鈥淥nce upon a time, I dreamed of being a children鈥檚 author,鈥� says Selma Storey, 鈥渂ut inspiration is like fairy dust. I鈥檝e tried everything, even random word generators, but I can鈥檛 conjure a children鈥檚 title from 鈥榣everage鈥� and 鈥榓sbestos鈥�. Help!鈥�
It鈥檚 debatable whether machines can be truly creative. But they鈥檙e certainly smart. Artificial neural nets are thinking machines modelled on the connections in our own brains. They can spot patterns within data, allowing them to predict earthquakes, steer self-driving cars and beat everyone at the board game Go.
Advertisement
There are lots of open source versions you can experiment with. You just need a few lines of code and some training data. To start, I gave my neural net a list of 200 children鈥檚 book titles, scraped from .
It got to grips with form, even if the results lacked flair. 鈥淭he Stories and the Things鈥� doesn鈥檛 scream bestseller to me, and 鈥淭he Magic Magic Mole Mole鈥� is a little repetitive.
Time to up the data set. This time I assembled 1180 titles. After a single pass, my neural net showed improvement, even if 鈥淭he Story of the Story鈥� was a bit recursive. After a few more passes, it was finding its own dry sense of humour, offering up 鈥淥liver and the Neverend鈥� and 鈥淭he Fantastic Profity Monster鈥� 鈥� something all publishers are no doubt looking for. Following an afternoon deep in thought, the neural net came up with 鈥淭he Adventures of Dunno and His Friends鈥�, 鈥淗ans the Mole Brother鈥� and 鈥淏ad Tomato Wench鈥�. Not bad.
In a final push, I spent a day compiling a huge data set of 3600 titles. Feeding this literary smorgasbord to the neural net, things turned sinister. It dreamed up titles like 鈥淭he Bear Made Me鈥�, 鈥淚 Want to See Christmas鈥� and 鈥淏ary Rots in Homewold鈥�. Still, 鈥淎vocado Baby鈥� is sure to appeal to millennial parents.
In the end, though, a number of passes through the long list got my neural net generating just what I had hoped for: titles like 鈥淭he Mysterious Church Stairs鈥� and 鈥淭he Adventures of the Wild Dinosaurs鈥�. With infinite writing prompts, how could you ever be lost for words again? Now, does anyone have the number for Random House?
- Let us solve your domestic trifles. No problem is too small for our resident tech wizards. Tell us what鈥檚 bothering you and we鈥檒l build a fix.
- For more Makes visit
