At The Overlap

My daughter saw a floppy disk in a museum display case recently. It was sitting up there on a timeline of data storage examples, including cassette tapes, CDs, and USB sticks, among others.

“What is that?” she asked.

“How do I explain this properly?” I thought, while feeling slightly strange. And old.

I remember using floppy disks during my university degrees. This was simply the way we saved data from a number of very finicky laboratory instruments.

My daughter, who is currently in elementary school, has obviously never used a floppy disk (nor a finicky old lab instrument). But she probably already knows what that little picture of a floppy disk means, to an extent.

She recognizes that it’s the little symbol that appears when you want to save something.

So she understands the symbol…without understanding the object.

When the object disappears

The floppy disk became the save icon for a fairly obvious reason. When Susan Kare designed the original Macintosh icons in the 1980s, saving to a disk was actually what people were doing. So the icon was a picture of the thing involved in the action.

Over time, how we store data has changed.

Floppy disks have all but disappeared. Hard drives became standard, USB sticks came and went, and cloud storage has now become king.

And that little floppy disk image? Still holding on.

What’s interesting is that Adobe has actually tested different alternatives over the years, including images of hard-drives, CDs and clouds. But the floppy disk was still more recognizable as “save”, even among people who had never used one.

At some point, the icon stopped being a picture of a physical object and became just a symbol for the action.

Because we don’t look at it and think, “Ah, a floppy disk”, do we?

We think, “Save.”

Why is that a frying pan?

What about that little magnifying glass you see everywhere on the computer? You know, the one beside a search box.

You know what it means. But did you ever think about the fact that a magnifying glass doesn’t naturally mean “search?” It means, well, magnifying.

So what happens when we don’t all see the same thing? Funny enough, Microsoft discovered just how non-obvious that association could be when it was developing the icons for Windows 3.0. (wow that feels long ago!)

One overseas office looked at the search icon and asked why Microsoft had used a frying pan!

So apparently the magnifying glass was not a universal symbol for searching. The most likely explanation was that cultural references that helped make it meaningful, such as a detective looking for clues (Sherlock Holmes anyone?), weren’t shared across the globe.

I like this story, not just because it make me smile, but because it exposes something we normally don’t notice.

A symbol doesn’t come with its meaning attached. We learn the meaning.

And once enough people learn the same meaning, we sometimes stop noticing that we ever really learned it.

We are surrounded by borrowed meaning

We are surrounded by numerous examples of this, but perhaps the most daily version is the shopping cart.

The physical cart has simply been moved into a different environment.

And computers are filled with similar examples. Folders on a computer? There is no physical paper inside it, but the idea of a folder gives us somewhere to put things.

We don’t have to understand everything beneath the hood to be able to use the computer. We get enough of a rough idea by what these symbols tell us.

Seems great. But there can also be a catch. Especially when the thing we borrow brings some of a different meaning with it.

When a shortcut becomes a mental model

A lot of my work around AI involves something that sounds pretty simple: helping people understand how they can actually incorporate AI into their work. And this is where the idea of borrowed meaning gets interesting.

For example, you might think “what exactly is an AI system?”

A tool?

An assistant?

A copilot?

An agent?

It might feel like it doesn’t matter much. That it’s “just” semantics.

But words can result in people forming certain expectations. For example, if I call something an assistant, I might expect it to wait for instructions. If I call it an agent, I might expect it to take action on its own. A copilot suggests that I am still sitting beside it, making the important decisions.

Like what we might ask an AI to do and how much we are willing to let it do for us.

Of course it’s not just about terminology. It’s far more complex than that. For example, AI systems aren’t all doing the same thing. An AI with access to a few simple tools and one that can search, write to other systems, make decisions between steps and carry out a longer task can genuinely have very different capabilities.

So it can go in different ways. Our mental model affects how we use the technology, while changes in the technology can make an old mental model more or less useful.

It is interesting to see how much research has actually been done on this topic, looking at AI and metaphors, symbolic language.

One example is a study published back in 2020, where researchers gave 260 people conversational AI systems described using different human metaphors, like a toddler, teenager and experienced butler (1). Even when the AI system behaved in the same way, the metaphor changed how people perceived its competence and warmth. And this actually affected peoples’ decisions whether or not to keep using it.

Words can be very powerful. Especially when it comes to mental models. When we reach for something we already understand to help us make sense of something we don’t.

The difference is that with AI, the thing we’re trying to understand is changing at a rapid pace.

Back to the museum

My daughter doesn’t need to know all the details of what a floppy disk was, for the save icon to function.

She may eventually learn. Probably in a history class. Or maybe she won’t.

But that’s the strange thing about symbols. You don’t necessarily need to understand where they came from to understand what they mean.

My daughter may never use a floppy disk.

But she already knows what it means.

---

1) Khadpe, P., et al. (2020). “Conceptual Metaphors Impact Perceptions of Human-AI Collaboration.” Proceedings of the ACM on Human-Computer Interaction, 4(CSCW2), Article 163.
https://doi.org/10.1145/3415234