LAST FM and the Music Multiverse

Mat Wall-Smith, Ross Rudesch Harley, and Andrew Murphie

[The Third Annual Art of Record Production Conference took place in
the Brisbane, Australia from December 10 – 11, 2007.]


The last few years have seen an explosion of online music services that challenge the dominant modes of digital music publishing and distribution, often claimed to be “pioneered” by Apple iTunes. This period has also seen the development of new software that filters vast quantities of data in order to make specific suggestions based on user profiles and purchase histories (a strategy made popular by Amazon’s tracking of customer habits to build personalised recommendations).

The rise in popularity of web-based music communities could well be seen as facilitated by, and situated within, the same technological framework as that of the Itunes or Amazon approach. But the rise of multi-user music communities can be usefully situated within a slightly different set of histories. That alternative history foregrounds collective effort, open databases and community participation. Although there are many social networking sites that are experimenting with the way we use and engage with music in our everyday lives (eg Finetune, Pandora, MOG, MyStrands, iLike, Myspace and iJigg), The social music network LastFM provides us with a really interesting example of how a particular approach to database systems can generate new ways to expand the listener’s sonic horizons. Rather than building a top-down expert system. LastFM has developed a bottom-up and agile system that opens up unexpected pathways to the discovery of music that listeners may not have found otherwise.

The way Last.FM, perhaps serendipitously, has managed to plug into a number of different ideas about listenership, taste, fandom, expert knowledge, databases and niche audiences makes it an interesting model for a vital and participatory engine of discovery and the basis of what we will call a new economy of affect. According to the work of Chris Anderson in the book of the same name we are living in a ‘long-tail’ moment that is radically expanding the depth and breadth of both back-catalogues and new recordings (1). The massive proliferation and free availability of artists, genre’s, and forms, that characterizes this moment presents a challenge to the traditional architectures of discovery upon which the established industry has survived. Although many punters and industry commentators are tempted to say that this sonic overload will simply be overwhelming, we see it more as the last piece of a bigger puzzle to do with the digitisation of music in general. Along the way, we want to ask whether the history of collectively created music databases such as Gracenote sheds any light on the possible difficulties LastFM will face since its purchase by CBS earlier this year. Does Gracenote’s transformation from an open-content music database to a for-profit proprietary system raise concerns about the leveraging of community or fan knowledge for corporate gain?

Production, Distribution and Discovery

In order to understand these developments, we need to see them as part of the radical transformation in the making and reception of music. The rise of the “hyper-active” listener (listeners have always been “active”, but maybe not so much as they are now) is part of major shift that has taken place in the making and distribution of music. For the purposes of this paper we frame our work by way of the broad categories of production, distribution and discovery. Along with others in the field, we see the main challenges and shifts at the moment to be taking place in the area of discovery. Although we will continue to see many new developments and work practices emerge, the area of production has already been massively transformed by digital methods, workflows and processes.

The “democratisation” of music-making as a result of the widely available and relatively inexpensive access to recording tools, software and hardware has been much commented on (1). Similarly, the genie is well and truly out of the bottle in terms of the distribution and massive circulation of music via an extremely wide variety of digital formats and systems. Although this transformation is still playing itself out, there is no doubt that the easy availability of a vast sea of audio recordings via the web, satellite, file-sharing, peer-to-peer, mobile networks, WiFi and a range of other networked technologies has changed the shape and nature of the recording industry. Though there are many battles still raging around the nature of rights, who should control them, and the conditions under which digital music should circulate, there is no question that we will continue to witness an explosion in the availability of all kinds of digital music, and that we will see a massive multiplication of artists and their recordings made available to the listening public.

But, how will we deal with this? As Roger Richards, head of Melbourne’s Extreme Records recently put it,

“Rapidly falling prices of music production equipment has resulted in a glut of music which conversely increases the need for new intermediaries, which is where the rise of online recommender services, online social networks and niche blogs are replacing the old world of print media, face to face record stores and traditional distribution. Likewise, as the biggest bands make the final transition into ‘brands’ and their releases become ‘advertising’ for their expensive live shows, the impact upon niche music scenes in which ‘live’ performances are rare and releases act as social documents, is unknown.” (Richards and Chan, 2007: p 9)

Although the idea of democratisation is problematic, we’re using it here to signal the extent to which access to the means of production and distribution has shifted away from the corporations, who have been losing their tight grip on studio recording and mass distribution of music. As music production and distribution drifts increasingly away from the orbit of the corporate recording industry, we see the status of music shift from product to process, from consumption to participation, from object to a network-of-relations.

So under these new and constantly changing conditions, how do we navigate the “cellestial jukebox” that so many new music pundits are keen to invoke — whether it’s Rick Rubin’s $19.95 per month cable subscription model, or the music-as-water concept that suggests music should be treated as a utility service? In trying to formulate one possible scenario for the future of the music industry in this context, Simon Frith and Lee Marshall paint the following picture that sets up our investigation:
“All music is now easily copyable and distributable by anyone with a computer, mobile phone or television. This has hit the major record labels hard — they’re much smaller operations than at the end of the twentieth century … [with] small independent labels serving localised niche markets. As a general development, music has become less mainstream. For consumers, while they freely copy recordings and download from P2P services, there is a feeling that music isn’t as important as it once was, that finding good music among all of the tunes swirling around is just too much hard work. And, with little possibility of gaining a deal with one of the old majors, and without taste-shaping media such as national music magazines any more, musicians find it hard to make their work heard or valued.” (Marshall and Frith, 2004: p 210)

The development of CDDB, GraceNote and FreeDB.
The rise of a networked music ecology can be framed as part of a larger history regarding the digitization of music and the effects of this digitization on the way we contain and categorize music as part of an extended information architecture. As recently as the early 1990′s there was no metadata system for Compact Disc. Placing a CD into a computer didn’t automatically provide any metadata for that content. The metadata normally associated with the content via its physical packaging had yet to be reproduced in a form that would make its automated reconnection with the digitized content possible. In 1993, the amateur software developer Ti Tan wrote a Unix-based CD player called XMCD. Not only could it play CDs on your computer, but it could utilize a disc recognition system to match CD track-listings with files kept on the user’s computer. Teaming with college-friend Steve Scherf, they developed an internet based service called Internet Compact Disc Database (CDDB), and enlisted an army of fans and volunteers to enter tracklists into the database. As CDs have no such metadata contained within them, the first person to submit the information to the database provides the information for the next person. “CDDB automatically gave information on any CD it knew, plus added new information as it came in from users.” (Fry, 2001) By 1997, CDDB had become the default music database, and in 1998 CDDB was acquired by the media player manufacturer Escient for just under a million dollars. From this point the CDDB system was licensed under the name Gracenote. Gracenote has gone on to become the most dominant player in an industry concerned with the management and capitalization of vast libraries online media information.

The CDDB system was, and still is, the result of the collective knowledge and labor of music fans who voluntarily enter track and CD metadata combined with information provided directly by record labels. Despite the collective production of Gracenote’s metadata, most of its revenue is provided by licensing access to the database. Many of the key online music services and hardware providers deploy Gracenote’s libraries. According to one source, CDDB has been used by more than 150 million people for a total of about six billion database searches. Clients include Apple, RealNetworks, Sony, Panasonic and Honda, who utilise the database in their propriatory technologies. ( Gracenote has grown to become one of the leading music recommendation systems for the music industry, and is focusing much of its current research on the problem of discovery. Their approach is based on “seeds” of user knowledge and what they refer to as “360 degree” personalized recommendations. Gracenote is the most ubiquitous music database on the planet but the capitalization of a collective knowledge base has raised some serious questions and complications – both for Gracenote and for other emerging systems that harness user-generated content for profit.
The use of what was initially created collectively as public-domain knowledge for profit, incensed many who had worked on producing data for the CDDB or had deployed it in the development of their software. As one representative of this view puts it, the “owner of CDDB and related trademarks and copyrights, as well as a number of software patents related to music metadata lookup… took a database that had been built by fans and turned it into a private company.” (“Gracenote: eminent domain over the public domain”) In addition to these corporate appropriations of collective knowledge, Gracenote has also been engaged in litigation with competitors such as MusicMatch and customers like Roxio (who decided to include FreeDB, the open content alternative to Gracenote, in their rip and burn software instead.)
In the outcry that followed the commercialisation of CDDB, a number of determinedly open source and community database systems were developed. ‘FreeDB’ remains the database of choice for developers and is considered a ‘clone’ of CDDB in terms of functionality ( FreeDB is also licensed on a General Public License and is committed to remaining open and free. Both projects use a ‘nearly’ unique identifier in order to identify CD’s and their tracks according to their published order. This is of interest because the taxonomy that’s being used to handle meta-data in this model is a hangover form the CD-era. Songs are attributable only as tracks on albums according to the tracklisting system of the Compact Discs on which they are distributed.

MusicBrainz was another database project that developed out of a strong community reaction to Gracenote. It provides an interesting juxtaposition to the more dynamic relational databases of ‘entertainment’ and ’social’ music sites such as LastFM and Pandora (simply in terms of populating a database). MusicBrainz is a Not for Profit Foundation that uses two different ‘audio fingerprinting’ technologies to allow individual tracks to be identified and associated with appropriate metadata according to their audio characteristics. This approach removes any reference to the original CD architecture. Instead of identifying tracks according to their position within a defined playlist, fingerprinting technology allows the track to be identified by audio analysis. This is also one of the main areas Gracenote, MuiscBrainz’s arch enemy, is moving into: audio waveform recognition. The non-commercial basis of MusicBrainz has resulted in relatively slow development cycles, and so projects like who harness user generated information are developing database back-ends at a much faster pace. Unlike MusicBrainz those systems are unhindered by an archaic infrastructure based on physical media and a mission that lacks relevance in a rapidly and ever-changing networked music ecology. MusicBrainz is stuck between its history as an open ‘meta-database’ for catalogueing and attributing user-submitted data to the future development of a useful relational database.

A New Economy of Affect

We have figured three steps are barriers in realizing a new music industry. The first barrier was production. The second was distribution. Both of these have been obliterated by cheaper and more professional home recording technology and an increasingly networked world. But this brings us to the issue of our third and final barrier to a new music industry; Discovery. The fact that a band like Radiohead can afford to go it alone actually tells us very little about the new music landscape. As was infered in our introduction all this psuedo-independence tell us, is that Radiohead has become its own brand. If any other emerging artist was to try the same stunt it’d obviously fall flat. Discovery is nearly everything to an emerging artist. It is for this Reason that Ross, Andrew and I understand discovery rather than distribution as the crucial battleground between established an emerging forms of music industry. We argue that the new ecology of music to which we have already referred demands a new economy of affect to augment if not supplant a broadcast economy made increasingly redundant by a massive proliferation and distribution of networked music.

Let’s begin by pulling apart this term ‘economy of affect’. We can simplify ‘economy’ by describing it as the relationship between supply and demand. This relationship is far from given – in fact the new music economy illustrated very well in Chris Andersens ‘The Long Tail’ seems to indicate that a surplus on the supply side can drive demand (1). His figures show that the more sounds that are made available, the more gets consumed. It is our argument that this distribution demands a more appropriate engine of discovery. ‘Affect’ here refers to the things that move the body. As opposed to logical thought out deduction or calculation Affect refers to the raw modulation that precedes thought; The stuff that moves the body to think, but also to fight, or to flee, or to dance.

Contemporary theory on the relation between thought and perception shows that much of what we think of as being rational descision-making is actually occurring at a more fundamental pre-conscious level. We are moved to thought and to act by a network of intense relations. I am motivated to collect the music that moves me, the music that adds something to the intensity of everyday experience. So the term ‘economy of affect’ here refers to the realization of this intensity in the service of selling stuff- music in this case. Music has to move me before I will buy it; the experience of music for the most part prescripts my consumption.

Broadcast technology lent itself to the mass/industrialized production of that affective surplus or excess – that added value of music. Record labels realized surplus via a star system that invested large amounts of capital in the production, distribution, and marketing of a limited number of artists. Earlier we described how distribution changed with the development of digital networks. But modes of discovery don’t necessarily shift along with this move without the design or emergence of alternatives to the broadcast mode of discovery. If we are still reliant on the old mechanisms of discovery then the free availability of content online does very little to realize a new economy of affect. Itunes is perhaps the best example of how the new distribution landscape looks without alternative modes of discovery – that is it looks a lot like the future looks a lot like the past without the realization of new ways to discover that content.

Here i’d like to look at two divergent approaches to this task of realizing a new mode of discovery and therefore a new economy of affect. Both systems are based on the development of relational database systems in order to create audio streams of recommended tracks. The first example is The Music Genome Project and its Pandora front-end and the second is a dynamic user generated system.


The Music Genome Project is a database of music metadata classified by paid musicians who analyse each track manually. The genome trope sets the scene for the Project’s approach – there is an inherent assumption that the experience of music can be practical reduced to its components parts and qualities; tempo, rhythmic qualities, instrumentation, personel and in excess of 400 other carefully defined types that are assigned by The Music Genomes expert taxonomists. The Music Genome as the whole DNA of music trope suggests is not subject to change. The effectiveness of the Music Genome is thus subject to the ability of its logging team to keep up with the almost limitless catalogue of music that currently exists and which thanks to the changes we’ve noted in production and distribution is increasing in density at an exponential rate.

Pandora is the streaming radio side of the Music Genome project. It works by deploying the genome metadata to make connections between tracks based on those 400 qualities and streaming recommendations. There is nothing quite so off putting as a recommendation engine that begins by asking you for your favourite artists but then can’t locate them in its database. On the other hand, there is nothing quite so engaging as inputing the three most fringe artists you can think of and having them return a result. With Pandora my experience is the former, with the latter. The Music Genome Project is the basis for what is called an ‘expert system’ approach to an ‘intelligent’ recomendation service. The problem with any expert system however is that they tend to work very well in a carefully defined territory where all data-objects can be recorded and all there possible movements mapped. This is a problem realized in a variety of other fields including so called ‘strong’ Artifical Intelligence. Once you move outside the map ‘expert systems’ tend to collapse. The problem with mapping out the so called ‘long tail’ of the new music ecology is that it keeps getting longer, growing mutant offshoots that interconnect in weird and wonderful ways.

The Pandora/Music Genome engine makes some key assumptions that limit the dynamism and I’d argue the effectiveness of its dynamic playlisting. It assumes that the style of music that I am listening to at any one moment is the one that I want all my listening to be guided by. That is to say it assumes my listening habits are orderly and predetermined. It assumes that if I had listened to John Butler, I’m likely to want to listen to Xavier Rudd next, and Ben Harper after that – a rather tedious prospect if it was to continue indefinitely. This may indeed may make Pandora an effective engine for a personalized radio stream for those who prefer a well travelled path, but as a model of discovery it shows the limitations of an expert system to confuse the always incomplete map for the territory.

As a final note on Pandora and The Music Genome Project we’d like to reinforce the point that music culture is not genomic, we are not even sure that music itself is – but music cultures definitely aren’t. Assuming that my music listening habits can be mapped and utilized in this way is akin to suggesting that our DNA can explain all the contingencies of human cultural development. In both cases we are negating the intense experiences and interactions that draw out and fold forward a multitude of possible futures from any particular quality – that define music and human as vital and in a state of continual flux and proliferation. began as two projects founded by two distinct groups. The first project was Audio Scrobbler which allowed users to populate a database and to receive recommendations within popular audio software and portable music players. The second project was which was based on dynamic user generated streaming. These two projects define the two aspects that make their combination as the current system such a revolutionary model of discovery. Since 2005 the two have been completely integrated under the banner. The real-time streaming adds a recursive intensity to the generation of the combined and Audio Scrobbler database as my interactions with the stream continuously fold back into the database realizing new connections and new possible streams. For this reason the system feel very live and active as a listening experience. An open API provided for the integration of the playing history of the iPod and iTunes to be ’scrobbled’ automatically. The ease of uploading and augmenting metadata according to experience, the ubiquity of compatible music software and hardware, meant that Audio Scrobbler and were able to develop a massive and completely dynamic and agile relational database based on a huge and actively productive community of ’scrobblers’. represents a model for the development of an alternative economy of affect to the broadcast/mass production model of the legacy industry. Firstly, Audio ‘scrobbling’ refers to the automated submission of metadata about what tracks you play on your computer or your portable player to’s database.’s database is populated with this data as well as the data that exists as a function of its own streamable catalogue. Unlike Pandora this means that there are no missing links or spaces in’s catalogue. The moment I ‘scrobble’ a track it is added to the database. If that track is badly named or has erroneous metadata or is simply an unreleased private jam it will still get scrobbled. However that errant data won’t link to any other user’s playlist. It will thus fall off the radar via a process of natural attrition; it will never be recalled in the dynamic network that is the collective consciousness of the system. The crucial thing to note here is a big difference in the type of metadata that preferences. Its not concerned with metadata based on a strict and static system of categorizations but rather data based on the communities experience of the data object – it treats the music as an active component in a vital community rather than a given object in a static reference system. Crucially it serves the whole musical spectrum equally and allows connections to be made between diverse music types regardless of genre or market position. My recomendations can thus moves from Merzbow, to Coltrane, to Moving Ninja simply because those tracks are connected via peoples active listening experience.

The original power in engine was its ability to move beyond the performance aspect that usually operates in any online profiling engine. By ’scrobbling’ data automatically based on the users actual listening history the engine avoided the usual quirks of a more ‘reflexive’ self-profiling. On the early scrobbler integrations (before the development of social bookmarking features) you ‘were what you listened to. The addition of the simple schema of ‘skipped’, ‘loved’ and ‘banned’ options meant that the data was being ’scrobbled’ according to an affective reaction to a track. The simplicity of that schema reduced the tendency to ‘intellectualize’ this reaction. On tags were effectively delimited to those three ‘affective’ categories and the relational power of was built on the connectivity that such a delimited schema provided between profiles and consequently between artists, between individual tracks etc. The power of lies in its ability to move beyond the assertion of a particular taxonomy or typology. The addition of groups, friends (as opposed to listening ‘neighbours’) and tags reintroduced the ’subjective’ structuring of musical relations and for many this ’social’ networking aspect has become the principle element of This hasn’t yet threatened the diversity of playlists but has added a degree of peer influence to profile generation. Cliques tend to emerge and people increasingly move into ‘closed’ neighbourhoods with the ability to listen to another members or groups stream by choice. That said the combination of these two original features, an open user-generated database system and a simple affective taxonomy, provide for a engine of discovery that really has changed the way I listen to music.

And this brings me to the final point regarding the model and its most important aspect in realizing a different kind of economy of affect to broadcast or to an expert and centrally controlled system – last allows for a distribution of intensity – what Andrew Murphie refers to as a kin to the ‘dynamic of the live’. is only controlled by people actually listening and by people you know,or get to know, as active listeners. This allows for very immediate listening experience that adds a certain veracity of experience to the playlists. The experience of feels is a vital and immediate social experience and perhaps marks the remembering of music as a generative social fabric.

In conclusion we;d like to suggest that offers a model for a system that is an agile enough system of discovery to deal with the new dynamism that is realized in the music ecology of the networked world. Moreover it allows people to share the experience of listening intelligently. Here though ‘intelligently’ actually means removing much of the intent and the performance that compromises the metadata of many user generated systems and which obfuscate the realization a more fundamentally embodied knowledge. There are numerous problems with the system and i don’t mean to laud it uncritically. It does tend to trap mainstream listeners in mainstream listening circles. At the same time however listeners on the fringe are more often than not those actively populating the database and making new connections between artists. also might well be criticized for its centralization and capitalization upon metadata in a manner that hands over the information generated by a community of users to a corporate concern – admittedly voluntarily and for a trade off in terms of functionality. The question remains, that having been sold to one the original powers in broadcasting, whether will continue to provide such an open engine of media discovery and an uncontained but very industrious economy of affect. Its both plausible and realistic to imagine a sytem like that is open-sourced and physically distributed so that the rights of artists and consumers alike can be ensured rather than freely handed over to a corporate intermediary. I’d suggest a combination of Bit-torrent (which will soon offer streaming), and a like system of discovery based on the user’s active listening experience, offers the potential for a reinvigorated music industry – or perhaps I should say ‘more reinvigorated’ for despite the big stakeholders of the music industry crying foul I’ve found that thanks to , emusic, and blogs like Cyclic Defrost, that the last few years have been the most exciting and most vital years of listening of my life.


(1) Chris Anderson, The Long Tail; The Future of Music: Manifesto for the Digital Music Revolution; The Future of the Music Business: How to Succeed with the New Digital Technologies.
Sebastien Chan and Roger Richards, “Antipodean extremities interview”, Cyclic Defrost 18, November 2007.

Simon Frith and Lee Marshall (eds), Music and Copyright, Edinburgh University Press, Edinburgh, 2004.
Jason Fry, Three veterans advise the next tech wave: it’s all about business”, Wall Street Journal, Decemeber 31, 2001.
“Gracenote: eminent domain over the public domain”,