Outlook cloudy: Designing for uncertain times
How weather apps attempt to provide certainty-as-a-service despite uncertainty-as-a-reality
Originally published August 29, 2023
We were sipping tea and talking shit about the weather. Well, weather apps. We’d all agreed: the apps are not all right. They’re getting worse. They give you one forecast in the morning, another in the afternoon, and then both are wrong in the end. We were tired of ending up too hot or too wet or missing a beach day. We went on and on until, finally, someone let rip the inconvenient truth: “It’s not that weather apps are wrong, it’s that our climate is off the fucking rails.” Oh… right. I guess it’s easier to rail on an app than it is to come to terms with the realities of the climate crisis.
What we crave is certainty in this increasingly uncertain world.
The uncertainty goes beyond the weather, of course (though the weather is fucking bonkers). The American project is on fire. Our bodies are microplastics. Everything is AI-ified but no one knows how it works. The UFOs are here — they’re just bored?
And so we crave certainty from the products we use, the things we read, and the people we listen to. We want certainty-as-a-service without nuance as a nuisance. We expect products to have done the homework, but to just share the answers.
To me, the weather app examples illustrates a few dynamics at play:
- Technology creates an illusion of certainty;
- More data doesn’t guarantee better predictions;
- We expect digital products to simplify and offer certainty;
- Yet products don’t have a UX or vocabulary for uncertainty
Grappling with and communicating uncertainty is even more important when you consider we’re barreling towards a world of LLM-powered probabilistic products.
The problem with predictions
A major earthquake is possible next week in San Francisco. At least according to a post on X:
Panic! at the group chat ensued. Well, sorta. Californians are used to doomsday predictions — and to ignoring them. Just days before, Hurricane Hillary landed not with a bang but a whimper. But the tweet felt legit, with all its data — a heat map! percentages! — and its stilted emergency-alert language. The account is followed by nearly 60k users and claims that its “earthquake forecasts are based on research from (2005 to 2023).” Maybe, just maybe, we’d finally gathered enough data to predict our shaky futures? Nah.
When I clicked through to Twitter (X?), a “context” module assured me that the tweet was textbook misinformation:
It got me thinking how rare it is to see a platform labelling fallibility or embracing harm reduction. This is one of the few examples of what I’ll call Uncertainty UX. Uncertainty UX is how a product signals its confidence level to the user. In this case, the module is saying that what you’re reading “might not be true.” It’s funny because that framing presumes that most of what you do read on Twitter is true.
The feature is a move in the right direction, but it’s plainly limited: the tweet itself can be shared and disseminated without the extra context. I’d bet Twitter’s algorithm treats it the same as any other tweet, too.
Twitter is nobody’s angel, to be clear. It’s maddening how long it took them to do anything about misinformation. We had to live through years of Trump’s lies before the company ever regulated him — or shipped actual features to tackle misinformation at scale. It’s all too-little-too-late, as many users have publicly left the platform for more trustworthy competitors, like the decidedly PG Threads or dorky Bluesky.
Principles of an Uncertainty UX
So, what would Uncertainty UX look like in practice? Here are a few principles that could work across different categories:
- Acknowledge the veracity of content, e.g., “This claim was not fact-checked” [Social media]
- Identify predictions as such, e.g., “This forecast is a prediction ” [Weather ]
- Cite sources, e.g. “This prompt was generated from XYZ sources” [ChatGPT]
- Note variability, e.g. “This step was confusing for X% of users” [Recipes]
- Consider user outcomes, e.g.: “This shirt fit Y% of users as expected” [Shopping]
What’s important is that a product acknowledges its own limitations and uncertainties throughout the user experience. In practice, this would require adding new UI elements — more data, more disclaimers, more friction. It would require businesses to acknowledge that all users don’t have the same experiences.
But businesses — and their product designers — don’t want that. They want their products to be as clean and straightforward as possible. We know that a cleaner UX means more users, means higher conversion, means more revenue. This conventional wisdom isn’t wrong, but it’s also an example of short-term thinking.
As I’ve argued before, it is a smarter business strategy to consider the long-term outcomes for your users. Consider the weather app. If the weather app were more open about its own fallibility — if it gave users the most truthful approximation of what the forecast could be and how often it’s wrong — then, over time users would trust it more. Apple’s Weather App doesn’t do that, and it’s losing users for it. They actually recently bought and shuttered their best competitor, DarkSky, a popular crowdsourced weather app. Darksky’s UI provides a good example of how to communicate the inherent uncertainty of weather forecasting:
Dark Sky elaborates on uncertainty by offering the user a great deal of more data and more context. It’s nowhere near as “elegant” as Apple’s single-page summary, but it’s more accurate and more useful to the user.
In many ways, Apple is the real culprit behind this dangerous impulse in product design. They invented modern product design and idealized interfaces that are simple, easy, “elegant”; their design embodies certainty. The most successful products follow their playbook, taking complex tasks — like banking or grocery shopping or watching videos — and rendering them utterly frictionless in a digital interface. When a product is “well-designed” it’s as “delightful” and painless as a water slide. But we know now that frictionless activities have all sorts of externalities. To take an extreme example, a Robinhood customer killed himself after thinking he traded away $730,000. We need more friction on everything everywhere all at once.
Uncertainty UX in an age of probabilistic design
I’m harping on this theme because so much of our current computing paradigms are about to shift towards probabilistic design. All of the apps I’ve mentioned above rely on traditional product development with deterministic outputs. Uncertainty UX is crucial in a world where most consumer products will soon be powered by Generative AI and LLMs, which are inherently probabilistic and uncertain. We can only expect more and more products to offer services that contain uncertain outputs.
ChatGPT is probably my favorite example of a company that buries uncertainty in a chic Apple-fied interface. Here’s what I have to say about that.
***
To get back to the weather, I can’t decide if forecasting is a good example or an extreme one. The reality is that we’re living in a climate crisis. Weather predictions can only get worse; weather apps can only get more wrong. So, given that, do you collapse complexity as Apple does, or do you share more nuance as DarkSky does? At what point does a sunny outlook move from willfully ignorant to just dangerous?
— XML
