I appreciate the nuanced take here. LLMs have definitely made certain tasks easier, and therefor let me build ou features I wouldn't have had time to before (i.e. the theme picker example from the blog post)... but I am increasingly finding myself susceptible to the "tumbleweed effect" you describe, where an LLM can start to feel like more of a crutch than a tool.
I feel that Bruce Dawson's blog falls into the same/similar category, though he hasn't had a great technical post in a while. I wonder what multithreading bug he'll find next.
Great read. The part about art needing to be created by a human particularly resonated.
There is a fundamental difference between creativity and novelty. Novelty is empirically measurable and is something an AI can do. Creativity is an expression of the human experience. It can't be outsourced.
If you really don't see any difference between "evoking emotion" and "emotional manipulation" then nobody will change your mind on this.
And like I wrote about, you can have very precise charts that can be extraordinarily manipulative.
The whole "charts should be about hard data" thing always sounds nice, until you realize that every visualization is an abstraction and brings in some human bias. You are always picking the scales, whether or not to use color, how to aggregate the data, etc. There is no such thing as a neutral chart.
Also... there are data visualizations outside of science. Of course science should be more precise than something like data art.
No reason those need to be mutually exclusive. Posted this in another thread, but I don't see a good argument against this project (https://www.dear-data.com) being both data visualization and being an act of personal expression.
Hah, so I actually went back and forth on that. The original studies around visual inference also included axis labels on the bar chart. The fact that you CAN label the axis on a bar chart that way is actually an argument people make not to use pie charts.
Author of the article here. Great points, thanks for reading and commenting.
I think that data journalism is a great example of that "put it in someone's brain" idea.
I also think there are all kinds of places we're seeing more novel approaches in too: data art, physical data viz (e.g. museum exhibits), visualizing huge datasets, etc. The Climate Stripes project, maybe the most famous data viz of the last couple decades, was made by a climate scientist.
But I definitely agree that this "more experimental than rigid" side of data viz doesn't capture the majority of the field, nor should it!
