Embracing AI as a collaborator in data visualisation design
Guest post by Andy Kirk
Rapid advancements in Artificial Intelligence have instigated introspection across the data visualisation field. Though swathes of AI discourse are characterised by hype and gimmickry, if you cut through the noise, it’s clear we are facing a significant new era of technological progress. It’s time to evolve, or die, as the saying goes.
But what should our relationship with AI be, particularly with generative AI? To what extent could our authentic craft be usefully augmented by an artificial one? How will it help and with what activities? In what ways might it pose risks?
I’ve spent time considering the elements of the data visualisation process, as detailed in my book, where AI could helpfully collaborate with data visualisers, augmenting our efforts to make tedious tasks more efficient and creative acts more ambitious. Rather than focus on the concerns I have about AI, I’m choosing to frame this list around a more positive, perhaps hopeful, perspective about ways I would appreciate enhanced support:
DATA
Data foraging: Receiving smarter suggestions for places to source bespoke data, especially unstructured or non-digitised data, perhaps for niche topics for which data may be fragmented and requires extensive searching.
Qualitative sources: Unlocking greater potential in the extracting of unstructured qualitative data from compelling material like text, printed artefacts, imagery, audio, or video sources.
Data handling: Having an auto-generated transcript of a given data handling process would be invaluable, noting all the relevant steps from initial gathering, to cleaning, to calculating, and through to transforming it ready for charting.
CREATIVITY
Colour themes: Although there are specific rules for certain colour-data associations, there is plenty of scope for subjectively selecting specific colours across all aspects of a visualisation. These could be inspired by suggestions based on subject matter and/or tone, with mockup variations to enable you to consider how they’d look.
Thematic inspiration: Using smarter searching, I would appreciate surgically sourcing relevant examples of visualisation designs for inspiration about a similar subject matter or to find other approaches to tackling common analysis challenges (eg. “how have others shown changes over time and location simultaneously but a static/printed output?”)
Signature style: Could AI learn your design techniques, style, or preferences, in order to suggest refinements each time you work through a visualisation process? This would be especially relevant when working in a corporate or commercial context with formalised design guidelines/branding rules to observe.
DESIGN
No-code interactivity: Enabling the creation of interactive visualisations without the need to code. Tools like ‘Make real‘, offer a promising sketchy-point-and-click interface to automate interface designs. Alternatively, using language-led prompts to describe the features you’d want - skipping even the step of vibe-coding - would be game-changing.
Text annotation: Smart-generated chart titles and captions, and suggestions for their styling and positioning within a chart’s spatial constraints - even if only as a first draft text - would be a huge time saver. Refining would be much quicker than doing from scratch. Extending this to generate copy for user guides, chart-reading instructions, data definitions, method statements, and FAQs - as well as auto-translating into other languages - could be invaluable.
Story building: AI assistance for developing designs to aid amplifying certain narratives/insights, not just in helping to select the right chart type, but in the broader design and styling of the overall communication. As demonstrated impressively in this video from Andy Cotgreave.
EVALUATING
Output previews: An enhanced way to envision a visualisation design as previewed in a wide range of different output formats, dimensions, and settings would be very useful, to see how outputs look on mobile/tablet, laptop/desktop, meeting room screen, home TV screen, through to conference halls. Extending this to simulating how they might look - or suggesting how they should look - in different settings like a printed document on a moving train, a newspaper graphic read outside a sunny cafe, a poster-sized map being viewed in a busy under-lit station at night.
Functional evaluation: Evaluating a nearly-finished work by occupying the mindset of a target audience to judge its effectiveness, perhaps with different viewer profiles - from vis-expert-level, domain-expert-level, and to lay-person. This could assess the design’s perceived trustworthiness, to surface any potential issues about the reliability of the data material (sources, assumptions, inclusions or exemptions?) and check the integrity of design (any distortions, misleading chart constructions?). It could also perform an accessibility check to assess suitability for different levels of visual, physical, technical, perceptual, and interpretive accessibility.
Stylistic evaluation: This will always be extremely subjective, even for humans, but a further valuable assessment would be formed around perceived visual appeal: is it beautiful, is it visually unified, is it memorable, is it sufficiently attention-grabbing, does it feel authoritative? All matters of contextual variation, and utterly subjective, but worthy of assistance if available.
Andy Kirk is an Independent Data Visualisation Expert delivering a range of professional services to clients around the world as a data visualisation design consultant and prolific trainer. He is a sought-after speaker, a four-times published author, the editor of visualisingdata.com, and host of the ‘Explore Explain’ video and podcast series. He is the author of Data Visualisation: A Handbook for Data Driven Design, now in its 3rd edition.
