So this summer when I’m not teaching, I’ve been working on a book based on one of the HCII classes, Designing AI Products and Services. I’m writing it with the two of the faculty most responsible for the content and research that went into the class materials: John Zimmerman and Jodi Forlizzi. Today, I pretty much finished the draft of the first chapter.
It’s an interesting challenge for me in that I’ve never had co-authors, and I’ve never written a book where I wasn’t the true subject matter expert—John, Jodi, and some of the PhDs they’ve worked with over the years are. I’m really just the writer—I think of myself mostly as a ghostwriter. I’m taking a lot of the raw material they have (class notes, slides, research papers, etc.) and turning them into a book that is geared towards professionals in the field working on AI products and services. Having written a few books for design professionals, the shaping and writing of those is something I’m good at, and it fits well with my Professor of The Practice title and responsibilities.
It’s also helping me get deeper into the topic. In order to explain something simply, sometimes you have to learn a lot about it and then pair it down to the essence. Which, having never coded an AI model, for example, I learned a bunch about it so I could figure out how to explain it to others who’ve never done so either.
It’s also a challenging topic because the field changes so rapidly now. In March, for example, there was a big AI headline literally every day. Trying to write a book that’s helpful now but still fresh in 2024 when this (hopefully) comes out is really threading the needle.
We started by coming up with a target audience (designers and product managers mainly, but also anyone on a team working on these products and services), then did several drafts of the Table of Contents. A lot of the TOC is from the structure of the class, which I tweak and massage as I write. I plundered most of it from the in-class presentation slides, then supplemented what I didn’t know (or frankly understood) with research as I write. Then John and Jodi review my drafts to add or edit things or to simply tell me what I’ve gotten wrong or what isn’t working. And then we had some others read the chapter over to see if what we wrote makes sense to the target audience. In this way, we churned through a first draft of the first chapter. Next, we have an introduction and a second chapter to write before approaching our preferred publisher.
Here’s a short excerpt from Chapter 1: The AI Innovation Gap:
Imagine you have a spare bedroom you’d like to rent out on AirBnB. How much should you charge for it on any given night? $80? $120? $300 if there is a convention in town? This was the dilemma of AirBnB hosts until 2016 when AirBnB introduced Smart Pricing. Airbnb Smart Pricing is an AI pricing tool. It predicts the maximum a host can charge and still have their place rented. It takes into account factors such as location, size, amenities, number of views a listing receives, quality of customer reviews, seasonality, local events, competitor pricing, historical booking patterns, days of the week, and more. Smart Pricing is designed to be hands-off, so hosts don't have to frequently and manually adjust their prices. Instead, the AI automatically adjusts the nightly rate, helping to ensure that hosts are earning the most they can for their listing without them having to keep track of the AirBnB market and everything happening in their town. While some hosts think the AI is too biased towards putting “heads in beds” (maximizing occupancy over income), many seem pleased with the results—especially because it’s free for hosts, unlike other competing services. “We use it and love it,” said one host in Durham, North Carolina on the AirBnB Community Forum. “It has increased our booking price on average of 7% and we're booked 92% of the time. While I used to agonize over the price and if I would get a booking, now I let Smart Pricing do it for me.”
Interestingly, successful AI projects like this are usually more accidental than intentional. In fact, almost all AI projects fail. Most fail quietly before launch. For the few that do survive and launch, some fail spectacularly, becoming media horror stories about AI or about the competence of a specific brand.
AI is terrible. AI is amazing. AI will change everything. AI is overhyped. AI will become sentient and kill us all. This is the hype the media, VCs, and tech bros are feeding us daily. Meanwhile, a quieter revolution is happening just under the surface of all this noise. AI is now in everything from the media we consume to the cars we drive to the healthcare, education, and legal advice we receive. AI has had a big and transformative impact across many industries. It can direct people through an inhuman quantity of information, such as surfacing a small set of Netflix movies you might want to watch tonight. While AI can generate images, video, and text, it can also co-create alongside users, such as by suggesting songs for building Spotify playlists. AI can make micro-content, tailored to personalized instruction far beyond what a human teacher could provide because it would simply be far too time-consuming to do. It can predict the weather and generate email responses, which are fine even if they aren’t 100% accurate. AI can replace tedious, mundane, or repetitive tasks like filtering spam from your inbox.
None of these rely on AI being all knowing or even having high quality performance. AI can be sucessful with moderate performance as long as it is valuable and low risk—for users and organizations. AI is not superintelligent—in fact it’s often quite dumb. But it’s also quite powerful when used appropriately in the right circumstances. This book is about finding and taking advantage of those circumstances in a deliberate, systematic way. It’s about improving your outcomes when you create new products and services that rely on AI.