Home , Hustling, Landing page, Spam blog

Spam

Finn Brunton

THE COUNTERHISTORY OF THE WEB what the rules are, and who’s in charge. And the first conversation, over and over again: This chapter builds on a larger argument what exactly is ‘spam?’. Briefly looking at about the history of the Internet, and makes how this question got answered will bring us the case that this argument has something to the Web and what made it different. useful to say about the Web; and, likewise, Before the Web, before the formalization that the Web has something useful to say of the Internet, before Minitel and Prestel about the argument, expressing an aspect of and America Online, there were graduate stu- what is distinctive about the Web as a tech- dents in basements, typing on terminals that nology. The larger argument is this: spam connected to remote machines somewhere provides another history of the Internet, a out in the night (the night because comput- shadow history. In fact, following the history ers, of course, were for big, expensive, labor- of ‘spam’, in all its different meanings and intensive projects during the day – if you, a across different networks and platforms student, could get an account for access at all (ARPANET and Usenet, the Internet, email, it was probably for the 3 a.m. slot). Students the Web, user-generated content, comments, wrote programs, created games, traded mes- search engines, texting, and so on), lets us sages, and played pranks and tricks on each tell the history of the Internet itself entirely other. Being nerds of the sort that would stay through what its architects and inhabitants up overnight to get a few hours of computer sought to exclude. Identifying and describing access, they shared a love of things like sci- spam, from the terminals of time-shared ence fiction and the absurd comedy of Monty mainframes in the 1970s to the elaborate Python’s Flying Circus. Alone at the termi- automated filtering systems of Gmail, meant nals, together on the network, they would vol- having to talk about what the network is for, ley lines from Python sketches back and forth SPAM 565

– the dead parrot sketch, the dirty fork sketch, – for instance (Larson, 1994).) The lottery the spam sketch. This last was particularly was an initiative to simplify and speed up the popular because most of the dialogue was process of getting papers to live and work in just the repetition of ‘spam’, whether sung by the United States as a foreign national: a sub- Vikings or shouted by the waitress, and it was ject of obviously narrow interest, which they therefore trivial to generate. You could write had broadcast to computers from Singapore a simple program that, at the right spot in the to Australia to the Netherlands – and, of dialogue, would post ‘SPAM! SPAM! SPAM! course, throughout the United States, where SPAM! SPAM! SPAM! SPAM! SPAM!’ over the vast majority of recipients were citizens and over, relentlessly and without pause, fill- already. To enter the lottery, furthermore, ing the screen, killing the discussion, and only required sending in a postcard, but often overloading the chat platform com- the message suggested that paid legal help pletely, kicking people offline. Jussi Parikka would be needed – a sleazy commercial and Tony Sampson (2009), in the context of misrepresentation. Abuse of the network’s their larger analysis of spam, have shown that many-to-many tools and global reach the sketch itself is built around a communi- had been combined with a moneymaking cations breakdown: the point where noise on scheme. the line overwhelms any particular signal. It In the process of trying to describe this was annoying, but playful and mischievous event, the global community on Usenet set- rather than malign, like unexpectedly blow- tled on spam as the term of art for their mes- ing a vuvuzela in the middle of a conversa- sage and their action: the American lawyers tion. This kind of noisy, frustrating behavior had spammed the network. The word had was dubbed ‘spamming’. jumped into the domain in which we identify The term came in useful in the ensuing today … or rather, it had come closer to our decade-plus – though not with reference to current understanding, in a way that is inti- advertising or commercial messages, which mately intertwined with the development of were their own category of etiquette viola- the Web. tion. ‘Spamming’ remained the domain of After assembling the whole history of noise and the indiscriminate, wasting time, spam, from an anti-Vietnam War message dis- attention, and bandwidth on redundant copies tributed on MIT’s early time-sharing system of messages, on overly verbose and off-topic in 1971, to a Digital Equipment Corporation postings, on tedious rants and cut-and-pasted ad on ARPANET in 1977, to present-day slabs of text. Dave Hayes, prominent on the phishing, comment spam, ‘spammy’ posts discussion system Usenet, wrote a mani- and social network activity, clickbait and festo in 1997 itemizing the forms of social 419 (‘Nigerian price’) emails, I arrived at a misbehavior online – with ‘commercial self- working definition. What is ‘spam’? Spam is promotion’ as a separate entry from ‘SPAM’, the manipulation of information technology which meant precisely being a high-noise infrastructure to exploit existing aggrega- low-signal attention hog over a precious tions of human attention. That is the meaning and expensive medium (Hayes, 1996). This of ‘spam’ once all the technological particu- definition shifted irrevocably in the spring of lars of search engine spamming or phishing 1994, when two lawyers from Arizona posted campaigns have been worn away: follow- a message across Usenet – that is, to com- ing the term’s broad application across the puters around the world, to many thousands decades – both in English and as a loanword of users, indiscriminately – offering their in other languages on the Internet – includes services with the process of entering the US commercial and noncommercial activities, Green Card lottery. (We know this event best criminal and legitimate, with many different through the responses that quote it at the time technologies and platforms, from email to 566 THE SAGE HANDBOOK OF WEB HISTORY

Twitter to content production. What remains with their lame message, treating the whole consistent, I argue, is the model: spam- network as a passive audience whose time mers identify already existing collections of was theirs to spend. (Some of the complex human attention, and imitate and manipulate distinctions inherent in spam, ‘trash’, ‘junk’, their particular properties to extract value. I and ‘waste’ are considered in the analysis want to say a few more words explaining this collected in Parikka and Sampson (2009), before focusing on the Web in particular. especially Galloway and Thacker’s ‘On Spam is an information technology phe- Narcolepsy’ (2008), and Gansing (2011).) nomenon. Across their many modes and Two consequences follow from studying domains, spammers push the properties of spam in this light. First, we can see the his- information technology to their extremes: tory of networked computing as a thread in the capacity for automation, algorithmic the history of the management and distri- manipulation, and scripting; the leveraging of bution of attention – Alessandro Ludovico network effects and vast economies of scale; (2005) has argued that spam is best seen as distributed connectivity and free or very one instance in a long history, from traveling low-cost participation. So many neglected salesmen to personalized bulk postal mail to blogs and wikis and other social spaces are eye-catching billboards, of trying to interfere out there on the Web: automatic bot-posted with our thoughts and provoke us into some spam comments, one after another, will fill form of consumer desire. Second, we can see the limits of their server space. What this in concrete terms how the nebulous shape of means – beyond characterizing spam as an community online used spam to define itself. activity – is that spammers take advantage The intersection of these two topics brings us of existing infrastructure in ways that make back to the distinctive history of the Web, the it difficult to extirpate them without making ways it gathered attention and formed com- changes for which we would pay a high price. munities, shown to us anew through spam’s Indeed, in Geert Lovink’s argument (2005), crass, hustling, inventive counterhistory. I spam is akin to other network failures like will break this counterhistory up into four identity theft in being inherent in the design – sections, which describe from spam’s side constitutional elements of yesterday’s net- how the Web became searchable, central, and work architecture. Spammers partially sur- social. vive by finding places where the potential value lost and effort expended in locking- down could exceed the harm they do – which reveals those places, and their value, to us. JUNK RESULTS More exactly, spammers find places where the open and exploratory infrastructure of the How, though, could spam have come to play network hosts gatherings of humans, how- a role in the Web? My brief sketch of spam’s ever indirectly, and where their attention is origins, above, is all social spaces: conversa- pooled. The use they make of this attention tion threads on Usenet, chat on time-sharing is exploitative not because they extract some computer networks, and of course email, value from it but because in doing so they where ‘spam’ as a concept and a business devalue it for everyone else – that is, in plain scaled up. The Web, though, was a kind of language, they waste our time for their ben- document navigation system at first – a efit. Recall the objections against the Green markup language and set of protocols for Card spam campaign on Usenet: it wasn’t authoring and exploring knowledge through simply that the lawyers were acting com- hypertext. It was a project suited to a poly- mercially but that they didn’t respect sali- glot scientific community: developed by an ence, barraging everyone indiscriminately English computer scientist, revised by a SPAM 567

Belgian, with the first site built in French, Brin and Page put it, that ‘some advertisers hosted on a Californian machine in the walls attempt to gain people’s attention by taking of a Swiss research institute, a knowledge measures meant to mislead automated search presentation and navigation tool for one of engines. … “Junk results” often wash out any the twentieth century’s biggest capital-S results that a user is interested in’ (2). Or, as Scientific communities. It’s a context, and a a paper published on the very same day more technology, in which you can no more imag- bluntly put it: ‘Some authors have an inter- ine spam thriving than you can imagine mold est in their page rating well for a great many growing on a space station. types of query indeed – spamming has come But mold does in fact grow in outer space, to the web’ (Pringle et al., 1998: 1). behind panels, on gaskets and insulation, The form that spamming took reflects the on walls, under clothes. As human atten- unique particulars of the Web and search tion condensed, collected, and pooled on the technology: it was designed not simply to Web, from Erwise to ViolaWWW to NCSA dominate a conversation, as in chat, or flood Mosaic, techniques began appearing to a channel with messages, as with email and absorb and exploit it. Take a year, from the Usenet, but to make assertions of relevance middle of 1994 to 1995, when we can see and salience. We can see, through spam’s many different factors picking up speed: the development, how generations of search global growth of users away from computer engines tried to model information on the science professionals to the general popula- Web in terms of what it meant for a user’s tion; the end of the noncommercial restric- query. The spiders that the first search engines tions on the network in the United States; the sent out would go through the HTML source shift in power from sysadmins to lawyers and of a page, using the structure of the markup to entrepreneurs as social arbiters; and events assess the significance of words with greater like the Green Card lottery spam on Usenet or lesser degrees of importance and relevance (with subsequent publicity in newspapers to a search. A word in a URL (for uniform – the first appearance of ‘spam’ in print – resource locator, the ‘address’ of the page) or and major advertisers scenting blood in the in the first header tag – which is the markup for water); Mosaic’s booming download num- what the human reader would see as the ‘title’ bers; and the publication of How to Make a of the page, as in

My Homepage

– Fortune on the Information Superhighway – a was probably more important than one in cash-in book by those same Arizona lawyers, the body text of a page and would be rated promising to teach readers how to market accordingly in the index. A set of elements across the global network and get rich quick called ‘meta tags’ were used in HTML spe- by exploiting the technology. cifically for the benefit of search engine spi- There were 20-odd websites in the fall ders, with keywords listed for the page such of 1992, 10,000 by the end of summer in that they would be invisible to the human 1995, and millions by mid 1998. There were reader but helpful to search indexing. Helpful so many sites by then that finding what you in theory, anyway: though meta tag elements were looking for – even knowing what was were popularized by early search engines available to be found – was an enormous such as AltaVista and Infoseek, they were so challenge, eloquently described in a paper aggressively adopted by spammers that meta- published on April Fool’s Day of 1998: ‘The data was largely ignored by the turn of the Anatomy of a Large-Scale Hypertextual Web century, with AltaVista abandoning the influ- Search Engine’, by Sergey Brin and Larry ence of meta tags on search results in 2002: Page. Others had already tried to solve the ‘the high incidence of keyword repetition and problem of abundance on the Web by devel- spam made it an unreliable indication of site oping search engines. The issue was, as content and quality’ (Sullivan, 2002). 568 THE SAGE HANDBOOK OF WEB HISTORY

What precisely was the business plan of Web browser from other platforms, like a these early search spammers, and what were phone, or a Braille display, making it possi- they putting in their web pages? Keywords ble to serve a compact page to the phone and were repeated in the meta tags and gathered text instead of images to the Braille device. in the page itself, hidden from the casual This means you can serve one page to a spi- human reader’s eye. One of the details that der, to be indexed and delivered as a search HTML can specify is the color of text, so result, and an entirely different page to the the page’s author could set the page’s back- user who clicks on the link. The signatures of ground to gray and make text the same shade spider requests, which trigger the cloak page, of gray, invisible on the human reader’s dis- proved very difficult to disguise from spam- play while appearing to be normal text on mers – which brings us back to Google’s the page as far as the spider was concerned. embryonic form in 1998: ‘a prototype of a Innocuous pages with some form of spammy large-scale search engine which makes heavy intent would have a mysterious gap at the use of the structure present in hypertext’, cre- bottom of the page. (Such techniques could ated precisely to solve the problem posed by also be used playfully or prankishly, part of ‘junk results’, spammy web pages – and cre- the toolkit of the ‘vernacular web’ of bespoke ating in turn a new way for spam to reshape HTML (Lialina, 2009).) The text on the page the Web (Brin and Page, 1998: 1). ended and there were no images, just a few inches of the gray background before the bottom. In that gap, in background-matching color and often minuscule font size, lay a MUTUAL ADMIRATION SOCIETY magma flow of obscenity and pornography, product names, pop stars, distinctive phrases, ‘The citation (link) graph of the web is an cities and countries, odd terms seemingly important resource that has largely gone plucked from Tristan Tzara’s hat, selected unused in existing web search engines’, because they happened to get good returns at wrote Brin and Page. ‘These maps that time. The text reads as though a Céline allow rapid calculation of a web page’s character worked for Entertainment Tonight: “PageRank,” an objective measure of its toyota ireland ladyboy microsoft windows citation importance that corresponds well hentai pulp fiction slut nirvana. with people’s subjective idea of importance’ Such blocks of text illustrate a recurring (3). Inspired by academic citation structure, theme in the development of spam on the Web they argued for reputation, essentially treat- and elsewhere: a matter-of-fact distinction ing links as a measurable expression of between humans and machines, with differ- social value. They were not the first to do ent strategies for dealing with each. Almost this – earlier search engine projects, trying every piece of spam, whether over email or in to beat the keyword-stuffing of the Web’s the context of spam blogs or comment spam, first spammers, had tried to use numbers of became biface, capable of being read in two links to roughly evaluate the meaningful- ways with very different messages for the ness of results. In response, spammers had algorithm and for the human. (We will return started link farms, pages of nothing but links to this distinction and its consequences for between spam sites, providing a cheap-and- the Web at the end.) Spamming the early Web easy boost to that metric. Part of Google’s exacerbated this process, with techniques like brilliance lay in the flaw in this strategy: ‘cloaking’. Search engine spiders identify spam pages are lonely. They may link to themselves by the way in which they request thousands of other sites, but the only a page. This identification is part of the set inbound links, as a rule, come from other of protocols that help to distinguish a normal spam pages. SPAM 569

Links, in theory, carry an implicit endorse- among ordinary users of the Web: a comment, a vote of relevance made by a person. ment in a blog post included the comment- The spam-fighting question is: who is the ers’ websites along with their names, to rack person, and how much does their endorse- up another link. Posting something without ment count for? Google’s PageRank equation including a ‘via’ link to the person you got it answered those questions with the behavior from – the ‘via’ being an additional outbound of the ‘random surfer’, an abstracted user of link as a kind of thanks for using their dis- the late-1990s Web. This rather depressing covery – became increasingly rude, the sign model of a person starts on ‘a web page at of an uncouth person. ‘Mutual admiration random and keeps clicking on links, never societies’ arose, huge link-heavy sets of sites, hitting “back,” but eventually gets bored and each page linking to many of the others – starts on another random page’ (4). The like- all sites kept carefully unspammy, maintain- lihood that this idle character, clicking ever ing the pretense of legitimate use of the Web. forward along the link graph, will land on Their business was not to produce spam sites a given page defines PageRank. This means themselves, but to charge for outbound links that other sites linking to your site matters from the society. They were renting out their – as does which sites link to those that link accumulated ‘votes’. But even those had a to you. It’s a reputational model that works characteristic shape: heavy cross-linking transitively, with links weighted differently within a group of sites, all with only a few by their significance: ‘pages that have per- inbound links (because spam pages are lone- haps only one citation from something like some), creating little islands of intense self- the Yahoo! homepage are also generally endorsement with no outside involvement. worth looking at’ (Yahoo! being, at the time, To analytic tools, it’s a pattern as obvious as a directory of human-curated significant the newspaper ads taken out by vanity pub- links.) What were spammers to do? Building lishing houses for their new releases with the on the algorithmic inference of social data, blurbs from friends and family – and easy for Google could make it ‘nearly impossible to Google to discount accordingly. deliberately mislead the system’. The only In 1999, a company called Pyra Labs workaround for spammers would be to build launched a service called Blogger. (Google their own artificial societies. would buy it in 2003.) Blogger’s goal, as A variety of strategies developed as of so many related systems, from Flickr to Google’s market share grew and other search Wikipedia, was to provide people with an engines around the world developed similar intuitive means for publishing their content models. Websites with a high PageRank were on the Web. It was remotely hosted, so you transformed into kingmakers. A link from did not have to own a website domain name them could move a site onto the first page or or pay for hosting; many of its processes top three returns of the different search sites, were automated, so you did not have to boosting attention and revenue. Sites took design it or do any coding behind the scenes; advantage of preexisting ideas for the human- and it had a useful and increasingly sophis- curated Web, like ‘Best of the Web’ awards, ticated Application Programming Interface ‘Top 100 Sites’ awards, and so forth; these (API) for connecting with other Web appli- awards included a badge, a little image, and cations and automating processes. With the a snippet of code to be copied into the win- boom in weblog popularity and the peculiar ning site – a snippet that included a link to the chronological publishing model of blogs, award-giving site. The human user saw a lit- came another three-letter acronym, RSS, tle badge image, but the search engine spider ‘Really Simple Syndication’, which makes saw an outgoing link: a digital endorsement. new posts or other changes on a site available New habits of use and etiquette appeared in forms that are easy to use. (For the sake 570 THE SAGE HANDBOOK OF WEB HISTORY of Web history completeness: RSS originally search engine spiders. The splogs only work stood for ‘RDF Site Summary’, which high- from a distance, appearing to be groups of lights its relationship with the history of Web people, the language and links functioning document formatting – but was retroactively in aggregate. Taken in statistical total and changed to the more straightforward mean- algorithmic analysis, splogs resemble the ing.) Feed readers can gather the latest entries patterns of a thriving community. Their posts from RSS-enabled sites (which blogs soon are pitched at precisely the level of complex- were by default), material can be forwarded ity the spider requires to accept their input to mobile devices, and a page can feature the as human, and they adapt human text for headlines or recent posts from other sites. other machines to read and act on; affecting What does this have to do with Web spam? humans happens only indirectly – boosting Consider the toolkit laid out by these devel- the search ranking of a spammy appliance opments: a content publishing system that review site, for instance, that makes money can be easily automated (new accounts, posts through ads and affiliate links, or leading a on a prearranged schedule, modified set- human searcher into a fraudulent destina- tings), without the detailed work and paper tion, whether a simple rip-off with ads and trail of registering domain names and paying no meaningful content, or a site that might for Web hosting, and – with RSS – a faucet of middleman a transaction, tacking on an addi- other people’s words, content that looked real tional fee or trying to force-download some and human because it actually was, unlike a adware. lot of spam production. Hooked together with This section opened with ‘Google’, a the right software tools, you can generate a new-minted concept for a ranking algorithm, new kind of mutual admiration and endorse- and ends with Google, a massively success- ment society with a network of spam blogs ful advertising company that runs a search – or ‘splogs’. engine. One question raised by this chroni- A splog production system will pull in RSS cle of spam’s relationship to the Web and feeds from other blogs and news sources, chop Web search is how complicit, or symbiotic, them up and remix them, insert relevant links, Google is with its own antagonist. Consider and post the resulting material, hour after splogs: built and hosted on a platform Google hour and day after day, with minimal human owns, using text for content drawn from supervision. You can turn the machine on and other sites hosted by Google, optimized to leave the room while it makes money for you. best fit Google’s search engine algorithms, With contextual advertising (including ads as to boost the results for Google searches for a launching point for browser malware) you sites that make money by hosting ads served can make money through pageviews and the through Google’s affiliate advertising pro- occasional click by running ‘excerpt model’ gram (which, of course, also makes money splogs, with fragments taken from other peo- for Google). Search engine spammers run- ple’s posts that are polling particularly well in ning their vast stables of spam blogs and sites Google’s keyword metrics. A more ambitious are not anomalous, quantitatively or quali- system is ‘full content’ splogs, cross-linking tatively; splogs now account for more than in their hundreds and thousands to distort the half of the total number of all blogs (Fetterly shape of the Web. Each splog is assigned a et al., 2004). They are the optimal users, set of keywords and feeds from which to pull from Google’s perspective, constructing a related text, and in turn links to other splogs, system in which all the extraneous matter which link to still more, forming an insular of people and conversation has been pruned community on a huge range of sites – a kind away in favor of the automation of content of PageRank greenhouse that is not in itself production, search results, clicks, and ads meant to be read by people, but solely by served. This system in turn puts Google in the SPAM 571 contradictory position of having to analyze schedule an appointment with the lawyers, and expel many of their most dedicated cus- just as, with much of the spam in the years tomers: those who overexploit, and acciden- following 1994, you could actually purchase tally overexpose, the financial and attention the quack weight-loss pills, the deadstock economies and technologies that underlie the toys, the counterfeit watches. Spam – spam contemporary Web. Google is hardly alone in of this era and this meaning – was loathed this problem, as we will see. and despised, but it was also still somewhat legitimate, if only by accident. It thrived in the regulatory shadow of direct mail market- ing, a powerful and moneyed interest that THE LANDING PAGE didn’t want a legal precedent set that could close off a future advertising venue, and in Another shift in the dynamics of spam was the novelty of the increasingly popular and developing, meanwhile, which reflects the commercial Web, growing faster than legis- Web’s role as what Christian Sandvig (2013) lation and defensive software could keep up. calls an ‘emerging essential’. It had been International guides to legal redress for spam- something that ran on infrastructure – ming sprang up, their constantly updated con- running on top of the Internet, which ran in turn fusion of potential laws – from CAN-SPAM on top of the infrastructure of telephony – in the United States to economic crime units which became infrastructure itself, in the in Norway to Canada’s Department of Justice negative sense defined by Paul Edwards task force on pyramid schemes – highlighting (2003: 187), ‘those systems without which the problem of figuring out where crimi- contemporary societies cannot function’. The nal lines were crossed (for example, Hollis, Web became a key component for banking, 2005). Many spammers were able to present health care, work and administration, and themselves as brashly inventive promoters, content creation and consumption, with with postal addresses and registered trade- browser standards and shared protocols marks, seeking recognition in the classic tra- becoming matters of urgent negotiation, dition of entrepreneurial hustlers. What they monopolistic strategy, and even public safety produced is still what many people think of (as in the push to standards like HTTPS) – when they think of spam: the enthusiastic with international implications from the top- pitches full of mangled grammar and implau- level domains issued within the United States sible stock photography, in the service of a to legal decisions on hate speech in the EU recognizable, even traditional, class of dubi- (Goldsmith and Wu, 2008). In other words, it ous pleasures from timeshares and self-help was not simply an aggregation of human books to diets and pornography. attention around documents and content, But as the Web became an infrastructural navigated through search and hyperlinks, but system – and spam, for reasons too complex a portal into many vulnerable and intimate to go into here, became a progressively more parts of our working and personal lives. To embattled industry with less easy money – a understand how spam shifted accordingly, it new predatory spam technique took shape, helps to briefly look back for the last time to to trick humans rather than machines. As far that Green Card lottery message in 1994. back as the mid 1990s, hackers and spam- What the lawyer-spammers were offering mers alike had been finding ways to fool was, technically, an actual service (a mislead- people into giving up their login informa- ing, borderline fraudulent one, but let that tion. Initially, the goal was to send spam pass), whereas much of the spam we receive from trustworthy-looking accounts, or now has a very different agenda. You could within closed networks (whose members call the real, working telephone number and were often more naïve and easier to exploit). 572 THE SAGE HANDBOOK OF WEB HISTORY

In early 1996, in the Usenet newsgroup for botnet machines can run semi-autonomously, the hacker magazine 2600, the term for this receiving command-and-control instructions technique makes its first appearance: ‘phish- for new spam campaigns and then spewing ing’ (‘mk590’, 1996). out messages in the millions, adjusting each A representative spam business in the mid one individually (‘per-message polymor- 1990s – who spelled it ‘fishing’ – used a simple phism’) and shifting strategy if they receive ASCII picture, ‘<><’, to note AOL accounts an unusually high number of rejections they’d captured to deluge people within the (Kreibich et al., 2008). In fact, a unique, tar- AOL network with spam ads (Brunton, 2013: geted ‘spear-phishing’ attack, like the one that 76). That was small change, though, com- got John Podesta’s Gmail login and disrupted pared with the uses to which phishing would the 2016 American election, is a flashback to be put. Targeted phishing messages used a more artisanal, personal time in the Web. It the same biface aspect of HTML – one side had a carefully designed trick URL (‘myac- of the markup visible to people, the other count.google.com- securitysettingpage. for machines – to send email purporting to tk’) and a beautiful HTML email and login be from a bank, a credit card company, an landing page, both mimicking Google’s email provider, or an employer, with a link style to the pixel. (Similar care was taken whose innocuous text (‘To resolve this block with attempts to get internal email from the on your credit card, click here’) disguises a campaign of Emmanuel Macron in France – suspicious URL (mastercard.l337haxx0r. a tactic now so common that his staff pre- ru, or whatever). Following the link reveals pared for it in advance.) That kind of attack a careful – or sometimes not-so-careful – is now the exception rather than the rule, the counterfeit of the original ‘landing page’ for human touch for high-value targets. When the legitimate site. Careful study of spammer we encounter a spam comment, a spam blog, landing pages reveals how much HTML and a spam email, a message on Twitter @’d to CSS – the markup and styling vocabulary of you by a bikini avatar with a high-entropy early Web design – could convey the ‘real- name, we are very likely the first people to ness’ of a particular online destination. Some have ever seen it. It is the product of layers crudely copy-and-pasted the HTML avail- of wholly computational work, for which the able through ‘view source’ commands for the humans merely set the parameters, assembled sites they were pirating; others, faced with and passed around the world on a chain of better-protected sites, reverse-engineered the mechanical writers and readers. This brings design of their counterfeit, replicating color us to the last chapter of the Web’s counterhis- schemes, trying to duplicate the placement of tory, the closing act of those linkfarms and images, and lifting or in some cases amus- mutual admiration societies: the rise of the ingly improvising the text. post-human social Web. Phishing sites are now often hosted on the compromised computers or servers that make up ‘botnets’, networks of many thousands of machines under the remote con- WE STILL BELIEVE THERE IS HUMAN trol of the spammer. These botnets generate INVOLVEMENT the great bulk of the spam that we encounter (and much that we don’t – spam can be The Web had always been social, of course, upwards of 85% of all email on the Internet alight with cultures of linking, authoring, at peak times, most of it stopped by filters sharing, and citing, with forums, boards, well before a person receives it (Messaging comments, and ‘virtual communities’. But as Anti-Abuse Working Group, 2011)). That a matter of terminology, the Web became verb, ‘generate’, is carefully chosen here: the ‘social’ after it became searchable and SPAM 573 increasingly central, when Friendster, pages and YouTube accounts, serving up MySpace, Facebook, Orkut, LinkedIn, Bebo, porn links and browser exploits and selling Sina Weibo, Twitter, the Marie Celeste ghost armies of Twitter followers, blocks of tens of ship that is Google+, and a million other thousands of Facebook ‘likes’ and YouTube social networks came to dominate much of views and upvotes. A huge portion of human the experience and use of the Web. This was time on the Web became devoted to interact- an aggregation, a pooling, of human attention ing with and producing content that sought on a scale beyond a spammer’s wildest the illusion of salience through popularity – dreams. and if there was one thing spammers were Furthermore, it was a model for aggre- good at, it was producing exactly that illu- gating human attention to which spamming sion. In a sweatshop model that recalls the came quite naturally. It was – and is – a ter- early days of email spamming, employees rain dominated by clickbait and linkbait, of ‘likefarming’ firms will ‘like’ a particular by eyeball-grabbing fake news (a technique brand or product for a fee. The going rate is pioneered by spam emails with links that a few US dollars for 1,000 likes (Schneider, launched malware downloads) and you’ll- 2004). Performed in narrowly focused bursts never-guess headlines bannered over the thin- of activity devoted to liking one thing or one nest content, by the endlessly refilled candy family of things, from accounts that do little bowl of meme culture, and advertisements else, this tactic is easy to spot, so they have to indistinguishable from old-school spam generate the appearance of casual use. They come-ons (weight loss, penile enlargement, do this by liking pages recently added to the predatory home-loan scams, and other drag- feed of Page Suggestions, which Facebook nets for dim fish). Even the legitimate human promotes according to its model of the user’s users developed spam-like approaches to their interests – they behave, in other words, as activity. Merlin Mann, a bemused witness to ideal Facebook citizens, heavy users con- the dot-com scene, dubbed this activity on stantly clicking the thumbs-up and endorsing Twitter personality spamming, the work of whatever Facebook’s recommendation algo- arrogating attention for oneself, using social rithm thinks they will endorse. media to build an audience – often a very Twitter, likewise, has an enormous bot carefully quantified audience of ‘followers’ problem. It must regularly conduct sweeps and ‘rebloggers’ – rather than a network to purge the bot accounts from its ranks, of friends. It is the socially acceptable but but the bots follow paying users in packs of aggressively eyeball-hungry work of those thousands to make them look important and who would be, or act like, celebrities, ‘influ- popular, as well as random humans to create encers’, or ‘thought leaders’. From the Web the illusion of normalcy for their other activi- 2.0 status culture analyzed by Alice Marwick ties. A too-successful purge is rewarded with (2013) to Whitney Phillips’s media-savvy outrage as users see their follower numbers trolls (2015), to Limor Shifman’s circulating plummet (and Twitter as a company sees its memes (2013) and Sarah Jeong’s ‘Internet of value drop, likewise, as the pool of active Garbage’ (2015), studies of the social Web users shrinks). The same is true of buying capture how difficult it can be to distinguish views for your YouTube video, listens for from what its own users call spam. your song on Soundcloud, or clicks on your Spam was a natural fit for this set of plat- ads. In a Web 2.0 version of Goodhart’s forms and practices – so much so that it Law (‘When a measure becomes a target, it produced a similar paradox to that faced by ceases to be a good measure’, or, ‘What gets Google, for which its most optimal custom- evaluated, gets gamed’), any metric meant to ers were spammers. Spammers jumped into describe human interest, esteem, or attention creating fake Twitter accounts and Facebook more generally will spur the development 574 THE SAGE HANDBOOK OF WEB HISTORY of customized code to take it over for pay whose work makes regular data entry look (Goodhart, 1981: 116). All of these vast social exceedingly pleasant by comparison, are Web platforms have created models in which essentially being paid to be human – to exhibit the spammers boost the metrics in exactly the a theoretically solely human characteristic. way they’re supposed to be boosted – just In that labor, and in the statement ‘We still not as legitimate human users. (What these believe there is human involvement’, we can models say about the expectations for the see a Web increasingly and finally dominated humans, those hoof-clicking attention cattle, by the activity and content not of humans but is left to the reader.) of software, and humans directly responding I would like to close with a final prob- to and directed by software. As Ben Light lem, a ubiquitous mark on the structure (2016) points out, human agency was always of the Web left by spam, one that points the centerpiece of Web 2.0 – but, observed towards the future: the humble CAPTCHA. more closely, it’s clear we miss the real story The CAPTCHA system – the deformed let- of the contemporary Web if we fail to account ters on weird backgrounds that only humans for all the nonhuman agency and activity on can read, in theory, to verify their non-bot it. This is an appropriately grim and para- status – is meant to block automated posting, doxical note for the close of this counterhis- commenting, and account-creation tools, key tory of the Web: with Twitter accounts made components in the contemporary spammer by humans solving problems for machines, arsenal. CAPTCHAs make it harder to start to provide other humans with the illusion of new Blogger blogs or open more free email social activity, of ‘Web presence’. accounts, and spammers have been working We have followed the work of drawing the assiduously on different fronts to overcome line defining spam and non-spam through them. In May 2008, the security company the history of the Web, from links and sites Websense documented a series of attacks to blogs and search to the exquisite fakes on the account-creation process of email that mislead users of the Web’s infrastruc- services. Many requests for accounts kept ture. Throughout, I have argued that the act hitting the CAPTCHA stage, and most, but of calling something ‘spam’ tells as much not all, failed (Whoriskey, 2008). The pace about what is being excluded as what is being (replies in six seconds) and the failure rate identified: junk results and salient searches, (nine to one) suggested that computers were personality spamming and meaningful social doing the solving. ‘We still believe there is content, fake-out landing pages and the real human involvement’, said the company’s thing, abusive advertising and legitimate statement. Botnet attacks on text recogni- applications. I hope I have also conveyed tion have improved enough since then that how blurry those categories can be. As we new forms of CAPTCHAs rely on identify- follow the movement of the word and the ing somewhat ambiguous visual informa- systems to which it is applied, spam exposes tion, like picking storefronts out of a set of how vague and tricky the distinctions can be, pictures of buildings. To solve this, spam- and will continue to be. Spam exposes the mers have turned to automating humans with failures and holes in models: how relevance services like Captcha King, which retrieves is calculated, what a valuable collection of the CAPTCHA images from things like the interlinked Web pages looks like, how people Twitter account-creation process for manual understand the pages they see and how their entry. An outsourced staff sits there all day machines construe them, what constitutes banging out CAPTCHAs, with a guaran- approval, interaction, relationships, society, teed ‘success rate of 95% with a response even humanness. In some of the cases I’ve time of less than 90 seconds’ (Krebs, 2012; described here, spam indicts the very systems Motoyama et al., 2010). Those poor souls, it exploits; it produces optimal users, pushing SPAM 575 business models to their logical extreme. Goodhart, C. (1981) ‘Problems of Monetary Following all these developments provides Management: The U.K. Experience’, in a different history of the Web – searchable, Anthony S. Courakis (Ed.), Inflation, Depres- central, and social – through the ways each sion, and Economic Policy in the West (pp. development created communities, attention, 111–146). Lanham, MD: Rowman & and the possibility of their own failure. Littlefield. Hayes, D. (1996) ‘An Alternative Primer on Net Which brings us back to the present and Abuse, Free Speech, and Usenet’ (http:// future of the Web, seen from spam’s point www.jetcafe.org/dave/usenet/freedom. of view: in which the humans involved have html). never been less important, mere fodder for Hollis, K. (2005) ‘alt.spam FAQ or “Figuring out content production and analytic stats, under Fake E-Mail & Posts”’ (Rev. 20050130, 30 the watchful eye of the platforms. The end of January 2005) (http://digital.net/~gandalf/ the line. spamfaq.html). Jeong, S. (2015) The Internet of Garbage. New York: Forbes Media. Krebs, B. (2012) ‘Virtual Sweatshops Defeat Bot- REFERENCES or-Not Tests’ (Krebs on Security, 9 January 2012) (http://krebsonsecurity.com/2012/01/ Brin, S., and Page, L. (1998) ‘The Anatomy of a virtual-sweatshops-defeat-bot-or-not-tests/). Large-Scale Hypertextual Web Search Kreibich, C., Kanich, C., Levchenko, K., Enright, Engine’, Computer Networks & ISDN Sys- B., Voelker, G., Paxson, V., and Savage, S. tems 30(1–7): 107–117. (2008) ‘On the Spam Campaign Trail’, Pro- Brunton, F. (2013) Spam: A Shadow History of ceedings of the 1st Usenix Workshop on the Internet. Cambridge, MA: MIT Press. Large-Scale Exploits and Emergent Threats Edwards, P. (2003) ‘Infrastructure and Moder- (http://cseweb.ucsd.edu/~savage/papers/ nity: Force, Time, and Social Organization in LEETStormspam08.pdf). the History of Sociotechnical Systems’, in Larson, W.L. (1994) ‘Re: Green Card Lottery – Thomas Misa, Philip Bray, and Andrew Feen- Final One?’ in news.admin.policy, 12 April berg (Eds.), Modernity and Technology 1994. (pp. 185–225). Cambridge: MIT Press. Lialina, O. (2009) ‘A Vernacular Web 2’, in Olia Fetterly, D., Manasse, M., and Najork, M. Lialina and Dragan Espenschied (Eds.), Digi- (2004) ‘Spam, Damn Spam, and Statistics: tal Folklore Reader (pp. 58–69). Stuttgart: Using Statistical Analysis to Locate Spam Merz Akademie. Web Pages’, Proceedings of the 7th Interna- Light, B. (2016) ‘The Rise of Speculative tional Workshop on the Web and Databases Devices: Hooking Up with the Bots of Ashley 67 (2004): 1–6. Madison’, First Monday, 21(6) (http://first- Galloway, A.R., and Thacker, E. (2008) ‘On monday.org/ojs/index.php/fm/article/ Narcolepsy’, in Jussi Parikka and Tony D. view/6426/5525). Sampson (Eds.), The Spam Book: On Viruses, Lovink, G. (2005) ‘The Principle of Notworking Porn, and Other Anomalies from the Dark (Concepts in Critical Internet Culture)’ (lec- Side of Digital Culture (pp. 251–263). ture, Hogeschool van Amsterdam), 24 Febru- Cresskill, NJ: Hampton Press. ary 2005. Gansing, K. (2011) ‘Spamculture: The Informa- Ludovico, A. (2005) ‘Spam, the Economy of tional Politics of Functional Trash’, in Miyase Desire’ (Neural.it, 1 December 2005) (http:// Christensen, André Jansson, and Christian www.neural.it/art/2005/12/spam_the_ Christensen (Eds.), Online Territories: Glo- economy_of_desire.phtml). balization, Mediated Practice and Social Marwick, A. (2013) Status Update: Celebrity, Space (pp. 89–109). New York: Peter Lang. Publicity, and Branding in the Social Media Goldsmith, J., and Wu, T. (2008) Who Controls Age. New Haven: Yale University Press. the Internet?: Illusions of a Borderless World. Messaging Anti-Abuse Working Group (2011) Oxford: Oxford University Press. ‘Email Metrics Program: The Network 576 THE SAGE HANDBOOK OF WEB HISTORY

Operators’ Perspective. Report #15 – First, Pringle, G., Allison, L., and Dowe, D. (1998) Second and Third Quarter 2011’ (http://www. ‘What Is a Tall Poppy among Web Pages?’, maawg.org/sites/maawg/files/news/MAAWG Computer Networks & ISDN Systems 30(1– _2011_Q1Q2Q3_Metrics_Report_15.pdf). 7): 369–377. ‘mk590’ (1996) ‘AOL for Free?’ in alt.2600, 28 Sandvig, C. (2013) ‘The Internet as Infrastruc- January 1996. ture’, in William Dutton (Ed.), The Oxford Motoyama, M., Levchenko, K., Kanich, C., Handbook of Internet Studies. Oxford: McCoy, D., Voelker, G., and Savage, S. Oxford University Press. (2010) ‘Re:CAPTCHAs – Understanding Schneider, J. (2004) ‘Likes or Lies? How Per- CAPTCHA-Solving Services in an Economic fectly Honest Business can be Overrun by Context’, Proceedings of the USENIX Secu- Facebook Spammers’ (TheNextWeb, 23 Jan- rity Symposium (August 2010): 435–452. uary 2004) (http://thenext Parikka, J., and Sampson, T.D. (2009) ‘On web.com/facebook/2014/01/23/ Anomalous Objects of Digital Culture: An likes-lies-perfectly-honest-businesses-can- Introduction’, in Jussi Parikka and Tony D. overrun-facebook-spammers/). Sampson (Eds.), The Spam Book: On Viruses, Shifman, L. (2013) Memes in Digital Culture. Porn, and Other Anomalies from the Dark Cambridge, MA: MIT Press. Side of Digital Culture (pp. 1–18). Cresskill, Sullivan, D. (2002) ‘Death of a Meta Tag’ NJ: Hampton Press. (Search Engine Watch, 30 September 2002) Phillips, W. (2015) This Is Why We Can’t Have ( http://searchenginewatch.com/ Nice Things: Mapping the Relationship article/2066825/Death-Of-A-Meta-Tag). between Online Trolling and Mainstream Whoriskey, P. (2008) ‘Digital Deception’, Wash- Culture. Cambridge, MA: MIT Press. ington Post, 1 May 2008.