A self-help book generated with machine learning algorithms.
“The solution is the source of pain and pleasure to the situation to ensure that they don’t do anything or on the way we evaluate the way they love to realize that they don’t want to realize that they have to do it.”
What Will Happen In Your Metaphor? is a self-help book generated with the Char-RNN machine learning algorithm, using Tony Robbins genre classic “Awaken The Giant Within” as source material.
The title of the project and the full content of the book, including chapter headings and motivational quotes are generated from the output of a trained software model.
The intention of the project is to not only to demonstrate the power of text based generative frameworks, but also to document the progress of such a model as it learns, with each paragraph in the text generated by another sequential training iteration. The book begins with an almost incomprehensible stream of characters but quickly transforms into fully legible, grammatically correct aphorisms.
In this sense ‘What Will Happen In Your Metaphor?’ exposes an uncanny transition, as a more clear voice gradually emerges from a soup of letters, asking the reader to consider exactly where the meaning of the words resides.
I saw my ‘role’ in the process as that of an editor rather than author, setting the parameters for the training process the model and formatting the output.
Char-RNN is a fascinating approach, developed by Andrej Karpathy and extensively outlined in his blog post: The Unreasonable Effectiveness of Recurrent Neural Networks.
It processes large volumes of source text to make predictions on a character by character basis, using multi-layer LSTM’s, or ‘Long Short-Term Memory’ a type of neural network architecture particularly well suited to this task. The learning model quickly builds – essentially from scratch – an appreciation of how digital text is structured and is before long out-putting seemingly meaningful novel sentences. Part of what is so interesting for me about Char-RNN is that it demonstrates how susceptible we are to infer from meaning from a text, if the grammar and structure it’s presentation seem to match our expectations.
The process of training a model involves establishing constraints about just how it should learn. Most of the generated book is the output of the model as it is ‘learning’ and more inclined to take risks in how it generates characters. The final chapter represents the output of the model when it has been fully trained and , leading to a higher degree of accuracy.
The source for the project was “Awaken The Giant Within”, a foundational and perhaps seminal text within the self-help movement, for which it’s famous author Tony Robbins is a noted figure. If the original book has a central theme beyond it’s stated aim of being a guide to “Take Immediate Control of Your Mental, Emotional, Physical and Financial Life” I would say it’s about the power of language to transform our thinking. An intuitively written guide for experiencing the world in a different way by actively reframing how we describe it.
NB: I have been asked about the copyright issues that are raised by such a project, a question that is both very interesting to me and ultimately unresolvable at this time. My best guess was that you might share copyright along a split, 40% to Tony Robbins, 40% to Andrej Kaparthy et al, and 20% to me.