ai creative writing

ai creative writing

AI Creative Writing

1. Introduction

AI research in creative generation has evolved to focus on understanding creative behavior and simulating human creativity, particularly in the cognitive science modeling tradition. Early work was dominated by knowledge engineering, using large representations of domain knowledge to align with computer science traditions of the 1970s and 1980s. Research has explored creative generation in specific domains, e.g., music, and also the development of systems that generate across a multitude of domains. Simulation of narrative has become a sub-field in its own right, with a specific workshop series at ACM conferences. Work in this area has often aimed to cross the divide with humanities research in the construction and analysis of narrative. In tandem with the wider AI research, there has been an increased focus on generating systems on interaction with end users. This has involved work in the generation of mixed initiative interfaces and interface agents. More recently, there has been an emergence of creative generation potential on the web, e.g., the generation of personalized art, poetry, or stories for individual users.

Artificial intelligence (AI) has the ability to evolve its own language and make up stories which could be dismissing the need for human storytelling at all. AI story generation isn’t a new concept and has been around for many years. You can think of the attempts by Demeester in the 1970s. AI story generation system can be a very complex process. It can be broken down into a few steps such as story generation, story authoring, and story comprehension. There are some AI systems that can generate stories that are generated by a retelling of a knowledge representation of a story rather than creating the story from scratch. An excellent example of this is Minstrel, which is designed by Carbonell, a program that generates stories under hard constraints of consistency and generality. Once the story has been generated, the AI system may then go onto the story authoring state, which could involve converting and translating the story into different mediums.

2. Benefits of AI in Creative Writing

The most fundamental way that AI can benefit creative writing is simply by providing writers with new ways to generate ideas. This can be considered as an automation of the classic strategy of using writing prompts. These prompts can be very varied, from specific tasks like “write a story on this topic,” to more conceptual or abstract suggestions. AI technologies can offer a form of this automation across a wide range of generative models, from simple rule-based systems to more complex, statistically driven methods. For example, the use of Bayesian networks or Hidden Markov Models, which provide a probabilistic framework for the generation of text based on an underlying model of word choice or grammatical structure, can be used to generate sentences around a particular theme. At a higher level still, AI technologies such as case-based reasoning systems can function to identify similar examples of a text in a known database and use these to make inferences for a new piece of writing. This might be useful for a fiction author seeking to clarify a narrative plot.

Perhaps the area in which AI technologies have progressed most significantly, and certainly one of the most exciting areas, is in the potential application of AI to various creative tasks. In contrast to the rules and logic-based expert systems which dominated AI research in the 1980s, many AI researchers now utilize the statistical methods and probabilistic framework provided by machine learning algorithms, in an attempt to make sense of less structured data. This has opened up new possibilities for AI to take on open-ended tasks which have traditionally been the sole domain of humans. One such area is creative writing. AI technologies have the potential to act as an aid to human creativity across a wide range of tasks and unleash new realms of possibility for human authorship.

3. Challenges and Limitations

When comparing the original and the computer-generated stories, readers rate the original stories as being of higher quality and clearer, but they offer no rating for creativity and originality, suggesting that those that were auto-generated were unclear and of lower quality, but their originality and creativity were equal to those of the human-generated stories (Kukich, 1983). Kukich discusses this as a challenge of evaluation – that there was no significant difference between the ratings of seen and unseen stories and this suggests that human readers and evaluators might be more critical of computer-generated stories, implying that a true assessment of generation quality should result in higher ratings for auto-generated stories that mirror original stories. In contrast, however, Rowe et al. explored the cognitive processes of those reading automatic and human-generated weather forecasts and discovered that whilst readers did identify the automated forecasts as being of higher quality, they were less memorable and led to poorer task performance, suggesting that automatic generation may be worse than human generation in some cases, and so the challenge of comparison is understanding whether automatic generation should match or best human generation, and in what areas. Task performance was also an issue in the study by White and Frederiksen (1998) where subjects were found to perform worse when using comprehension strategies to read an automatic summary and then answering questions on it. This suggests that more complex types of writing may actually be worse comprehended when auto-summarized, posing another challenge as to what types of generation should be used for certain types of material.

4. Ethical Considerations

AI-generated writing is a revolutionary development that has grown substantially in the last five years. In fact, OpenAI now has an API which this very text has been written in. AI writing has the potential to be faster, more comprehensive and more mainstream than human writing, expanding the capabilities of the professional landscape. It is then pertinent to investigate AI writing and its implications within the context of professional and creative writing. Published by Oxford University Press, Ai Med aims to publish the insights and experiences of medical professionals. Created by the GM and Board of Directors, this publication targets medical professionals, primarily internists, who use artificial intelligence (AI) and machine learning to discover patterns and analysis to improve patient care. The emergence of AI-generated text has the potential to affect the viewership and contribution of clinicians to this site. This case study will employ a CASP Checklist tool to examine the validity and significance of an article, Comparing AI vs. human breast ultrasound. The CASP tool is in the form of a questionnaire that has the ability to appraise the quality of a randomized control trial; the article in question, comparing a double reading of breast ultrasound (an independent variable) by AI with the current practice of a single reading by a human radiologist followed by a double reading comparing the first reading by the same radiologist and the second reading by a different radiologist at a later date. This is to detect and characterize ductal cell carcinoma in situ (DCIS) and/or invasive breast cancer in symptomatic women, where the population is well defined (listed inclusion and exclusion criteria). The aim is to gauge the sensitivity and specificity of machine learning with the detection of breast cancer in contrast to human reading. The exercised RCT can be found on the Ai Med publication and is held in regards to Dr. Ian W Turnbull’s Reddit post. This case study will provide a comparison to critique the validity and significance of the article in question.

5. Future of AI in Creative Writing

The progress of AI writing is a double-edged sword for human writers. On one hand, it offers massive possibilities for language generation and translation software, adding to the developments of AI in speech recognition and virtual reality. The visually impaired could “read” a book purely through writing it. Steps are already being taken to replace human-authored content with autogenerated content. News articles, for example, have been written by AI in an attempt to reduce bias and save money on wages. A large enough development in AI creative writing could see a slump in demand for writers as their machine counterparts become more proficient and cheaper to hire.

AI writing in the present day relies on supervised learning using a large corpus of human work to improve its own. However, the end goal for many scientists is a general-purpose unsupervised machine, which, given a certain topic or idea, will write in just the same way a human author would with no external input.

Certainly, a fascination for Veale’s envisioned “machine purposes of language” has led to a period of rapid growth in the AI’s application to creative writing. This growth has been catalyzed largely by the development of neural network models which emulate the pattern recognition of the human brain, allowing for data learning and thus more human-like output.

AI has become increasingly relevant to a number of creative fields, with writing being no exception. In his 2012 paper “The Automation of Creativity,” Tony Veale considered the implications of AI in creative writing and concludes that “the power of natural language generation is bringing the worlds of human language and machine computation ever closer, to the point where machines may one day do useful things with language, not just for the sake of communicating with humans, but for their own machine purposes.”