DESCRIPTION OF SITE:
The Persistent Data Confidante is a www site allowing for the anonymous transfer of secrets and confessions. The work asks users for a secret after which they receive one previously contributed from another user. They are then asked to rate the secret they have received. Each secret's "popularity" or intrigue increases its probability of being re-told in the future -- thus the best secrets will "live-on" while the more banal will "die-off."
ARTISTIC CONCEPTS:
The main points of reference for the work are:
1. Like a confessional talk show, the Persistent Data Confidante uses pop imagery and dramatic rhetoric to incite participation and fascination with charged "moral" issues.
2. The work plays with recent US censorship policies noting that for www art galleries, a curated site places the responsibility to censor upon the site developer, whereas in an uncurated site the onus of responsibility rests on the contributor. In this strictly user - curated site, it places the judgment of "good taste" on anonymous users who's approval ratings connote consent. On the other hand, secrets of a too banal or dull nature are the most offensive to this community of internet users, and will most likely be democratically eradicated.
3. As secrets are retold (age), they become capable of producing sexual offspring. Each day, secrets that are mature (have lived to experience 100 re-tellings), and which also fall into the top 10% in popularity (according to viewer ratings), will seek a compatible mate (one with similar content.) These two secrets will produce a new offspring which is composed of major clauses from these two parents. Newborn secrets will then be forced to compete for user approval in the rigorous environment of the database if they are to live on to produce their own offspring. This ALife component is of course referencing the manner in which secrets evolve in our verbal world getting confused with others and mutating uncontrollably as they are re-told. Our use of such a system is attempting to establish a parallel, verbal sociology of the net.
The work is rather straight-forward for the individual participant, yet we hope to "set the stage" for a very complex, richly interactive social phenomena. Will users come to the site to live vicariously through the secrets of others; as exhibitionists to achieve pleasure in their anonymous confessions; or for reasons in-between these two poles? Which secrets will live to breed, and which will not be able to survive in the truly rigorous environment of the internet? What are the actual desires of principal internet users, when authorship and curatorial decisions are confidential and democratic.
TECHNICAL DEVELOPMENT OF SITE:
Preliminary work one the site began with the creation of a Perl script to handle basic secret input, re-telling, rating, etc. This script writes to a database, an example is shown below.
Anatomy of a secret (database structure):
| index number | # of re-tellings | rating (%) | secret body |
| 1 | 4 | 4 | "Once, when I was 19..." |
| 2 | 21 | 54 | "I just found out that..." |
The more complex scripting involves the breeding of secrets. To spawn new secrets by combining existing texts, the PDC incorporates intelligent processing methods, based on the cutup method pioneered in the 1960's by Brion Gysin and William S. Burroughs. Separately chosen texts are spliced together, sentence-by-sentence or phrase-by-phrase, to achieve a whole which is artistically greater than the sum of the parts. This technique figured prominently in Burroughs' earlier works (Naked Lunch, Soft Machine, Nova Express), and we have modeled in previous web-sites (http://erewhon.mt.cs.cmu.edu/Cutup/).
We intend to explore various existing language technologies in support of the PDC, for example: text categorization and clustering (to select similar texts for user feedback and reproduction); text extraction (to identify relevant semantic concepts as a basis for categorization and reproduction); and natural-language understanding and generation (to ensure fluency of PDC-generated texts). To date, we have implemented an initial clustering method, based on a frequency model of salient word occurrences, and an initial "reproduction" method, based on existing cutup techniques (implemented in Perl). As the database of secrets grows, we intend to focus and refine our methods, taking advantage of the emerging characteristics of the text corpus to deploy more sophisticated natural language processing techniques to improve the performance of the system.
Examples of current re-assembled secrets (using our first generation algorithms.)
Whomever shall read that I received a public park and like to take small boys back and like to know. You gotta believe me to maintain my petting zoo.
Now I've never felt so bad you would do things. These video loops drive my social drugs in the choice and think of my favorite things. Whomever shall read this.
NEED TO FIND A women was fun, with moist, yet another.
AUTHOR HOMEPAGES:
Paul Vanouse -- http://www.contrib.andrew.cmu.edu/~pv28
Lisa Hutton -- http://art-slab.ucsd.edu/ARTSLAB/LisaHutton/LLHpage.html
Eric Nyberg -- http://www.lti.cs.cmu.edu/ehn/