Dear friends, Alberto here.

Today we’re launching a new course in the Open Visualization Academy, titled Tidy Data, Tidy Infographics. It’s an elementary overview of the design process for a simple static infographic.

If you are an absolute beginner, this course, along with my intro to visualization, is for you. I go over some basic data wrangling —transposing from wide to long format— and design with RawGraphs and Adobe Illustrator. This new course is, in part, a gateway to our lengthier and more comprehensive tutorials about those tools (check them out: 1, 2).

In the new course I briefly demonstrate how generative AI (GenAI) can be integrated into a design workflow; other courses in the OVA go deeper into that. In one of the videos I discuss a tentative personal heuristic about GenAI: I'd consider using it for a design task…

A. If you can clearly envision your desired design and the decisions and process that would lead to it if you were using non-AI tools instead.

B. Even better: If you—alone or as part of a team with sufficient time and resources—could recreate that same design with non-AI tools.

Xaquín G.V, someone whose judgment I trust since we met two decades ago, pointed out that B sounds a bit repetitive. Maybe, although what I’m trying to reflect is the subtle and blurry differences I see between ideation and execution, thinking and making.

Say that I wish to design an interactive graphic that requires complex JavaScript+D3. Before I even think about GenAI, I should be able to envision the design—its components and how they should be arranged, which I can draft in a static mockup—and also the reasoning process that would lead to it. This is the bare minimum to consider integrating GenAI in the workflow.

To actually create that visualization (without GenAI), I’d need to collaborate with people who know much more than I do about the intricacies of code. That’s B, which is an extension of A.

When I shared the heuristic—which I’ll likely keep tweaking in years to come, as my views about this stuff are a work in progress—in social media, someone asked what happens if we use GenAI for ideation. I agree with Frank Elavsky’s eloquent arguments, so I recommend not to do it, at least too early in the process. Make an effort of imagination and creativity first; what your brain is capable of may surprise you. Trust it.

In any case, let’s say that we do use GenAI for ideation, and that it offers us an alternative design that we like but hadn’t thought about. The heuristic still stands: at a minimum, we should be able to justify the design, the reasons for every element in it (“Why is the baseline of this graph at 0?” “Why this color palette and no other?”), and the steps that we, not the AI, would take to make it real.

In the early 2010s, Bret Victor delivered a series lectures —Drawing Dynamic Visualizations comes to mind— where he demonstrated prototypes that to me felt like a dream come true: “Wouldn’t it be great if we could draw code, instead of writing it?” I wondered.

That’s why I love tools such as iNZight, which is essentially an R graphical user interface (GUI), and I’ve always been fond of the idea of natural language interfaces (NLI), which could translate my musings into code for visualizations. GenAI fulfills that dream. We can now use iterative language instructions to design charts.

The word iterative is key because it implies that we shouldn’t take what the tool spits out at face value —that’d be cognitive surrender; see below— but, instead, we ought to deliberately guide the tool, step by step, to accomplish our envisioned goals. As the good folks at The Economist data and graphics team wrote recently, “make sure that you know how to count and code, at least well enough to steer your AI agent of choice in the right direction and catch when it leads you astray.”

In other words, if you know what you are doing, and if you are able to verify the outputs the GenAI delivers —arguably a massive if, but that’s a topic for future articles,— it can be useful. If not, it’ll be destructive. Eva Sibinga has a pretty great article about all this at Nightingale, and MIT has recently published a report about AI in education that offers sensible advice applicable to many other areas.

At its core, my tentative heuristic above is a warning against the phenomenon that this Twonks cartoon depicts:

In a paper titled “Thinking — Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender”, University of Pennsylvania researchers Steven Shaw and Gideon Nave explain the difference between cognitive augmentation and offloading—consciously delegating specific tasks to automation—and cognitive surrender, which is a tendency to “defer judgment, effort, and responsibility” to devices, “particularly when that output is delivered fluently, confidently, or with minimal friction.”

This reminded me of the difference between “centaurs” and “reverse-centaurs” that Cory Doctorow has discussed in recent articles and in his latest book. A centaur is a human being who has sufficient knowledge and skill to decide what to automate, how, when, and for what purpose. A reverse centaur, on the other hand, is a human who becomes (or who is forced to become) part of a process led by an automaton; a cog in the machine.

Whenever you see a designer rhapsodically singing their praises of the latest shiny tech toy, there’s a high chance they are a centaur. And, Doctorow says, centaurs tend to be oblivious to the corrosive effects that the tools they love may have on those who don’t have the privilege of being centaurs yet.

“This, I think, is the explanation for the paradox of AI: the AI users who are being immiserated and precaratized by bosses who have been convinced to fire their colleagues and pile their work on the terrorized survivors of the layoffs hate the AI, because it makes their life worse in every way. Whereas the people who choose when and how to use AI — the centaurs — are only using AI to the extent that it is useful, and throwing it away when it’s not. They may make poor choices about the AI, but those choices are theirs, they are not imposed from on high."

Back to Shaw and Nave, they write:

“Unlike cognitive offloading, which is typically strategic and task-specific (e.g., using GPS to navigate), cognitive surrender entails a deeper transfer of agency. Whereas cognitive offloading is a strategic delegation of deliberation, using a tool to aid one's own reasoning, cognitive surrender is an uncritical abdication of reasoning itself. It reflects not merely the use of external assistance, but a relinquishing of cognitive control: the user accepts the AI’s response without critical evaluation, substituting it for their own reasoning.”

Based on numerous —albeit anecdotal— first-hand observations, I share this concern.

—

I recently learned that Tony Iommi, founder and guitarist of Black Sabbath, is about to release a new album. He’s collaborated with one of my favorite singers, Jørn Lande, so I couldn’t be more excited. Here’s the first single, World Alone (Iommi is 78, and still rocking hard):