A design space for AI writing tools (CHI’24)
TLDR: A new paper maps the human-centred design space around AI writing tools, plus an interactive tool to explore the literature behind the design space:
Mina Lee, et al. (2024). A Design Space for Intelligent and Interactive Writing Assistants. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’24), May 11–16, 2024, Honolulu. ACM, New York, NY, USA, 34 pages. Open Access Preprint: https://arxiv.org/abs/2403.14117 [Interactive tool]
Writing is thinking. My career-long fascination has been with how computers can support thinking, spanning digital tools for writing and diagramming ideas/arguments. We clarify our thinking by seeing if we can externalise our thinking coherently. When we can’t, it’s a signal to raise our game, or switch tack.
The GenAI writing invasion. Everywhere we look in the world of digital writing, AI is wriggling its way into the apps, from the longstanding, fully featured tools like Microsoft Word and Overleaf, to the niche products like Grammarly, to the myriad new kids on the block targeting specific commercial sectors with the promise of “writing productivity” from GenAI [one of many review listings]. Here at UTS of course, we’ve been deploying and refining our own AcaWriter web app since 2015, building our understanding and research-informed evidence around what works with both students and teaching teams.
Educational implications. From an educational point of view, we are now grappling with the profound questions that Generative AI raises around how we teach and assess writing. The fact that the documents that until now served as plausible proxies for intellectual work, may have been co-authored with/ghost-written by a machine, forces us to ask how and in what ways our students need to demonstrate their writing competency. I’ve reflected on this in The Writing Synth Hypothesis and other GenAI posts. Assessment reform for the age of AI is the challenge.
Design Spaces. A software design space is a framework that clarifies what the key options are that designers can choose from, around key elements of the digital artifact. How many ways are there to provide the user with critical functionality? My PhD 1988-92 was working with Rank Xerox EuroPARC on Design Space Analysis, a form of design rationale capture (Graphical Argumentation and Design Cognition) — so it’s been fun to return to this now.
The AI writing design space. So — what is the shape and size of the design space for AI writing tools? In a new paper we map that space, and will present this in May at CHI24, the leading international conference on human-centred computing. Kudos to Mina Lee and the others in the lead team who coordinated the team of 36 authors who mapped this space, reviewing 115 papers from HCI and NLP, covering the different levels of such tools.
Figure 1: Our design space for intelligent and interactive writing assistants consists of five key aspects—task, user, technology, interaction, and ecosystem—that are interconnected and interdependent. Within each aspect, we define dimensions (bold texts) that represent fundamental components of the aspect and codes (examples associated with bold texts) that represent possible options for each dimension. When necessary, we group semantically relevant dimensions together within each aspect and use a prefix to denote the group name; the interaction dimensions are grouped by user, user interface (UI), and technology; likewise, the technology dimensions are grouped by data, model, learning, and evaluation.
An interactive tool helps explore the literature behind the design space:
