Maskinskrift

I min iver att inte slänga bort min tid och på samma gång inte lösa några problem har jag utfört en serie intressanta experiment. Tyvärr måste jag konstatera att resultatet av dessa var en aning nedslående, men jag känner att det är värt att rapportera om ändå.

Det talas så ofta om litteraturen att jag ibland undrar om vi inte pratar om den mer än vi faktiskt läser den. Något som säkerligen sker för sällan är samtal med litteraturen och det är just vad de senaste dagarnas experiment handlat om. Med hjälp av konversationssimulatorn MegaHAL har jag samtalat med tre klassiska svenska romaner: Röda rummet av Strindberg, Kallocain av Karin Boye och Hjalmar Söderbergs Doktor Glas.

För att lära en roman att tala matar man in hela textmassan i MegaHAL, ett tämligen användarovänligt konsolprogram utan grafiskt användargränssnitt. Mac- och Iphoneanvändare drabbas förmodligen av svår chock vid blotta anblicken av något så gräsligt och primitivt. Rent praktiskt så behöver man ha tillgång till romanen i textformat – jag laddade ned ovanstående romaner från Project Runeberg. Därefter måste man omvandla texten till en träningsfil som innehåller hela romanen med en mening per rad. Träningsfilen används, som namnet antyder, för att träna upp MegaHAL i romanens språkbruk och stilistik.

Nu lär sig ju programmet förstås inte ordens mening och betydelse, utan det handlar om en inlärning av matematiska och statistiska samband mellan ord och fraser. Upplärning sker dessutom kontinuerligt under samtalens gång genom att programmet analyserar inmatningen från användaren. MegaHAL implementerar en ganska primitiv form av ”artificiell intelligens” (AI) som bygger på så kallade Markovkedjor. Min tilltro till mer avancerad AI är dock fortsatt låg.

Experimentet krävde mer jobb och förberedelser än jag trodde, bland annat för att jag behövde konfigurera och hacka programmet för att hantera svenska korrekt. Min förhoppning och förväntan om goda samtal vänner emellan drev mig dock tålmodigt vidare.

Experimentet blev på det stora hela en besvikelse. Samtalen med romanerna blev ofta förvirrade och inte sällan var de till och med otrevliga. Doktor Glas och Kallocain gav jag upp om ganska snart, för hur jag än försökte var de mer eller mindre deliriska. Jag tog fram mina mest älskvärda, charmerande och lismande sidor, men inget tycktes hjälpa. Röda rummet gick i sammanhanget betydligt bättre men dessvärre talar vi inte längre med varandra.

Här följer ett komprimerat sammandrag från ett av samtalen med Röda rummet (mina repliker i fetstil):

Hur står det till, bäste bok?
Jaså, ja, det läste några diagonaler och det gör en bok?

Nej du, några diagonaler gör ingen bok.
Jag går sönder hans – ja!

Pärmen är något sprucken, men så värst sönder är du inte.
Jag förstår!

Jag vill bara prata lite.
Han ska inte prata smörja om att verkställande direktören sin röst, som icke gått långt borta vid drivbänkarne i skydd av planket sutto två människor som icke var ett fint huvud, må du tro, den är något finare?

Kanske skulle man ändå satsa sina pengar på obligationer.
Nå, jag erbjöd herrn; vet herrn vad det nu haft under min vård var förlöjligat och måste befinna sig bland boskapen, blir respektabel, får du pengar på obligationer.

Dina logiska resonemang förvirrar mig, är de ockulta?
Vad var det för dig!

Jag tycker du har en väl hård ton.
Du är så rörd att han icke ha hänt förakt!

Jag är inte rörd och jag förstår inte vad du menar.
Jag menar att det var blott förruttnelsens doft, och han hade kört hem och lät brodern njuta av varandras sällskap.

Jag pratade just med Kallocain, men det var inte direkt givande.
Du var inte direkt givande.

Du kan vara ganska otrevlig.
Jag skall gå i trapporna, vi ha all anledning att påminna henne om, att jag i dag kan vara ganska otrevlig.

Låt oss bedöva oss tillsammans. Det tycks vi båda behöva.
Det är för oss som en trampad mask, och du behöver vi några år sedan, en statskalender, som var högfärdig, därför att man har hustru och barn och jag kan göra det!

Hur står det till med förståndet?
Jag förståndet?

Folk pratar inte som du gör längre. Tack och lov för det!
Tack, hjärtligt tack!

Jag funderar allvarlig på att ställa ned dig i tvättstugan.
Allvarlig men glad, uppriktigt glad på hela sitt liv, och nu har jag fått stå för slag?

Ja.
Han sade tvärt nej!

This is what I said (and what I should have said) during my speech in Gothenburg last week.

1. What is Feeding Frenzy?

Since I was informed that I was one of the winners of FFKP”s ”idea contest” I have happily told friends and family about it. Of course I also mailed them my application text and the press release. The problem is that nobody understood what the idea is about. I don’t blame them. The text I sent to the contest is rather cryptic using vocabulary inspired by the american author William Burroughs, conspiracy theories and alchemy. I thought that was fun and suited the idea quite well. Anyway, here is a more straight forward explanation.

Feeding Frenzy is a piece of conceptual software. The basic idea is to manipulate the text content of web feeds (i.e. RSS and Atom) by using different collage techniques or text tranformationen algorithms. Basically it is a web application where you can register the URL to any feed (e.g. blogs, news web sites, etc) and choose which text transformation algorithm you want to use. This will result in a new URL – a kind of ”remixed” text feed – that you can subscribe to with your favourite feed reader client (e.g. Google Reader, Thunderbird, etc).

Feeding Frenzy is so far the project name during the development. It might change – and probably will.

2. William Burroughs and the Cut-up technique

The idea is much inspired by the works and thoughts of William Burroughs, especially his ”cut-up” technique from the 60s. Cut-up was discovered by an english artist named Brion Gysin. What Brion did was that he copied out phrases from different sources, e.g. news papers and magazines. He then cut out pieces from different phrases using a razor or a pair of scissors, and rearranged the fragments. This produced a new text that was written down on paper or with a typewriter. Brion was a friend of William Burroughs who soon became obsessed with this new technique.

William Burroughs, an american avantgardistic writer, was quite an extreme and eccentric person in many senses. He was a notorious drug addict, a low-life criminal obsessed with guns and an almost militant homosexual. He was also a well educated, well dressed professor type admired by the beatniks. During the 60s he experimented a lot with cut-up and other similar text processing techniques. This resultetd in, among other things, the Nova trilogy – a non-linear, fragmentary story built of so called ”routines” and cut-up passages. The plot in the trilogy is basically a hallucinatoric story about a gigantic intergalactic conspiracy against humanity.

In books, articles, lectures and interviews Burroughs has given many different explanations to what cut-up really is about, what the technique means to him. First of all Burroughs wasn’t much of a theorist. The most important thing for him seems to have been the actual practice, or process, of cut-up – to spread the know-how and make people start to cut-up on their own. He even said: Don’t theorize, try it.

It was also a way to make the act of writing, or producing text, more concrete and physical. A way for a writer to become more like a painter that is constantly handling materials and colours, getting his hands dirty.

Burroughs had ideas of the cut-up process as some kind of psychoanalytical machine. He seams to have meant that during the process you create sentences with sub-consciously hidden information. In a lecture at the Jack Kerouac School of Disembodied Poetics from 1976 he says:

Now, how random is random. We know so much that we don’t consciously know that we know that perhaps the cut-in was not random. The operator, on some level, knew just where he was cutting in as you know on some level exactly where you were and what you were doing ten years ago this particular time. But you couldn’t – most of you couldn’t, a few freaks who can – make that knowledge consciously available. And the same way, while you’re doing the tape – on some level you know just exactly where your words are. So cut-ups put you in touch with what you know and do not know that you know.

In the Nova trilogy Burroughs introduces a bizarre theory about human language as a virus spread by aliens at the dawn of man. The reason for this is to take control of the human race, using the language to destabilize the society using ”feedback loops” that eventually will result in a cosmic nova (!). Cut-up is in this context a sort of counter measure or guerilla tactic. By cutting up words both with scissors, razors and tape recorders you are actually hacking the language itself – hacking the control machine – and liberating yourself from the control mechanisms that keep humanity enslaved. Through out the Nova trilogy Burroughs repeats combat instructions:

Calling partisans of all nations – Cut word lines – Shift linguals – Vibrate tourists – Free doorways – Word falling – Photo falling – Break through in Grey Room -

Burroughs has also suggested that cut-up can be used to predict future events. From the same lecture in 1976 he says:

While you experiment with cut-ups over a period of time you’ll find that some of the cut-ups and rearranged texts seem to refer to future events. I cut-up an article written by John Paul Getty and got ”it’s a bad thing to sue your own father” – this was a rearrangement and wasn’t in the original text – and a year later one of his sons did sue him. I mean, it was just purely extraneous information and meant nothing to me. I had nothing to gain on either side. We had no explanation for this at the time. It just suggested that perhaps when you cut into the present the future leaks out. But we simply accepted it and continued the experiments.

Burroughs is an unreliable source, though, even when it comes to his own strange domains. Reality and fiction seems to constantly blend together. What is fact and what is fiction? I am sure the question wasn’t even meaningful to Burroughs.

3. Motivations behind Feeding Frenzy

First of all, I am aware that this idea with Feeding Frenzy might sound strange and pointless. It might even end up being very boring and uninspired, but still I want to try. I don’t want to sound anti-intellectual, but we must remember that this is an experimental practice. Again I quote Burroughs:

I would like to emphasize that this is a technique and like any technique will, of course, be useful to some writers and not to others — In any case [it is] a matter for experimentation not argument.

My main motivation behind the idea is a bit cryptic and hard to explain. Basically, it is about lost mysteries and bad fiction.

The purpose of much modern technology and invention seems to be to solve problems, increase transparency and overcome long distances. Some problems that have been solved during the last twenty years are organizing photographs, remember meetings, write nice looking reports, etc. Our lives are constantly becoming more transparent with tools such as mobile phones, chat and presence clients (e.g. MSN, ICQ, etc), calender sharing and synchronization (e.g. Outlook, Google Calender, etc) and of course social networks (e.g. Facebook, Twitter, LinkedIn). Communication technologies, airplanes, GPS and navigation systems and instant news reporting have made the world feel a lot smaller since grandpa. This is impressing, but also a bit boring and sad. The way problems are solved with technology, things tend to get a bit too organized, accessible and static.

Storm the reality studio!

What we consider to be reality is always a fiction. The reality we are forming now is made up of live broadcasts from wars and elections, browsing through news sites, reading feeds, chatting, weekends in London, synchronized calenders, Google searches, Facebook status lines, Twitter, conversations over Skype and cellular phones, etc. The technology that delivers this simplified reality and transparency is combined with a belief that we always know what goes on and that it is true. Reality turns into daft fiction.

I am not a nostalgic person and I don’t think we should turn down modern technology in a kind of Theodore Kaczynski way. My view on reality is aesthtic and I think that technology could be used very different from now, used to remystify the reality and reinvent the irrational and strange, add some noise and glitches. Literature is one way to achieve this, but why can’t it be done by software too? I will try by cutting up feeds.

I am also very fond of the ”remix” concept. In music the process of remixing and musical reuse in form of sampling is widely used, especially in electronic music. Today the remix concept is used in a more broad sense than before, people talk about a remix culture which is about derivative work and reuse in general. Text transformation can definitely be seen as a remix technique. My hope is that Feeding Frenzy can be an interesting contribution to the world of creative reuse – a sampler for texts and feeds.

As you all know remix culture is today under attack from many directions. Law makers and the industries – the ”control machine” in Burroughs terms – have never really grasped the concept of derivative works. Lately even web syndication has been under attack. It seems as if the Associated Press (AP) and other news medias don’t always like what is done to their feeds. We will have to see how they feel about being cut-up.

I also hope that Feeding Frenzy itself will be ”remixed”, or misused in an interesting way that no one could anticipate. A great example of ”creative misuse” is the Roland TB-303 synthesizer. It was built during early 80s and originally marketed to guitar players for bass accompaniment while practicing alone. As a bass guitar emulator it is horrible. It doesn’t sound the slightest as an electric bass guitar and I don’t think it was very successful in that sense. Quite soon DJs and producers found the little machine and discovered its very characteristic sound – the musical genre Acid house was born. Even today you often hear the buzzing sound of the 303 in a lot of electronic music. In my opinion, being reused is a wonderful destiny for all inventions.

4. Implementation

Some words about implementation. My idea is to implement the whole system in Python. A lot of the source code will obviously be dealing with text processing and Pyhton has excellent support for that. In my opinion it is also a quite neat and elegant language. I will probably use Django to implement the actual website and its features. Django is a popular web application framework written in Python.

I will concentrate on implementing a ”good” cut-up algoithm. An important question is: what is a good transformation technique? For me the answer is simple: One that produces interesting results. There are no rules here, no way of cheating. There are only boring and interesting results. Remember that this is experimentation – it should be interesting and fun.

There are of course many other interesting text tranformation candidates. Since the scissors and razors of 60s new tools have emerged – computers and software. New more advanced text transformation algoithms can quite easily be constructed. E.g. multiple translations: You start with a text and run it through a translation software (e.g. Yahoo! Babel Fish and Google Translate). For example from swedish to german, then from german to french, followed by french to russian and then back to swedish. It can produce quite funny results.

Another example is mash-up techniques using search engines. E.g. you can programmatically cut a text into pieces then run a search with the text pieces and combine a text with help of the search results. It is just an idea. I haven’t seen it in use or tested it. It could be interesting. You can also mash-up many different feed sources which somewhat resembles the old ”fold-in” technique.

To begin with I will try to deploy the system ”in the cloud” using Google App Engine. App Engine is a platform for building and hosting web applications using Google’s extensive infrastructure. It supports both Python and Django. I am not completely comfortable with the idea of using a cloud service, though. Cloud-based platforms makes you dependent, you literary put your data in the hands of someone else. Talk about a control machine (in terms of Burroughs)! The reason why I still consider this is of course due to simplicity and practical details.  No servers to administrate, no failing backup scripts and no concerns about bandwidth. (How easy it is to sell out your ideals and principles – it only takes an ice-cream…)

5. Finally

This idea – Feeding Frenzy – doesn’t contain anything new. The practice of cutting up texts goes back 50 years as we have seen. And collage techniques in other art forms are much older. The idea of implementing cut-up and other text transformation techniques with software isn’t new either. You can find many cut-up machines by a simple Google search. Not even the idea of combining a cut-up machine with feeds is new. A couple of years ago a swedish artist (or maybe an artist collective?) created Poloblogg – an automatically generated blog that mixed together text pieces from many other blogs. It sometimes produced rather fun nonsense articles complete with pictures and hyperlinks.

There are some problems with these systems, though. With Poloblogg you could never alter the input in any way. New texts were generated by combining unknown sources. Both the input and the method was hermetically closed and even if the result could be fun and interesting there was no participation in the process. Other web based cut-up and text manipulation machines give you access to the input, but here another problem arises. The connection of input, the actual text transformation process and the output is slow and you often hesitate over which texts to use as input. It is a rather static and boring process that lacks flow.

The old analogue cut-up method is of course the most elegant and dynamical. It is hard to compete with. But there is one thing where it really can’t compete – again, the flow. Traditional cut-up is very slow and not always fun. What might be new with Feeding Frenzy is the possibility to connect a never ending stream of input to a configurable output in order to create ”remixed” texts.

And if Feeding Frenzy ends up being just boring, at least it doesn’t solve any problems.