Algorithmic Uncertainty
An Mertens, Gijs de Heij, Manetta Berends
A research trajectory to develop a poetically-observing book that is in dialogue with a machine-learning model.
An Mertens → writer, member of Constant vzw, Brussels
Published at De Bezige Bij, 2012
Gijs de Heij → Open Source Publishing (OSP)
Manetta Berends → freelance designer
Machine Learning
Machine learning is a technique of recognizing patterns, to develop knowledge on the basis of data. It's a way to automate the process of transforming data into information. The belief that this is possible is the basis of technology.
- long history, currently a new popularity wave
- different types of machine learning:
rule based
supervised learning
unsupervised learning (also called deep learning).
Each being based on another type of human-machine collaboration.
Cqrrelations (2015)
"Cqrrelations was a work session lasting from 12 till 23 January 2015 in deBuren, inviting data travellers, writers, numbergeeks, programmers, artists, mathematicians, storytellers, and other tech creative souls to explore the world of digital non-relations, desnalysis, blurry categorisations and crummylations in the Big Data that shapes our daily reality and language."
Cqrrelations (2015)
The Annotator (2015)
http://www.cqrrelations.constantvzw.org/1x0/the-annotator/
Uncertainty Detected
An Mertens & Gijs de Heij (2016)
Uncertainty Detected
An Mertens & Gijs de Heij (2016)
Uncertainty Detected
An Mertens & Gijs de Heij (2016)
Algolit
a workgroup around algorithmic literature, free code and texts
i-could-have-written-that (2016)
Graduation project Manetta Berends, Piet Zwart Institute, Rotterdam
i-could-have-written-that (2016)
Graduation project Manetta Berends, Piet Zwart Institute, Rotterdam
i-could-have-written-that (2016)
Algorithmic Uncertainty
Reporting on moments of uncertainty in a text-based neural network machine learning model.
structure
Stimuleringsfonds Startsubsidie to test the feasibility of the project, and develop a first dummy of the book to be able to contact publishers.
- context
- visualisations
- fiction
moments of uncertainty
To detect the gray borders of accuracy & certainty in a neural network model.
- discourse (scientific world, media, technology companies)
- problem task (approach)
- training process (converting text to vectors)
- test phase (validating techniques)
- algorithms (statistical models & linear algebra)
- the code of the software
- development of the model (compromises that need to be made)
Research method
- developing an understanding of the technique
(1.) to speak a shared language with machine learning experts
(2.) being able to validate moments of uncertainty
(3.) and translate it into a language that is understandable for a broad audience - software experiments
- interviews and continuous conversations with experts
Positioning
Our relation with the technique:
ranging from being fascinated & surprised, to annoyed & frightened.
The angle of the book:
[en] poetic / contemplative / observative / artistic / functional / speculative / specular / signaling / meditating / reflecting / activating / learning / studying / a meeting book
[nl] poetisch / beschouwend / observerend / artistiek / functioneel / bespiegelend / bekijkend / signalerend / mediterend / reflecterend / activerend / loerend / bestuderend / een ontmoetend boek
- een artistiek beschouwend boek / an artistic contemplative book
- een artistiek bespiegelend boek / an artistic specular book
- een poëtisch beschouwend boek / a poetic contemplative book
- een meervoudig beschouwend boek / a multi-sided contemplative book
- een meervoudig observerend boek / a multi-sided observational book
- een bespiegelend observerend boek / a reflective observational book
- een bespiegelend reflecterend boek / a specular reflective book
context
criteria
- concrete application
→ situated
→ partial and incomplete
→ constitutive (knowledge/information is framed by rules that make something happen or exist, what would not exist without these rules) - Dutch / Belgian roots
- a topic that appeals to a general public and reflects in substance on the motivation to work with machine learning techniques
- community use
- possible to publish our code and data under a free & open license
- Mike Kestemont, assistant professor in the department of literature at the University of Antwerp in Belgium
- Juliette Lonij, Research Software Engineer at the KB Lab of the National Library (Koninklijke Bibliotheek), Den Haag
Asibot, Mike Kestemont
An additional chapter to Isaac Asimov's book "I, Robot", written by Ronald Giphart & Asibot, a neural network text generator, commissioned by Nederland Leest.
Asibot, Mike Kestemont
Frame Generator, Juliette Lonij
Frame generator, Juliette Lonij
Visualisations & software experiments
- framework level, space for code experiments (Tensorflow, Theano, PyTorch)
- computer technical level (structure of a network)
- mathematical level (linear algebra & statistics)
word2vec (algolit)
word2vec's graph generator
word2vec (algolit)
saving data objects, such as temporary dictionaries
word2vec (algolit)
writing logfiles
word2vec (algolit)
excluded words
Fiction
Translation by analogy written from the perspective of the model.
A story from the Algorithmic Forest (draft)
planning
- prototype / dummy
- contacting publishers & partners
- funding proposal for a Stimuleringsfonds project subsidy in 2018
Exhibition Algoliterary Encounters
Maison du Livre, Brussels
Friday 9 - Sunday 11 November
workshops on Saturday & Sunday
public lectures on Saturday evening
Monthly Algolit sessions in Brussels