The Plan Is to Improvise

Dear Felienne!

Oh, the irony! Last week, I complained about researchers using AI to do their work and called them imposters, and this week, I made the mistake of using AI to finish my own homework! After completing my notes, I asked Gemini for feedback, and ultimately ended up using an idea it gave me. I felt weird about it ever since. Both like a hypocrite and, indeed, like an imposter. I have considered re-recording the homework for the podcast and removing the AI’s suggestion from this text, thus pretending it has never happened. But maybe there is some value in publicly admitting this.

Why did I resort to AI? It probably has to do with conflicting fears and competing feelings of discomfort. Yes, on the one hand, I feel discomfort about using AI and have many fears about using it. To just mention a few: I am concerned about being inauthentic, violating my core principles, deskilling, sustainability, supporting big tech, producing average-sounding garbage, and losing autonomy. But at the same time, now that these tools are available, I feel ever more insecure about my writing. Crazy thoughts enter my brain such as: What if I put something stupid out there, and I could have prevented it with AI? Or: Maybe I am just lazy. Shouldn’t I put in this minimal effort and fix those problems that I could have easily found with the help of AI? Or even weirder: Considering that AIs generally love everything people produce, better check if it likes my output as well. What if my text is the one exception that AI admits to hating, despite me asking if there is anything good about it five times in a row?

Yes, AI has served as my comfort blanket when leaving my comfort zone! My use of AI has to do with our podcast and my homework being slightly out of my comfort zone, which is exactly where and when it gets tempting to use AI. But maybe I can put a positive spin on this and see this ‘incident’ as yet another reminder that I should strive to accept discomfort, uncertainty, and imperfection over trying to achieve comfort, control, and perfection. So here it comes, my imperfect homework…this is one more of these instances where I did not really do what you assigned for the homework, but went off course instead.

An Employee Card is No Loyalty Card!

I re-watched Suchman’s brilliant videos about the computer scientists trying to use a copy machine (people can watch them on your blog)! Naturally, I got curious if I could use the copy machine in front of my office. My ego needed to know if I could do better than Turing Award winner Alan Newell (who struggles in Suchman’s videos!). Kidding aside, I was genuinely wondering whether modern copy machines have gotten any easier to use.

I set out to print three student theses. The first challenge was to get the documents from my computer to the machine. Our university has a website for this, which features a designated drop area where you can drag your files. Unfortunately, if you drag a PDF just a teeny tiny bit outside the indicated area, the browser opens the file and leaves the printing website. So I needed to try again, more carefully dropping the PDFs the second time.

Next, I went to the printer. But, of course, the printer had no idea who I am. How do I convince the printer that it’s me, Hanna? Apparently, you can type your university ID and password on a digital keyboard on a tiny digital screen. But this keyboard looks so different from my normal computer keyboard that my muscle memory doesn’t kick in, so I had no chance of recalling my password. (Yes, I am one of those people who remember their passwords in their hands, rather than in their head.)

Luckily, you can also authenticate yourself with your employee card. In theory, that’s easy. Unfortunately, I had the wrong mental model about how this works. That’s because our employee cards look like the loyalty cards from our supermarket: both have this black-and-white barcode on one side. Hence, I thought they are used in the same way. At the supermarket, the checkout terminals have a little scanner that reads the barcode. Then they know it’s you; you are probably properly profiled for advertisements, and all data about you is sold — but at least you also get your discount and some Air Miles in your name.

Surprise, surprise, using your employee card at the printer doesn’t work this way. Here, the process is based on RFID, so you need to hold the card really close to a special spot on the front of the printer. The barcode is irrelevant. (You could draw over it if you wanted to!) So while holding the card to the copy machine itself is easy, it’s still kind of confusing when you have a mental model from the supermarket’s loyalty card in your head and try to apply that.

Once the machine knew it was dealing with me, I got offered the expected overview of (1) printing, (2) scanning, and (3) copying functionalities. I chose printing and got presented with an overview of my three documents. I wanted to print all of them, but an option to select all was nowhere to be found! That changed, however, once I had selected the first document, and a new “select all” button suddenly appeared out of nowhere. I clicked it, considered the many colorful graphs my students tend to use, and hence went with the machine’s default settings: simple A4, two-sided, color prints.

The printing itself was super easy (just press the biggest button). Yet, there was a problem: my copies came out in black and white. I double-checked the machine’s default: it is set to color! Wait, what? I had actually found a bug! Or maybe not a bug but a money-saving feature? I wonder, how many people will actually print a document again, just because it was printed in black and white?

Naturally, I could not accept this unexpected behavior and needed to find a workaround! (Yes, I could have read the documents in black and white; but it became a matter of principle… Me against the machine!) What I tend to do in such situations is to put myself in the machine’s shoes and imagine how it might ’think’ and work. Also, I try to think of the programmer and consider what kind of obvious mistakes they could have made. What came to mind is that there could be a disconnect between the default settings in the machine’s frontend and its backend. In other words, if you do not change anything, the printer interface will initially say “color”, while the printer internally is set to start out with “monochrome". So I decided to toggle this setting from color to monochrome and then back to color to see if that helps. And indeed, there you go. To print in color, you need to switch the printer from color to monochrome and back to color!

Overall, I consider this a success! But I think printing would not have worked out so well without putting myself in the programmer’s shoes and trying to emulate in my mind how the machine works. In fact, if I had simply treated the printer like a human, this entire process would have failed completely. And this is also one of the cool topics in Lucy Suchman’s book¹. I actually have found a quote related to this. It’s essentially her starting premise. On page 189, she writes:

[…] the project of building interactive machines has more to gain by understanding the differences between human interaction and machine operation, than by simply assuming their similarity.

If I apply this to the situation at hand, yes, there’s a difference between how humans interact with humans and how humans interact with copy machines! Understanding this difference will help us design better (copy) machines. And apparently there’s still a need to do so, because our copy machines remain kind of weird. This weirdness is not the result of these machines not properly acting like humans. The machines are weird because they are not acting like proper machines!

How to Build Better Copy Machines

As someone designing interactive software and teaching about it, I naturally also asked myself if some problems I encountered could have been prevented in the first place. Of course, user studies would have been a great thing to do! But I also think many problems could have been prevented by doing what we in HCI call a cognitive walkthrough.

A cognitive walkthrough is a usability inspection method invented by Wharton et al. (1994)², which can be used to find problems in interfaces. It assumes that people approach a new interface with a goal in mind of what they want to achieve. They then search the interface for actions that might bring them closer to that goal. If they find an action that looks promising, they choose it. Then, they evaluate whether they indeed made progress toward their goal. The cool thing is that researchers turned this model into four questions that can help spot problems in interfaces. One just needs to ask those four questions for each tiny step a user has to take.

The questions have been reformulated in many ways. But if you ask me, it boils down to asking the following questions at each and every step a user needs to take:

1. Whether users, at this step in the process, will actually try to do the right thing. For example, will they even try to authenticate themselves to the printer? Essentially, it’s about considering what people have set their minds on achieving at this point in the flow.

2. Whether users will perceive the action or element that allows them to do what they have to do at this step. For instance, will they see that there’s this special area at the front of the printer where they can hold their employee card? Maybe this area is not clearly marked, and they will not notice it at all? So this question is about perception, i.e., what humans perceive with their senses. (One remark: Please consider blind people and, generally, people with disabilities.)

3. Do users understand that this element or action is what is needed now to complete the current step? Maybe they see this special area at the printer, but they do not associate it with authenticating themselves. For instance, maybe they think holding the card to the printer will allow them to get some printing discount? So this question is about the users’ understanding: whether they make the link between the available action and their intention.

4. Once users do the right thing and complete the step, do they (get feedback that allows them to) know they succeeded? Based on the machine’s reaction, is it clear that users are closer to their goal? For instance, after holding their card to the printer, does the user know they are now authenticated and thus have made progress?

Whenever someone creates software where a user has to carry out some step(s), one can simply ask these four questions for each tiny step the user needs to take. As soon as one gets a “no” as an answer to one of the questions, one knows one has a problem. And then one can (try to) fix it. This is something every programmer can do themselves. (In the official walkthrough, it is done with a team of experts, though! That’s even better, of course!) It’s very cheap, and it allows one to deal with a lot of low-hanging fruit before starting user tests. (One should still do those, of course!)

Simply put, I believe this process is about trying to put oneself in the users’ shoes. Do they have the right thing on their mind, perceive the things they need to perceive, understand the things they need to understand, and get the feedback they need to pursue their true goals? (And of course, “logging in”, “typing a password” or “reading a manual” is never the true goal of a user. At least, I do not know anyone who ever thought to themselves “I really feel like logging in somewhere, yeah, let me log in to a few websites right now…”!) If the programmers of our copy machine had put themselves in my shoes, then I likely would not have needed to put myself in their shoes to handle the machine’s weird quirks.

In the Heat of the Moment

Something I find beautiful about this cognitive walkthrough is how it ties in with Lucy Suchman’s ideas of actions and learning being situated and happening in the heat of the moment and being somewhat improvised. The cognitive walkthrough, like Suchman’s book¹, assumes that users will not read a manual! They will not memorize steps and then execute the steps as described in some user guide. Instead, they come up with their actions along the way, as they go along in the process. In this sense, the walkthrough perfectly aligns with Lucy Suchman’s view!

Yet, maybe — and that’s what the AI pointed out to me — this cognitive walkthrough idea is also fundamentally different from how Lucy Suchman sees human-computer action. The cognitive walkthrough treats the user as someone who follows clear, predetermined steps. People always determine the next subgoal, search the interface for actions that match this subgoal, carry out those actions, and then evaluate if they made progress. Then they start all over again, essentially looping through four predefined steps. And that idea of users having clearly outlined steps beforehand is very much opposite of what Lucy Suchman describes in her book. After all, she sees no concrete, rigid plan that exists in advance!

So maybe the AI is right, and the cognitive walkthrough fundamentally conflicts with Suchman’s ideas. But then again, maybe there is no contradiction here at all. Maybe we humans can have a plan that simply states that we’ll work out a better plan when the time comes? Maybe we can have meta plans that describe not what to do and how to do it, but rather how to figure out what to do and how to do it. Such meta plans are no rigid plans, and they do not seem at odds with Suchman’s ideas at all?! For me, the cognitive walkthrough is about people having such meta plans rather than concrete plans. Also, a loose meta plan is what I use to write my homework posts. I have no idea what to write at the beginning and no rigid plan in place. Instead, I plan to figure it out by using a mix of thinking, writing, reading, deleting, and revising my text. The plan is to improvise! Okay, that got longer than intended again. I guess that can happen when you improvise. Talk to you soon.

Cheers,

Hanna