Automated storytelling is a process involving the use of artificial intelligence (AI) to create written stories. Although automated storytelling is far away from being able to construct genuinely creative and insightful novels, the process has been steadily improving to handle basic applications.Content Continues Below
Automated storytelling tools use a combination of AI, machine learning and big data to create content. Typically, automated storytelling can be used for anything from writing headlines, financial reports and weather updates, to screenplays or short stories. A practical use of automated storytelling today would be to use the process to ‘write’ a more technical headline or report and have human writers focus their time on more creative stories that may be less structured.
Automated storytelling starts with collecting large amounts of data into a database. This data can include information such as hundreds or thousands of different stories or headlines. Tools such as natural language understanding (NLU) will then scan through the data parsing it into more structured data. Templates are created by humans so the AI can replace the information in the template with its own. The templates can be lower-level, meaning they can be simple points where data values are replaced by the AI (for example, fanatical reports), or higher-level templates intended for more complex and meaningful writing. Natural language generation (NLG) is used to automatically generate text-based summaries from the database.
To tell longer meaningful stories, automated storytelling may try to structure the stories put into its database by pattern. Templates can be created by patterns that commonly appear in literature and storytelling mediums. Stories usually share characteristics such as details on settings, characters, character development or conflicts, all aimed to elicit an emotional response. With these details, a person can create a template for a story so the AI can parse through the stories in a database to create something unique. This can be done, although we are still quite a ways away from AI creating something equivalent to what a human can write.
Examples of automated storytelling
Researchers at the University of Vermont developed a machine learning program to identify emotional arcs in stories that appear in text, video and scripts to then create arcs of its own from. Similarly, a collaboration between MIT’s Lab for Social Machines and McKinsey’s Consumer Tech and Media team lead to the creation of a machine-learning model which uses deep neural networks to parse through film, television and online video content. The machine learning model can then determine either positive or negative emotional content found in structures such as in plot, characters, close-ups, music and dialogue.
Journalistic outlets have also started using automated storytelling. For example, The Washington Post has started using an AI called Heliograf to write more than 850 articles. Heliograf has been used for large-scale data-driven news event coverage.
In the past, Heliograf has been used to help with coverage of the 2016 Rio Olympics, US elections and even high-school football games in the Washington D.C. area. In the 2016 Olympics, Heliograf was used to generate short, multi-sentence updates, event schedules, medal results and medal tallies; which were then posted on The Washington Posts’ blog and Twitter.