AI-powered robot can read novels and write poetry

Tiếng Việt

AI-powered robot can read novels and write poetry

A new book series by RMIT academics Karen ann Donnachie and Andy Simionato carries a cautionary label on each copy: ‘Warning! This book was not designed by humans.’

Well, the originals were.

The new versions though - featuring Haiku poetry and eerily cute images where the book text once sat - were autonomously generated by Artificial Intelligence (AI).

Dr Donnachie and Dr Simionato’s reading machine - on display at the Melbourne Art Book Fair in Australia - uses computer vision and optical character recognition to ‘read’ books.

It then uses machine learning and natural language processing to select a poetic combination of words on the page to form a Haiku, while erasing all other words.

Dr Andy Simionato [pictured left] and Dr Karen ann Donnachie’s [pictured right] AI-powered reading machine reconstructs novels as Haikus. Image: Peter Clarke

Next, the machine searches Google for an image to match the text and sends itself off to an online printing service. Creation complete.

The reading machine has transformed more than a dozen books already, from Margaret Atwood’s The Handmaid’s Tale to TrumanCapote’s Breakfast at Tiffany’s.

The results have been puzzling, often cute or even hilarious, and at times profoundly poetic.

“We run several versions of each book and they always produce different results,” said Dr Simionato, a lecturer at RMIT’s School of Architecture and Design in Australia.

“Each formulation unlocks new puzzles and combinations that have always been there but which we could never see.”

The reading machine's take on pages 16-17 of Truman Capote's Breakfast at Tiffany's.

Their idea with this project is to get people talking about poetry in the age of AI.

"This, really, is our attempt to grapple with the future of the book, an object we’ve had around us for so long and that’s so central to many cultures, that’s changing so fast," Dr Donnachie said, a lecturer at RMIT's School of Art in Australia.

“How the written word, and culture more broadly, is produced and read has now shifted so profoundly with social media shorthand, memes, emojis and AI summarising massive amounts of text on our behalf every day.

“So what does the creative process of poetry look like in that context?”

Dr Donnachie [pictured left] and Dr Simionato [pictured right] are interested in the future of books, after radical changes to content creation and consumption. Image: Peter Clarke

RMIT Vietnam Information Technology Lecturer Dr Dang Pham Thien Duy said while AI for content generation is becoming more and more popular, the implications that may develop will be interesting to observe.

“If a painting is generated by AI, who should hold ownership over the painting? The AI itself, the software developer creating the AI algorithm or the one who uses that algorithm?

“What will happen to journalism when robots can write news stories as proficiently as human journalists? How can we increase the readers' trust in auto generated news? These are questions we need to think about,” Dr Duy said.

According to Dr Duy, the use of AI to generate new content and natural language processing are “exciting trends” in the AI field and clear examples of how AI can contribute to other disciplines outside computer science.

However, he believes there’s still a long way to go until AI can be on par with human intelligence.

“The fact that AI is able to read yet unable to fully understand text and generate meaningful contents is a motivation for scientists.

“A lot of efforts still need to be made to develop AI that can get closer to artificial general intelligence,” Dr Duy said.

Story: Michael Quin and Ngoc Hoang

Science and technology

18 March 2020