Monday, November 7, 2011

Filters: our Shared Responsibility

A final look, I promise, at "The Filter Bubble", by Eli Pariser. (Part 1, Part 2)


There's a huge question about whether unbiased content is really a meaningful thing to talk about. In the same way that Pariser skewers the myth of disintermediation, the idea that there's an unmediated set of search results for some search term, an unbiased newspaper front page, or a listing of TV shows that everyone in the world would agree is value neutral probably needs skewering too.

If you've every written a web app of any complexity, you'll know that getting everyone to agree on what should appear for the result of a search can be difficult, even for the simplest of things. Actually, especially for the simplest of things. When you factor in the different ways in which the list can be ordered, well, the idea that there is some neutral way that Bing (thought I'd give Google a break) can be expected to index and organize the world's information starts to look a little ... lacking in rigour. It's already well-filtered before it gets to you. It better be. That's the value of of a search engine. Looked at this way, any further filtering in the form of personalization that you apply to most web content is merely an extra lettuce leaf placed on top of a huge pile of lettuce leaves (I'm trying that phrase out, see how it goes).

Kickin' it old school

As an aside, there's one particular reason above all others I'm endeared to "The Filter Bubble": cameo appearances by great writers. To be honest, sometimes in books like this you can get a little bit lost in a 25-page chapter dealing with some subtle point about the problem of induction: good meaty stuff, very worthy, but if I'm tired or distracted I may just drift a wee bit. But someone who pays homage to our shared literary canon by bringing in some great author as regularly as Pariser does snaps me out of it and wins me over every time.

It's also an effective pedagogical technique, if quite a stretch, to introduce the likes of Dostoyevsky at the tail end of a critique of algorithmic prediction techniques. But you have to admit, in "Notes from Underground", the great man pretty much nailed it: "All human actions will be tabulated according to these laws, mathematically, like tables of logarithms up to 108,000 and entered in an index...". Clustered, hopefully. Also making guest appearances in "The Filter Bubble" are Asimov (not so strange I suppose), Kafka, and Nabokov. Perhaps inspired by such exalted company Pariser rises on occasion to some tasty penmanship himself: at one stage he opines that personalization offers "a return to a Ptolemaic universe in which the sun and everything else revolves around us." Bravo!

I love this: there are plenty of books written about technology which barely acknowledge writers and thinkers of the past. Why would they? All this new technology is so unprecedented, future-shocky and paradigm-shifty that how can anyone over the age of 30, let alone some dead old Russian guy, have anything to say about how people interact with machines?

Sometimes though, Pariser overstretches, but even then there's a good story to be had, at least. In a chapter called "The Adderall Society", we're given the strange story of Russian defector Yuri Nosenko and the mishandling of his case by the CIA for 6 pages, simply to make the point that people can get their view of the world distorted quite easily. It's interesting stuff, but ... what does this have to do with me moving my Yahoo News widgets around again?

The LEGO Turing Test

A place where personalized filtering would be a boon, and I would happily pay for it, is if, for example, YouTube could work out that the intelligence interfacing with it from a particular machine was a child. These days my kids wake up very early, charge downstairs and use my Motorola Xoom to search for Lego Star Wars videos. Love it. I'm happy they're using magic technology that I couldn't have dreamed of when I was their age to light up their pleasure cells like a Christmas tree. It makes them happy, and they're learning to use the tools their world is rapidly filling up with. But it'd be nice to think that YouTube will keep a protective, avuncular eye on the little tykes, and stop them from being served up some video nasties by accident.

At the risk of infantilizing adult aficionados of Lego Star Wars (although I would argue it's too late for that) it shouldn't be too hard to make a short-term inference that the person watching all these kids videos is a kid, and to tailor the site's content accordingly: "The Babysitter Filter".

So, what's the problem again?

Before I finish, it's worth going over again what exactly the problems are with personalization, in a nutshell. In the shallow end, "there's less room for the chance encounters that bring insight and learning." In the deep end, excessive personalization is no less than a threat to democracy itself, as people exclusively inhabit their own bubbles, rarely if ever exposed to differing opinions, counter-arguments or dissent, the sine qua non of an informed worldview.

If that all sounds too crude and obvious, Pariser predicts that in the example of China there will be a rise in second-order censorship: the manipulation of curation, context, and the flow of information and attention, all assisted by the filters we gratefully sign up to use. Direct censorship is sooo 1989.

I read all this, and I'm not sure what to think. By what law of the universe should Google, Amazon, Bing, or Yahoo search results show the exact same information to each and every one of us? Speaking to friends about it, I find they're often surprised to find out that results are personalized, but then we usually end up muddled about whether this is good or bad. There's definitely a whiff of a 'good old days' argument in this book, for instance when Pariser reminisces about the time "when Yahoo was king, and the online terrain felt like an unmapped continent", but at the same time he openly acknowledges that the web is no different to other media insofar as it's growing up fast. It's just that, as he says, it was supposed to be different. Was it really, though? And if it was, have we really lost it so soon?

You can always still go to Wikipedia's home page if you want a more-or-less randomised inventory of interesting historical and scientific facts and (what seems - but it would, wouldn't it - to be) neutral news stories. Same with Twitter. Just hang out on the front page and watch the world go by. And really, the people who are interested in the world will seek out a diversity of opinion, the same way they always have. Those that aren't won't. It's up to you to find out if your service of choice is likely to be preventing you from seeing the big wide world, same way it's up to you not to be too much of a dick online, and up to you not to get ripped off by phishers and scammers. There's no magic formula: it just helps to be skeptical when things seem too good to be true.

Personalization and Counter-Personalization

A look at "The Filter Bubble", by Eli Pariser, Part 2. (Part 1, Part 3)


A good old-fashioned positive feedback loop

One way to conceive of all this creeping personalization - the subject of "The Filter Bubble" - is to see it as a positive feedback loop. You search for something in Google, you click on a link in the search results, and that result is more likely to appear higher the next time, which makes you more likely to click it the next time because it's slightly more in your face. This feedback loop has been in place for a while with ads: the activities of the users influence whether the ads are shown more prominently - the more clicks, the higher the placement (all other things being equal) - and the more prominent the placement, the more clicks the ad will generally get. It's one of the factors that has made AdWords successful. But I think people see a big difference between ads and 'organic' search results. We've always been told that there was one.

Amazon is going even further with their drive towards relevance, which is another way of seeing personalization, by taking it into the once private realm of reading. According to Pariser, when you use a Kindle, "the phrases you highlight, the pages you turn, and whether you read straight through or skip around are all fed back into Amazon's servers and can be used to indicate what books you might like next." It's funny to think of reading like this, isn't it? Reading under a tree in the park at lunch time, away from the noise of the office. Reading on the toilet. Reading to your kids. Reading about Paris on your way there on the morning EuroStar.

Bookmarks matter

As an aside, speaking of the Kindle, I still find the whole bookmark in the cloud idea amazing and delighful. As I've started to synchronise my Kindle account across a couple of devices - iPhone, Motorola Xoom - I'm always charmed to see the 'Loading' icon as the device fetches my furthest read position from one device to another: my virtual bookmark. I can accept the transition of books from real to virtual entities - their value is really only in the words they contain after all - but for me bookmarks have always been real, physical, holdy objects. One of my favourites was from a trip I took one May Bank Holiday in 1988 to the London Dungeon, a pseudo-leather green one with a frayed end and a big scary skull on it. Recently Gav made me a great bookmark, a solid, bruising bookmark, like a sample of one of his recent montage paintings, which imposes itself on the book. Leaves are an excellent choice too. Bookmarks matter. But I'm adapting to the Kindleverse. My bookmarks are in the cloud now. I'm letting go.

A personalized take on Gmail

At times the book takes a cynical approach where others have been more generous. Here's how he sees Google coming up with Gmail: "The challenge was getting enough data to figure out what's personally relevant to each user... But how? In 2004 Google came up with an innovative strategy. It started providing other services, services that required users to log in. Gmail, its hugely popular email service, was one of the first to roll out... By getting people to log in, Google got its hands on an enormous pile of data."

This is very different, and a lot less inspiring, from the way Gmail is explained in Steven Levy's "In the Plex". There the creation myth of Gmail is one of a maverick programmer tackling the problem of email by seeing it as a search problem where no one else had. Certainly not his bosses at Mountain View, who wanted him to stop messing around with email, a business they were mostly definitely not in. Along the way the irksome problem of periodically having to clear large files our your inbox was solved by giving users almost infinite storage space. Before they were finished, they made a decent stab at eradicating spam too. "The Filter Bubble" and "In the Plex": two different books, two different takes on our beloved Gmail.

But the book's take on the what Pariser calls the myth of disintermediation is more on the money. The idea that the open, democratic web eliminated all the nasty middlemen who ran big newspapers and media outlets like TV and radio stations is entertainingly dismissed, along with web luminaries like Dave Winer and Esther Dyson, as naive. "Once upon a time, newspaper editors ... decided what we should think ... Then the internet came along and disintermediated the news ... The middleman dropped out." Pariser's analysis is typically sinister: "the effect of such naivety is to make the new mediators, such as Google News or CraigsList, invisible. But while we've raked the editors of the New York Times and the producers of CNN over the coals for the stories they've missed and the interests they've served, we've given very little scrutiny to the interests behind the new curators."

The Man behind The Man

The commoditisation of your click signals doesn't stop at the site you happen to be visiting. In many cases, cookies are sold in a secondary market to huge, faceless third-parties in service of the sinister-sounding business of "behavioral retargeting", whereby ads "follow you around the internet". They may even encroach on your in-flight entertainment, even though you may never have flown before. Though the tone of "The Filter Bubble" is one of relentless pessimism, if @QantasAirways wants to serve me ads about MVC3 books rather than Omega watches that's ok with me.

Personalization is based on a Faustian pact: sites give you a nice customised experience and you give them information about you. So don't come the innocent when Mephistopheles sells your cookies to the man. You always knew that would happen. You did.

Save us from ourselves

Probably the most interesting argument against excessive personalization that the book makes is that, left to our own recommendations, we tend to duck the difficult stories in favour of salacious, trashy ones. That's just how we are. But we need to know a certain amount about boring, complicated subjects like the causes of poverty or the war in Afghanistan: it's good for us all. Not in any spiritual, ill-defined way, though: the idea of a communal front page of a popular newspaper site is a community good, one where bad deeds can be exposed to the light. If everyone sees something different because editors pander to viewers' basest desires then that virtual town hall has been knocked down, and its place has been taken by a shopping centre, complete with KFC and AdultWorld.

Counter trend of conformity

As persuasive as this book is, there are still artefacts of web culture that are pushing against the prevalent personalization. On many websites you'll see a 'Most viewed' list of links, often grouped with 'Most emailed', acting as a sort of communal aggregation of information, the opposite of personalized content. These are articles that a lot of people clicked on, so maybe you should too. I'm noticing a similar communality being brought about by the e-reading experience. I've been reading a polemical ebook on the Kindle that has attracted a regular outburst of communal sharing every 5th page or so. At the culmination of some important point, a sentence will be underlined, indicating it has been shared by at least 10 people. This crowdsourced highlighting effect instills in one a powerful urge on the bus home to accept the mob's judgement and join in their sharing of this sentence with the web. After all, this must be a very important sentence indeed if so many other people thought so.

There are plenty of other counter-personalized examples you can think of. You're probably more likely to follow someone on Twitter who has a lot of followers. For a page of YouTube search results, you're highly likely to watch one that has a high view count. These are powerful inbuilt centripetal correctives to the centrifugal pressure of individualized content, pulling us back into the centre, back into the bosom of the crowd, back where we belong.

(Continued in my next post)