Shirky’s Law and why (most) social software fails

Shirky’s Law states that the social software most likely to succeed has “a brutally simple mental model … that’s shared by all users”.

If you use social software like Flickr or Digg, you know what this means. You can give friends a simple and compelling explanation of these sites in seconds: “it’s a website that lets you upload photos so your friends can also see them”; “it’s a community website that lets you suggest interesting sites; the users vote on submissions to determine what’s most interesting”. Of course, for each Flickr or Digg there are hundreds of failed social sites. The great majority either fail to obey Shirky’s Law, or else are knockoffs that do little not already done by an existing site.

To understand why Shirky’s Law is important, let’s look at a site where it’s violated. The site is Nature Network, one of the dozens of social networking sites aspiring to be “Facebook for scientists”. Like other social networks, Nature Network lets you connect to other users. When you make a connection, you’re asked whether you would like to connect as a “friend” or a “colleague”. Sometimes the choice is easy. But sometimes it’s not so easy. Furthermore, if someone else connects to you, you’re automatically asked to connect to them, but given no immediate clue whether they connected as a friend or as a colleague. The only thing shared in the users’ mental model at this point is acute awkwardness, and possibly a desire to never connect to anyone on Nature Network again.

I don’t mean to pick on Nature Network. It’s the most useful of the social networks for scientists. But it and most other social websites (apart from the knockoffs) don’t even come close to obeying Shirky’s Law.

Why is Shirky’s Law so hard for developers to obey? I’ll give three reasons.

The first reason is that developers often have a flawed mental model of their own software. Imagine you’re developing social software. You spend hundreds or thousands of hours on the task. In your mind’s eye, you imagine the user interacting with your software, and reason that if the user is given more capabilities, they’ll be happier.

There’s an implicit mental model being used to make decisions here. It’s a mental model of a system with two parts – the software and the user. A real user’s mental model is quite different. It’s them, the software, and the entire network of other users. How they use the software is strongly conditioned on their mental model of how other users use it. If they lack confidence in that mental model, they have less incentive to use the software themselves, as the Nature Network example shows. The more social the software, the stronger this effect.

Most developers are not stupid, and intellectually they know the user experience involves both the software and the network of other users. But their own experience, day in and day out, is of being a single user working with the software. At this stage the network of other users is a theoretical abstraction. It’s easy to get sucked into doing things that would make a single user’s experience better, but makes the experience of a network of users worse. This is a large part of why it’s so important to build a base of beta users as quickly as possible, and to release early and often.

A second reason developers fail to obey Shirky’s Law is the desire to do impressive-seeming things. Software that looks complex is much more impressive to other hackers, venture capitalists, and non-hacker friends and family.

I’ve heard hackers brag that they could have built Twitter over a weekend. Underlying this boast is a misunderstanding of what is truly impressive. Coming up with Twitter required only a small amount of technical knowledge. The hard part was the social insight to realize such a tool would be useful. This is a social insight the bragging hackers didn’t have.

It’s no accident that many of the people who’ve been most successful at building social software have strong interests outside computing. Mark Zuckerberg, the founder of Facebook, studied both computer science and psychology at Harvard. Alan Kay, arguably the father of modern computing, has a list of recommended reading. There’s barely a technical book on it. It’s all psychology, anthropology, philosophy, and so on.

A strange consequence of all this is that much of the most successful social software was invented by accident.

Ward Cunningham invented the first wiki because he was tired of responding to user’s requests to update a website he ran. To save himself time, he made the page editable, and told them to update it themselves. He was shocked when this small change utterly transformed the dynamics of the site.

One of the first widely used pieces of blogging software, Blogger, was originally a small part of a much more ambitious project management system. The project management system never caught on, but Blogger took off.

The team that developed Flickr wasn’t originally building a photo sharing service. They were building a multiplayer online game, and decided to let players share photos with one another. When they realized the players were more interested in sharing photos than playing the game, they dumped the game, and built Flickr.

Shirky’s Law does not mean the software itself needs to be simple. Social software like Digg and FriendFeed uses complex algorithms to rank the relative importance of submitted items. But the complex parts of the software are hidden from the user, and so do not add to the complexity of the users’ mental models.

There are some apparent exceptions to Shirky’s Law. For example, Facebook is now a successful and complex piece of social software. But in the early days, Facebook was extremely simple, and this simplicity fueled their rapid growth: “it’s a site where you can connect to your friends, and show them what you’re up to”. Complexity can only come later, when users are already confident in their shared understanding.

The third reason developers fail to obey Shirky’s Law is that it’s difficult to do. The most successful social software starts out doing one task supremely well. That task is simple, useful, and original. It’s easy to come up with a task which is useful and original – just combine existing ideas in a new way, perhaps with some minor twists. But finding something that’s also simple is hard. It has to be a single task that can’t be reduced or explained in terms of existing tasks. Inventing or discovering such a task requires either a lot of hard work and social insight, or a great deal of luck. It’s no wonder most social software fails.

Further reading

This essay is adapted from a book I’m writing about The Future of Science. If you’d like to be notified when the book is available, please send a blank email to the.future.of.science@gmail.com with the subject “subscribe book”. I’ll email you to let you know in advance of publication. I will not use your email address for any other purpose!

Subscribe to my blog here.

Subscribe to my FriendFeed here.

Acknowledgments

Thanks to Jen Dodd for feedback that improved this essay.

32 comments

  1. I must say, this is a very well thought-out essay, and it goes a long way towards explain why some sites succeed and others fail. One consequence of all this, though, is that sometimes I find social services lacking subtlety that would disturb that shared understanding. For instance, marking people as “friends” is often grossly inaccurate– they may only be acquaintances or people that you are socially pressured to mark as friends.

    Since there’s no (obvious) way to work such subtleties into shared understanding, it’s not surprising that I have yet to find a site that successfully deals with them. As a result, it isn’t too uncommon for friend-network metadata to become confusing and contradictory. Of course, it’s not likely that hard to find other examples of unintended oversimplification resulting from Shirky’s Law.

  2. Agrrggh – self referential crisis, do I comment here or FriendFeed? Anyway had a very interesting conversation last night with some web geek types trying to explain why Friendfeed was good. They didn’t get it at all and preferred Twitter. But I think part of the reason it has been successful for ‘our’ community (whatever that is) is exactly this issues of a partly shared mental model. Part of the reason I think it is now becoming less useful (perhaps its good the comment is here) is that for this community that shared model is starting to break down.

    There is a side issue here about perceptions of what people are seeing which I think is a serious one. I personally don’t tend to consider enough the fact that someone may be seeing what I am seeing in a very different context (e.g. through twitter rather than friendfeed). Do we need a service that shares our view of other services so people can understand our viewpoint? Or does it not matter because our mental models will adjust to deal with this relativism? And should I go back to Uni and do a degree in philosophy so I can understand this better?

  3. I guess then scientific social networks should identify those things that would be their core feature that should be done extremely well. Examples: suggesting reading material, facilitating scientific collaborations (find me the person closest to my network working on X).

    @Cameron – I am not sure I understand your point. In general in real-life conversations, in FF and in the loosely coupled blog discussions, we need to assume some amount of shared context. We draw a cut-off and explain in detail from there. This notion of shared mental models was already a big part of blog discussions, FriendFeed makes it more explicit because I can assume that you might read something that you mark with “like” even if you don’t comment.
    One possible limitation of FF will be when each person finds in FF all the different types of people that share the different facets of this person. This means that with increasing number of users, each individual user will tend to broadcast/like/etc on a broader set of topics. Since we cannot subscribe to a “facet” of a person , only to the different life streams, this might increase the noise with time. I think the Rooms are meant to be used for this, to isolate the discussions by topics.

  4. Hi Michael, I’m the product designer behind Nature Network and have a psychology background too. Thanks for your kind words about the utility of the site. Clay’s suggestion that sites be brutally simple is a good one, but within science there is the complicating factor of competitiveness between researchers, especially in biological sciences. Organisational scale also affects the kind of product that gets made, the three examples you gave all came from small startups. Facebook and Myspace are not small simple products either, but ther is work to be done for NN to simplify the experience I agree.

    The addition of friend / contact serves two purposes. it will allow a means to filter the content and may support a layer of private and public content. It is something we are exploring currently. How can you share some information without telling everyone on the site something.

    Flickr and delicious are more simple cases, there is no competition over sharing photographs, but there is a privacy issue. The Pownce approach has a lot of merit to it in terms of semi private information sharing

    @pedro I like jaiku for this reason, you can turn on and off facets of the content a person is sharing.

  5. Was with you, right up to that twitter remark and then my misgivings continue from there. Essentially I have two issues with what you say:

    1) By what standards would you call twitter a success? If you judge it on traction, reputation, brand value, real value, revenue or impact then it’s not doing especially well.

    2) Twitter, Flickr, Youtube all started out as completely different applications. Their usage was shaped by the community in ways not anticipated by the technical founders.

  6. Jof,

    On point 1: Twitter’s problems have mostly been caused because they have not scaled well. Shirky’s Law is only a necessary, not a sufficient condition, for success. You can have a great idea that obey’s Shirky’s Law, and then fail to execute. Friendster is another example that comes to mind, in a similar vein.

    On point 2: nothing in my essay says otherwise. In fact, I explicitly make this point.

  7. Chris – Thanks for your thoughtful comment. Yes, it’s tough to express the subtlety of human relationships without the sort of social cues we have in everyday life. Some sites are pretty good about this, though. The “like” button on FriendFeed is a beautiful little idea – it’s kind of like smiling at someone in appreciation after they’ve done something good.

  8. Hi Gavin,

    Thanks for the comment. It sounds to me like you’re mortgaging Nature Network users’ current experience against the uncertain (and, to most of your users, unknown) possibility of an improved future experience.

  9. Cameron – Yeah, resharing is funny. When I bookmark something on del.icio.us, it is seen by at least 4 different audiences: me, later; my del.icio.us network; people who read my blog; people who subscribe to my friendfeed. Not to mention the possibility if it being reshared by others, in ways I can’t control. It’s tough to write for all those groups simultaneously.

    “And should I go back to Uni and do a degree in philosophy so I can understand this better?”

    Hah! I often feel I need degrees in sociology, anthropogoloy, economics, computer science, and political science as well, these days. Interesting times.

  10. @Michael – again, I disagree. At their core Twitter’s problems are not technical – they are business-related and the reliability issues they’ve been having are a side-effect of that. Had they not been drinking the Kool-Aid they might have already turned this into a successful business with decent revenues and scale.

    On the second point, I may have mis-interpreted the tone of the piece; my bad.

    BTW, I also disagree about the complexity of Digg, etc. I doubt their algorithms ARE complex as complex algorithms often don’t scale. But I see your point though… only nit-picking. I’m probably just grumpy because after two weeks my iPhone STILL isn’t activated.

  11. I’m not sure how Friendster and Orkut work with this theory. These sites were to market before Facebook, I don’t remember them failing to execute in any dramatic way, and they fell by the wayside. I think they obeyed Shirky’s Law as well as Facebook.

  12. Jof,

    I think we both agree Twitter is having trouble with execution. I’m not close enough to the problem to know to what extent this is a techical problem, and to what extent it’s a business problem (or both). If you think it’s a business problem, I won’t disagree.

    The salient point for the essay is that obeying Shirky’s Law is a necessary but not sufficient condition for success. Twitter is unusual in that they’ve built a useful and novel product that obeys Shirky’s Law, and their troubles are being caused by a failure to execute.

    On Digg etc: I’d like to improve these examples, for the reasons you describe. The ranking algorithms they use don’t appear that sophisticated. It’s possible that the search functions are more complex, but I haven’t explored enough to know. What I really had in mind with this comment was Google, whose simple search box hides extraordinary complexity. I didn’t use this example because Google isn’t exactly social software. Nonetheless, the general point holds: it’s possible to build very complex software that obeys Shirky’s Law.

  13. Michael
    Every product grows and changes, even simple ones. On Twitter hashtags came and went; @replies came and were incorporated; the with others page has vanished; it has gained an API. Flickr has grown and changed too, recently it has gained a lot of geo features.

    What we are doing with Nature Network is no different, but we are working with a community who are less willing to openly share content online. this gives us different problems. You cannot transplant one way of working into a different domain. A flickr for selling houses would fail, as there is competition amongst the viewers of the photographs.

    Having a brutally simple idea is a great place to start and will help you to hold course as your community demands their own special set of features. Learning to say no is important as you continue to develop your application.

  14. Human social networks are almost infinitely protean … a glimpse into a strikingly successful global science/technology oriented network is provided by Boeing’s 787 Technical Progress Website.

    Boeing’s global social network contrasts with Facebook mainly in its focus upon a single, shared goal—delivering the eight hundred 787 Dreamliners that have been advance-ordered.

    Obviously, Boeing’s network is founded upon shared mathematics, science, and technology. It is striking that almost all of this technology is classical.

    Could a similarly large, global network be founded upon quantum science and technology? There is a fundamental reason to think that the answer is “yes”. Namely, the promises that quantum physics makes about quantum systems are (mathematically) much stronger than the promises that classical physics makes about classical systems.

    It is therefore reasonable (IMHO) to foresee that the design tools of quantum system engineering will evolve—in fact, are already evolving—to become much simpler and much more powerful than the design tools of classical engineering … and that these tools will support global enterprises much larger than the 787 Dreamliner enterprise.

    Ever since Deak Parsons write his famous memo “Home Stretch Measures” to Oppenheimer in 1945, system engineering considerations have exerted an increasing powerful influence upon social organization in science.

    There are good reasons to foresee that this increasing influence will become even more pronounced, as the tools of quantum system engineering become even more capable.

  15. Great essay! I think you hit the nail squarely on the head.

    At the end of the day, the success of social software is judged by the number of users it attracts. Twitter has a variety of limitations and technical issues, but also a lot of users. Why? Because of (a) the way it allows you to interact was unlike anything else widely used and (b) a large, growing community of frequent users.

    Shirky was right, and so are you. It was a very astute observation that software developers do not think from the community perspective. Maybe you should send this post for inclusion in some of their trade publications.

  16. Great post, Michael. It’s just appeared, word for word, here >

    http://www.simpledog.net/blog/

    without any acknowledgement whatsoever. I’ve posted a comment there, (which is awaiting moderation), that the least they could do would be to acknowledge your original post.

Comments are closed.