SXSW 2008: The Science of Designing Interactions

by danhon

The Science of Designing Interactions

Room C
Sunday, March 9th
10:00 am – 11:00 am

Andreas Weigend Principal, People & Data
Ming Yeow Ng Discoverio

AW: Goals of today’s session – want to show you situations where design itneractions can be crucial, show examples of underlying patterns extracted and then will work with the audience and give us other examples and insights that we don’t have.

7 design patterns:

1. Focu on designing interactions
2. Build, experiment and measure
3. Give user metrics of his standing – helps users to play

4. Help the user decide actions – provide an economic framework of cost
5. Frame interactions as costs, reward, risk
6. Introduce currency for interactions
7. Create mechanism for discovery

(Slides will be up at discoverio.com/chat)

From participation to interaction – I’m not an economist but I know that the cost of communication, the cost of collaboration has dropped dramatically. The effect of that? We’re exploring the effect of this dropping cost. That means that interactions have dropped – the cost of interaction, the physical cost, other costs have become the most imoprtant cost. Collective intelligence is one of these elements that has been possible now. CI is in 3 things: problems that you can decompose, e.g. Mechanical Turk on Amazon – decompose a problem into small problems, farm it out, no interaction needed. Second one -t he portfolio approach, is where do you benefit form average – people have a portfolio, in order to come up with less risk for the same expected return. Thirdly, where do interactions actually pay a difference – prediction markets, people interact with each other, as a physicist, if you crank up the volume, there’s positive feedback, how do you creat e itneractions to create truly emergent properties.

We are not just architects of participation, but also of interaction and in the same word – as a design pattern.

Focus on designing interactions. We only talked about what type of interactions we’re considering – it’s not just UI but it’s itneractions between people. Interactions also ahve become very simple, what we have here is that there is both a quantitative way that you have many more interactions than before and a qualitative smaller spectrum that we’ve had. For the qunatitative, Twitter – it’s such an easy thing – I just twittered ‘where’s the party’ and someone told me we’re hanging out at Six. These are low weight interactions, the chatroom for the session is already filling up. but not only this, secondly there are lots of qualitatively different kidns of interactions – is it’a universe or a jungle of interactions?

Communcation cost is zero, thus we can design lots of interactions.

Fact number two is building stuff is cheap, so we can be bold, we can create, experiment and measure. What should we measure? The metrics of interaction aare also because there are so many things we can measure – there is a spectrum of actions that can be measured – interactions with content only through to with people. For yourself – annotating for yourself, starring something, primarily content interactions, like commenting, amazon reviews, sharing somethign with an audience, intrinsically both – forwarding and twittering, and with others only, walls and fans.

Now, when we talk abotu metrics, many of these metrics which people invented ten years ago are metrics which are of interest to the company – pageviews, uniques, all of these things from print, when you know how many copies you’ve shipped, but they don’t tell you anything about engagement. So here are some examples of metrics of engagement and what I want to shift here in my third point, is focus on metrics for users – truly measure user engagement. It’s engagement we’re interested in not just stuff about the page.

MYN: What Andreas just ran through was metrics for companies, but now we’re focussing on metrics for users. What we really need to do is give users mtrics for themselves so that they can change their behaviour in accordance with metrics. There’s a massive hierarchy of this, I propose that we need to provide them metrics: ranking achievements, popularity, feedback on past. [Showing Maslow's pyramid].

E.g. Yahoo Ansewrs – give points and leaderboards.

What’s more interesting is that metrics of popularity achievement. Standard over the past two years – how many friends do I have on Facebook, and then moved from that to have one very binary way of looking at it – popularity, we’ve moved from that to become more granular to see how someone can perceive popularity in the world – popularity on Twitter. I’ll give you another example which is Yelp – Yelp gives the user many different very small fine grained metrics that let them measure them by.

Popularity – how many fans, compliments, all these things are light one-way actions taken by other users, you can have one compliment. Another metric is achievements – restaurants that you’re first to review, are you ’08 or ’07 elite, this is *what I want to be* this is *what I need to get there* – are my reviews useful, funny or cool – *all these things are positive*. Although they accumulate to a good holistic feeling of themselves. It’s interesting how that’s compressed into one small box.

We talked about points int he first slide, what happens when we combine. Here’s an example from Facebook – Friends for Sale. Check it out, you can buy me as a friend. What you do is that you buy friends, you get cash every time you do that, every time you get bought and sold, you gain in value. Cash is a representation of power, your value is your desirability, so what – as you talk to a creator, he taught us that people are going crazy over points, to get people to log in every 4 hours to get points, someone in SA getting someone to pay $10k for points so he can buy attractive girls. This thing is pretty powerful. I know most of us don’t like Facebook apps but these are good interesting examples. So to round off this point, Facebook friends – Friends wall have 27k postings, that’s way beyond, 99% of the postings are please buy me, please buy me! 99% of postings are to be bought/sold or for pets.

Clear metrics: facebook, total number of wall postings.

Softer metrics of popularity. I think one thing when Facebook first got started was wall posting – total number does nothing for posting, why do I care about total number? But it’s a measure of people interested in you, it incentivises you to write on other peoples’ walls so they write back on to yours.

AW: Do you have any other suggestions or ideas, about short summaries of metrics. It’s easy to build, measure and to interact. What other ideas have you seen beyond the ones just presented?

Slashdot/digg rating systems – they’re good indicators for people and content, but not people to people.

Distance to the most popular people.

Rate my blog.

MYN: Next point – once we observe across a site what’s going on, then we can build models and make suggestions about what they might want to see – here are some examples, the first example – this is – how do we observe user behaviour across the whole site, this is interesting on Facebook. Suppose I post on a girl’s wall and then after that a guy posts above my posting, I’m 30% more likely to post again on her wall. So what will FB show me in my news feed?

Linkedin: Profile completeness that poitns out deficiencies.

Users with complete profiles are 50 time more likely to receive opportunities through LinkedIN.

Point out deficiencies – first establish what you think is a standard or norm and then tell them they’re deficient.

So another eaxmple is 50% zombies for biting opposite sex, 73% for vampires. Somethign he found out was that for zombies, if you’re a guy, it’s 50% bitten by a guy, but for vampires, it’s 73% you’re more likely to bite a girl. Vampire bites have a sexual connotation. Guys bite pretty girls. Girls – it’s something they wanted to do. You find interesting statistics when you go beyond analytics levels. Against gender, demographics, different user profiles.

AW: Looking for gender, different from looking for conversion rates. With this richer ways of interacting, both qualitatively many different apps, within apps there are different ways of tracking and with the richer way of quantitative things that you can add up or binary or we have the ability to truly have, be able to create interesting analytics across the site.

Understanding the economics of interaction: cost, reward and risk. Reward is expected reward, and for those of you with a background in machine learning, reinforcement learning is for where state action pair in a given state you have expected rewards and you might want to optimise those. Risk is the uncertainty of the reward at the moment and tells us what is the worst case, what happens with e.g. a 9% probability.

So, reward.

MYN: One clear point in rewards is becoming popular and cool, you really need to institute metrics for users, there have to be visible. It has to be easy for other people to see that you’re cool. You can’t make it a difficult process. Another thing here is give clear, immediate incentives. So Stanford Facebook, hey, Steve, we found your best matches out of 40m, the best ones for you. Invite your friends. This is powerful, a clear incentive. THe actino is clear as well.

AW: Two remarks. Those people were picked at random. Good matches – well, there must be some. The same way with Gold Box. We put things there at random, but we think they’re cool.

MYN: It’s like Yelp, you should buy this and you see lots of people buying it, so maybe I should buy it.

AW: At MIT, just came up with wonderful book: Predictably Irrational, experiment with students, taught one group, how much would you pay me if I ask 50 of your friends a certain set of questions about you – $5. Then the other group was told, here is the envelope and in that envelope are answers of 50 of your friends, how much do you want to pay me? I’ll pay $100. There’s something there already. So mayn of thet hings here we’re talking about have deep underlying values in behaviour, they’re not predictably rational, we evoled heuristics that help us deal well with typical situations, which of those help us deal well with online.

MYN: What you’re saying is if this said: “invite your friends to find your best matches”.

Incentivise a user – one last thing before you post. Posting a new topic on getsatisfaction – we’ve estimated the likelihood of your question getting noticed. Also, “I have this question too”. Suggest the action and the expected reward.

AW: Maybe we should say a couple words with NYT. It is the idea that you want to move the conversations to where the people are, as opposed to move the people to a website. E.g. United Airlines. We can write a response – it’s 4%. It allows you to have the conversation on the getsatisfaction site, then says it’s up to the company to respond.

The power of reciprocity. Werewolves. It’s likely that if you bite someone, someone will bite you back. Lightweight. You want to make it easy for the receipient of the beit to bite you back. And top friends – I call you a top friends. Are you going to call me a top friend back? Taps into a deeper desire of people to reciprocate.

AW: So that was rewards. Now the costs – lower effort and cognitive cost. We live in a world where if the product is good, transparency is good. Our advice is to reduce effort wherever you can.

MYN: You really want to make it very easy for the user. Capture the moment, when someone’s reading the review now, you need to make it easy to compliment the user from the review. Capturing the moment is really important. Here’s a good example of kickstarting – crowdvine. Talking to the CEO Tony, two simple buttons – fan/want to meet. What that guy gets – MY is a fan of you, come and see who he might be a fan of so it’s a positive and genuine, it’s very lightweight and one-way indiciation of itnerest. Fan, and “hello”. The problem with the online world is that too often we make things hard, everything must be formal email, it must be lightweight interactions to give people reasons to interact.

So Discovery.crowdvine.com. A network. Crowdvine is well designed. There is no shortcut to playing with it to understanding those points.

MyBlogLog: recent readers.

Xing: see which members have recently viewed your profile. people want to know who is checking me out? What Xing had two years ago and used to show it in talks and I didn’t know that someone could figure out I was checking Andreas out – people are curious about themselves. Yahoo bought myBlogLog.

Risk.

AW: Reduce and mitigate risk.

MYN: Hypothesis – the reason why Twitter succeeds is that we can tell everyone what we’re doing without incurring the risk of pissing people off. Twitter is a good way of letting people express themselves – optional replies and lightweight replies. Another thing is how do you reduce risk? Reduce the personal commitment that someone makes. Zoosk – you’re presented with a picture – you can flirt, wink or press next. I found that I wink 20 times and flirt 0 times and I don’t want to commit. Some of my friends say that the flirt – even if no clicks on it, makes people click on the wink. It’s about framing. This is lightweight in comparison. It’s about psychology can help you shape interactions in your app. This is behaviour economics. Adding choice – you can shift the behaviour. The Economist – by adding a choice that nobody picks, just Print, more people buy Print and Online. How can you frame things to get people to do stuff – Starbucks have XXL that no one buys and then just XL.

So 2 minutes talking abotu Match.com.

AW: Currencies. We went to match.com and one of the problems there is that people who have lots of time on their hands and are not that attractive hit up attractive people and who don’t ahve time on their hands. How do you deal with assymetry and imbalance? Same problem at LinkedIn those people desperate for finding a job are those that if you’re succesful you don’t want to deal with. So you create currency, we give people virtual currencies. You have to think about can you bank, can you trade, can you buy. Introducing a whole system allows people to then get through and say, if you want to talk to Jeff Bezos, you get one golden bullet a week – this is the one I have this week, I really want to talk to you.

Finally here, we have other examples of currencies. Standard example is reputations on eBay. We want to have shortcuts of predicting what the outcome is, ebay can’t tell you who’s a good guy, but what they have the past, they can tell you about behaviour in the past and people can make decisions based on aggregate data from the past.

Xobni – looks at the past emails. If you want to send me a message, but never send me a message, but the fact that you and MYN are friends, makes it higher probability that it will get through to me.

Discovery is the new cocaine. The world is changing whether you are there or not. That makes you come back. The same is the case with Facebook – did anyone kiss me? What happened? You have a stake in it whether it’s daytrading or facebook and it draws you back to see what’s happened.

MYN: So some examples of discovery: as users, we want to discover ourselves, people around us, and be discovered. It’s discrete components of discovery.

What others think of you: personality tests – eg. ENTP. THis is a new personality test, but you’re not taking it, other people are taking it for you. Characteristics. Again, Facebook is a good example of this, Compare People.

Flickr: interestingness, Etsy, many ways to shop. Helping users discover. Flickr is the most common example. Etsy – color, treasury, pounce, geolocator, etc. Unpredictability in the system, step out of traditional boundaries.

Digg: discovering via homepage vs discovering via users. Algo-based? Digg is algo based. As they shift to allowing users to discover other users – that’s more personal.

Crowdvine – updated profiles and comments are on the front page. You would want to update and comment because you want to be discovered. Once you do those things, you’ll be on the front page, that increases chances of being discovered. That’s powerful.

Facebook Feed: Timebased, to more intelligent. It knows who I’m interested in, My hypothesis is this is why people keep going back.

Discovery alone is not enough, I have discovered everyone in the room today, who should I talk to, connect with? This next point – give people a reason to interact. Reviews are the connecting point on Yelp, there are lots of people to connect, there are great starting points, also the Apache Social Network. What’s interesting is that there are people that work on projects, but they don’t know who they’re working with. What are your favourit projects, what have you contributed to? They discover who they’ve worked with.

I think we’ve moved from appraisal by pure interactions to really creating elements of addiction, things happen even when you’re not around, things are out of control, you put unpredictability and surprise elements. That will separate one generation of apps from the others.

Q: Currencies are typically particular to the system and the site. I wonder if it’s possible now and in the future to have a central bank that has a clearing authority. A currency of identity and reputation and in some way, the currencies play off of insecurities about ourselves. Would it be possible to create a central bank?

AW: The easy answer – money will be exchanged and the exchange will be floating. Facebook – deeply the government may well take that and not a for-profit company. If you think about how we started, it’s easy to communicate, it’s cheap to build apps, we may well have a high-dimensional construct of identity. Having new algorithms to show membership which does not allow us to tract down individuals.

Q: Then everyone would be a central bank.

AW: Virtual item trady, Thailand. Throttled internet connectivity. Too many Thai teenagers trading.

Q: Eighth design pattern – timeliness of interaction.