Data Storytelling, Generative AI
Scientific Focus on the Paper: Why is AI not a Panacea for Data Workers?
My thoughts on the scientific paper Why is AI not a Panacea for Data Workers? An Interview Study on Human-AI Collaboration in Data Storytelling, by Haotian Li, Yun Wang, Q. Vera Liao, and Huamin Qu.
Data storytelling is challenging and time-consuming since it involves various steps, such as data collection, analysis, visualization, narration, and presentation. Therefore, it is natural to wonder if and how AI can assist data workers in this task and the benefits and drawbacks of human-AI collaboration in data storytelling.
Working on the possible integration between data storytelling and AI, I have noticed that there are very conflicting opinions on the matter. On the one hand, some argue that AI is the panacea for every problem, including data storytelling. On the other hand, there is a particular fear of introducing it into the data workflow and storytelling.
For this reason, I decided to delve deeper into the problem and see if there was something in the scientific literature. I found this fascinating article, which you can find at this link.
The article tries to answer the following questions:
- Where would humans like to collaborate with AI in the data storytelling workflow?
- How would humans like to collaborate with AI?
- Why and why not do humans prefer to collaborate with AI?
The authors conducted an interview study with 18 data workers from both industry and academia and asked them about their current data storytelling practices and their expectations and concerns about collaborating with AI.
Where would humans like to collaborate with AI in the data storytelling workflow?
First, the authors identified the storytelling workflow as consisting of three successive steps:
- Planning — in this phase, the core message is decided, the data is collected, and the outline of the story is compiled
- Execution — in this phase, the pieces of the story are prepared, then integrated, and, finally, a style is given to the story
- Communication — you share the story.
Each step is made up of several steps. For example, the planning phase involves message decisions, data collection, and an outline definition.
Respondents prefer to use AI in the data collection phase and execution phase. It could also be helpful in the case of one-way communication, for example, to adapt stories to various media (video, texts, etc.), but not in two-way communication when there is direct communication with the audience, such as in the case of a meeting.
How would humans like to collaborate with AI?
The authors identified four ways of using AI in data storytelling:
- Creator — the AI carries out the task completely (for example, it collects all the data without the help of humans)
- Optimizer — AI does not execute the task but refines the human-generated content
- Assistant — AI assists the human in executing a task
- Reviewer — AI evaluates human tasks and suggests possible problems or improvements.
Respondents prefer to use AI as creators and assistants.
Why and why not do humans prefer to collaborate with AI?
Interviewees were very interested in the use of AI in data storytelling. For example, AI can be used to perform repetitive tasks. Furthermore, it is low-cost, so that it can provide suggestions faster than a colleague. Finally, compared to a human, it never gets angry, so it is possible to ask him anything without getting tired.
In reality, the interviewees also demonstrated a certain skepticism towards AI, as it often does not understand the context (the article was written at the beginning of 2023 when it was not yet straightforward how to do fine-tuning), has limited communication capacity, and still requires skills on the part of the human to use it correctly.
Final thoughts
Personally, I found this article very interesting, as it is one of the first scientific articles that brings together human-AI collaboration in data storytelling.
This article contributes to the research field of human-AI collaboration, especially in data storytelling. It highlights the challenges and opportunities for designing AI-powered data storytelling tools. It also raises some critical questions for future research, such as balancing the trade-off between automation and humans. Indeed, integrating AI, especially generative AI, into data storytelling will be a slow process. We’ll see :)
I appreciate the authors’ efforts and look forward to reading more of their work. :)
What do you think? In your opinion, can data storytelling be integrated with AI? Write what you think in the comments!
Thanks for reading this article, and see you next time!
