Social computing systems can do a pretty good job of modeling tie strength between users on the site - Newsfeed rankings can take into account predicted tie strength - Friend suggestions draw on notions of networks - Ties don’t tell the whole story however One-on-one communication from strong ties improved psychological well-being in a study, but all other stuff was just kinda meh and didn’t improve anything

Lecture 7 4/22/19: - Studies have shown that colocated teams that share a space are twice as effective as a distributed team, even with all the collaboration technology that exists - Lots of theories of best ways to create good teams, lots about why others are wrong - Now, even colocated teams use some mediating systems/tech - How can we make tools that will make distributed collaboration as effective as colocation? - Features of many existing tools - Easy to use solo or as a team, smooth UI, version control, compartmentalization, integration

-

-

-

-

-

Teamwork is harder in isolation because group actions are interdependent and depend on things others are doing - Need to be able to ask questions, reveal ignorance, monitor one another, hold each other accountable - Big theme in many platforms in awareness - allow each other to see what others are doing...but don’t want awareness to go too far and reveal too much - Social Translucence - give enough information to let natural social cues take over, without giving too much - ST also requires accountability - others know that their activity can be seen and are held accountable for not goofing off - If done correctly, ST supports interdependent work while maintaining plausible deniability Designing to make it as if you were right there looking over someone’s shoulder is the wrong goal - Too many subtle cues in the situation of being there to accurately recreate it Instead, find ways that technology offers affordances that allow you to be better than real life - Skype can do real-time translation between languages - Accessibility - Referring back to past interactions - Tracking dynamics - Adding metadata - Anonymization Grudin’s paradox - Socio-technical systems get abandoned because they seem to be benefiting everyone - except the people who are expected to use it - Manager wants one thing, engineers don’t care as much about it - We want API to be documented, but the people who know how it works already know how it works - Try to avoid by providing benefits to all users - not asymmetric benefits - Companies are often later adapters of these technologies because they start out wary of the new thing A lot of these things depend on the team and what they want/need in a system

Lecture 8 4/24/19: - Mike Krieger- guest speaker, co-founder of Instagram Lecture 9 4/29/19: - Dude in 1906 found that the average of everyone’s guesses would get within 1% of the true value (of the weight of an ox lol) - Wisdom of Crowds! - Wisdom of the crowd can be applied to things with: (according to Jeff Howe) - Diversity of opinion - Decentralized

-

-

-

- Aggregation function On the flip side, can’t solve problems where people think the same thing, people can communicate, and there’s no way to combine the opinions General algorithm: - Ask a large number of people to answer a question (independent answers from people with a basic understanding of the phenomenon in question) - Aggregate results (mean/median/etc) Accuracy reduced if you show people running total as more people are voting If initial guesses are inaccurate and public, the crowd never recovers Crowds are more consistent guessers than experts! Charles Babbage hypothesized that 2 people doing the same thing in the same way will make correlated errors Bayesian Truth Serum: have people answer a question, and also predict what percentage of people they think will choose each choice - Calculate population frequencies xbar k for each option k and the geometric average of predicted frequencies ybar k - Evaluate each answer according to its information score: log (xbar k / ybar k) Crowdsourcing - term coined by Jeff Howe in 2006, when someone raises a question or problem and opens it up to the world to solve - Powerful but also easy for a couple malicious individuals to mess things up - Need to have effort to vandalize >>> effort to undo vandalism - Can try to gauge quality of someone’s answers by interspersing “gold standard” / “attention check” questions that you know the answer to in order to make sure people aren’t just screwing around - Could also try to train a ML model to predict whether or not someone is paying attention based on time delays, backspaces, mouse clicks, etc - More common to have mental model misalignment (people don’t understand intended goal) than strategic dishonesty (deliberately screwing with your results)

Lecture 10 5/1/19: - Parallel, independent contributions work well for crowdsourcing tasks - But interdependence between integrated contributions is harder to manage - Requires direct collaboration between participants - Find-fix-verify - Design pattern for open-ended tasks - Find possible problems - Offer different fixes for the problems - Verify validity of fixes - Crowd production - workflows designed to facilitate multiple people working together on a project - Peer production - decentralized conception, diverse motivations, results treated as commons (non-rival - one person’s use doesn’t inhibit others’ use), no contracts for work on project

-

-

-

-

Benkler said peer production outperform traditional firms when there is a strong intrinsic motivation and work can be broken down into granular and easy-to-integrate tasks Open source governance models - BDFL: benevolent dictator for life who makes all final decisions for project - Meritocracy: top contributors are granted decision-making rights, policy decisions via committee vote - Liberal contribution: allow as many contributors as possible and use consensus-seeking for policy seeking Peer production is limited most by its modularity - its ability to be broken down into small independent components - Many projects can’t be predicted well enough to modularize them Hybrid peer production: peer production projects backed by traditional organizations - Backing companies help manage more complicated / coordination aspects

Lecture 13 5/13/19: - Premeditated anti-social behavior: trolling - People take advantage of social media to harass, spread fake news, etc - About half of online users have been harassed - Some sites are turning off comments because they had gotten too out of hand - Trolling: intentional disruption of an online community / any behavior falling outside acc...