Our journey towards the reinvention of writing for games moves ahead at great pace and we are for the first time able to hire more people into our project. For now we’re looking for a front-end and a back-end person to come along for the ride. More below, because first comes a little recap of the last weeks.
It was Tuesday evening that we both went to bed feeling a little bit under the weather. By Friday we knew the virus1 had found its way into our bodies. Yet, we had also recovered enough to run our second online workshop on Experimental Writing with AI. Together with 20 wonderful guests we explored the depths of language models, encountering stochastic parrots, and arson, and tea:
V V V V Arson? V Tea? V
The above poem is one of the completions created by the tiniest of GPT-Neo models after being fed a diet of dialogues. We searched, but never found out, how arson connects to tea. Yet, we accept that the two concepts are potentially poetically intertwined. Much more reasonable is the following reaction of Dr. Delrio to his own death:
“This morning’s news broke in with the reports of the deaths of Dr. Delrio, but the stricken man did not come to a room. He was unconscious, and a policeman questioned him about the circumstances of his involuntary breakdown.”
Exploring the space between H.P. Lovecraft, Jane Austen, and their own writing, our workshop participants uncovered pearl after pearl:
"A note on the subject of the future: I am fancied.
I'm thinking about the future, and I am fancied!
I am fancied!"
Beyond the entertainment value of wonky text generation, question-answering often hints at truths that are already expressed in a text. You ask the machine “Who was the murderer?” and it sometimes hints at a possibility you have not consciously thought about. If your model was trained on your own data, that possibility is extrapolated from your own thoughts by advanced statistical means. In a way, language models make the unconscious conscious, the hidden overt. It’s a curse when they bring the worst biases mankind has to offer to the front. It’s a boon when they uncover the murderer in your murder mystery before you consciously decided who it is. A lot of co-authoring with machines will have to deal with this challenge – and a key to creating good tools for co-creation will be to find ways to turn bias into a strength.
Our third Experimental Writing with AI workshop will be on March 25th. We still have a handful of seats available. We think that we will change the format of the workshop after that one, moving to using our own software instead of publicly available tools but let’s see.
Finally: Jobs! Jobs! Jobs! The spaceship2 we launched in is ready to take on two more crew thanks to the generous support of Innovationsfonden. If you are passionate about inventing and experimenting in the augmented writing space, and can wield the powers of contemporary web development, you might be exactly who we’re looking for. We have the budget to spend a few more months running exploratory workshops, developing experimental software, and putting together a credible pitch for an actual product. For that, we need autonomous individuals who want to create a future of games writing with us, where humans and machines work together in peace and intellectual prosperity. Apply if that sounds like you.
It feels awkward to write this newsletter when a country just invaded another one right here in Europe, displacing and endangering friends and colleagues. Yet what can we do but try to build a better future? Solidarity with Ukraine in these troubled times. If you want to glimpse into the multicultural past of the Black Sea region, we both can highly recommend Olga Tokarczuk’s Books of Jacob. There can be a better world.
I wrote a totally objective review of the Covid experience on Twitter dot com. It scored 7/10.
The painting above is by Victor Ash.