Project Update: Reflecting on Data Collection Through a Focus Group Session

I would be willing to bet that I have participated in far more focus groups than the average person. When I was hustling to pay off my student loans in the 2010s, I discovered that the city of Atlanta had a slew of market research companies that would pay me for my time and opinions on a number of different products: restaurant foods, scratch-off lottery tickets, pilot episodes of television shows and commercials, deep sea fishing excursions, NASCAR race events, household appliances. Those are just some of the focus groups I remember participating in.

There is a bit of a dirty secret to my former success in selling my opinions to market research companies though, and I wouldn’t say that I am proud of it now. I had to lie…rather, I chose to lie in order to qualify for most of the studies that I participated in. If a market research company’s screener was recruiting for subjects to taste a new menu item for a fast-food chain that sold seafood, then they needed to find people who ate at these establishments. They would call all the people on their database list of potential subjects and work to fill quotas that typically required some diversity in the group of subjects (age, race, gender, income, etc.), but the subjects all had to have one thing in common: they had to be a target audience for the product, likely users or consumers of the product. The market research companies didn’t want vegans in their Chickfila focus groups, unless they were testing out a new non-meat option or something.

But there is a critical flaw in the way these focus groups operate. It is a problem that opens up an avenue for unscrupulous people like me (old-me, anyway) to easily exploit their system. The company that wants the focus group data typically outsources this task to third parties. Company X hires a market research firm to collect data, and that market research firm often relies on another market research company to assemble the groups of participants. (At least this is how I saw it working in the 2010s.) The real breakdown seemed to occur with relying on the initial screeners to produce a group of people who fit the specifics that Chickfila or their hired market research firm decided they wanted data from. The call screeners’ job was to find participants, and back then, I quickly realized that doing so was not always easy for them to do. Sometimes I would get through a 10-20 minute screening process (answering questions about demographics and purchasing behavior) only to have the screener say, “Sorry, you don’t qualify” or “You do qualify, but I already have enough males in their 30s.”

Experienced screeners worked more efficiently and could figure out if I qualified much faster by asking the right questions in the right order (obviously diverting from their scripted questionnaire). And the occasional desperate-for-subjects screener would be intentionally leading in their questioning: “Are you sure you haven’t bought a scratch-off lottery ticket in the past week?” I think the screeners were even incented financially to fill groups, earning bonuses for quantifiables that encouraged them to ignore quality. Either way, once I caught on to the different motivations of the parties involved, I started to play the game. I would be intentionally vague in my screener call responses, allowing “good” screeners to lead me to the desirable answers. Then I would show up to the focus group, participate as much as I could, offer legitimate and genuine feedback whenever possible, and then collect my check or gift card. Rinse, repeat.

My dishonesty involved lying about mostly my consumer behavior. It wouldn’t have been easy to lie about demographic information because much of that would be obvious when I showed up to the focus group and they verified my date of birth and address. My race would have been relatively obvious too, although, one time my wife (accomplice and fellow liar) did get herself qualified for a self-tanner product and showed up as the only non-orange person at the group! I guess she didn’t think that one completely through, but she still got her honorarium.

After participating in a focus group, my wife and I would often discuss the performance of the focus group leader. As educators, we both knew what leading a group discussion entailed and we marveled at just how incompetent some of these market research professionals were at their jobs. There were also the rare few who were excellent, but the bad ones were always worth gossiping about. In my hubris, I almost always thought I could do better. Considering my level of experience as a participant toward the middle and end of my multi-year run as a frequent focus group imposter, I was probably somewhat right.

When it came time this semester for me to collect data for my current UX project, I thought about my experience with focus groups and decided that doing a dozen interviews at once would be an efficient way to go about this step. It was time for me to see if I really could run a focus group better than the professionals I critiqued years ago.

I was prepared. I had two pages of questions all related to the generative AI system that my students-turned-informal-research-participants had used just days prior. I also had 15 years of experience leading discussions on all types of subjects with diverse groups of students, including high schoolers, college students, and even incarcerated men in a state prison. I was ready to prove myself as a market research genius.

Ultimately, I failed at achieving true focus group leader glory for two reasons:

#1. It is difficult to keep the participants on topic. My list of questions was well thought out and logical. One question or set of questions logically led to another, but my research subjects were not privy to this planning and therefore jumped ahead and backwards many times. This made it harder to elicit the feedback I was seeking. Fortunately, I recorded the session so I could sift out the data, but I believe there were lost opportunities and lost data because of the way I failed to control and redirect the respondents at times. It was a balancing act because more feedback is generally preferable to no feedback, and I didn’t want to shut the respondents down.

#2. I found it extremely difficult to avoid leading my respondents toward my own opinions and observations regarding the user experience of the product (ChatGPT). Hubris strikes again. I could not remain an impartial enough recorder of responses. I saw opportunities to seek additional feedback from respondents that would confirm my own opinions, and I couldn’t help myself.

Despite these shortcomings, I did collect data that will allow me to better assess the user experience of the target product. If I could do the focus group again, I think I would start with more open-ended questions and allow the discussion to go where the respondents lead it, rather than trying to rigidly stick to the list of questions I developed.

Book Review: AI and UX: Why Artificial Intelligence Needs User Experience

AI and UX: Why Artificial Intelligence Needs User Experience, by Gavin Lew and Robert M. Schumacher Jr. 1st edition 2020, Apress. EBSCOhost, search.ebscohost.com/login.aspx?direct=true&AuthType=ip,shib&db=cat06559a&AN=ggc.996791144102945&site=eds-live&scope=site.

Why did I choose this book?

My current research interests revolve around all things generative AI, especially its applications for writing and teaching. I am trying to learn as much as I can, and I think an understanding of the UX aspects related to AI is beneficial. This led me to seek a UX book that might provide insight into how AI systems and products are designed with end users in mind.

Because the release of ChatGPT in November of 2022 drastically altered the entire AI landscape, I was hoping to find a recently released book on UX and AI, but I was unable to find one that looked like legitimate scholarship. This is not entirely surprising since it takes a while to research, write, and publish. So instead, I settled for the most recent book on the topic I could find that looked reliable. There were not many options to choose from, but this book, despite its 2020 copyright date, proved a good read for someone like me who is neither an expert in UX or AI (yet).

Who are the authors?

The authors for this book are Gavin Lew and Robert Schumacher Jr. Instead of trying to summarize their already relatively short biographies in the book, I will include them in full:

Gavin Lew has over 25 years of experience in the corporate and academic environment. He founded User Centric and grew the company to be the largest private UX consultancy in the United States. After selling the company, he continued to lead a North American UX team to become one of the most profitable business units of the parent organization. He is a frequent presenter at national and international conferences and the inventor of several patents. He is an adjunct professor at DePaul and Northwestern universities. Gavin has a Masters in Experimental Psychology from Loyola University and is currently the Managing Partner of Bold Insight, part of ReSight Global, a globally funded UX consulting practice across North America, Europe, and Asia.

 Robert M. Schumacher Jr. has more than 30 years of experience in academic, agency, and corporate worlds. He co-owned User Centric with Gavin from its early stages until it was sold to GfK in 2012. While at User Centric, Bob helped found the User Experience Alliance, a global alliance of UX agencies. Also, he founded User Experience Ltd, a UX agency in Beijing. He is co-founder, co-owner, and Managing Partner of Bold Insight, part of ReSight Global, a global UX company. Bob was the editor of and contributor to The Handbook of Global User Research (2009). He has several patents and dozens of technical publications, including user interface standards for health records for the US government. He also is an Adjunct Professor at Northwestern University. Bob has a Ph.D. in Cognitive and Experimental Psychology from the University of Illinois at Urbana-Champaign.

Basically, we have two writers with plenty of UX experience (related to technology and other fields) and backgrounds in psychology. It might have been nice to have an author with more of a computer-science background paired up with someone who knows the psychology behind UX, but these two authors also have a long-established working relationship which enhances their ability to communicate throughout the book.

Summarizing the Chapters and Some Highlights:

In the preface, the authors state the following:

“Our perspective on how AI can be more successful is admittedly and unashamedly from a UX point of view. AI needs a focus on UX to be successful.”

This is a central theme in the book. The authors recognize the role UX must play in the development of AI systems, tools, and interfaces. Having now had some experience myself with a few of the generative AI platforms, I think the authors are correct, and an emphasis on UX for AI tools won’t just make those tools easier and more pleasant to use, but a better UX experience can actually save these tools from being written off by the general public as novelties or passing fads. The failure of AI to live up to hype in past decades did lead to these kinds of dismissals, but the latest wave of advancements may have reached a tipping point that insulates AI from another major cultural setback or lengthy pause.

Chapter 1: Introduction to AI and UX

This chapter does a respectable job of making the important connections between UX and AI. The authors prove that they know enough about these connections to be credible voices from which the reader can learn.

Drawing from their significant UX work, Lew and Schumacher tell us that “For any product, whether it has AI or not, the bare minimum should be that it be usable and useful. It needs to be easy to operate, perform the tasks that users ask of it accurately, and not perform tasks it isn’t asked to do. That is setting the bar really low, but there are many products in the marketplace that are so poorly designed where this minimum bar is not met” (16).

Throughout the book, the authors make a good case for the application of pretty much all general UX principles to AI products. Chapter 1 just lays out the landscape and major connections.

Chapter 2: AI and UX: Parallel Journeys

As the title implies, Chapter two provides a nice historical walk through AI and UX development. Particularly interesting is the focus on “AI winters” that followed periods of overhyped AI performance in the 1960s and again in the 1980s.  Also, they mention the “domain-specific AI winter” for AI personal assistants which followed the overhyping of Siri in the early 2010s.Part of the reason for these AI winters is that the developers of the systems were not focused enough on user experience.

I appreciate the differentiation the authors try to make between HCI (human-computer interaction) and UX in chapter 2:

“Where HCI was originally focused heavily on the psychology of cognitive, motor, and perceptual functions, UX is defined at a higher level—the experiences that people have with things in their world, not just computers. HCI seemed too confining for a domain that now included toasters and door handles. Moreover, Norman, among others, championed the role of beauty and emotion and their impact on the user experience. Socio-technical factors also play a big part. So UX casts a broader net over people’s interactions with stuff. That’s not to say that HCI is/was irrelevant; it was just too limiting for the ways in which we experience our world” (50).

The way I interpret this is that HCI is akin to a substrata of UX.

Chapter 3: AI Enabled Products are Emerging All Around Us

And

Chapter 4: Garbage In, Garbage Out

These two chapters are where the book shows its age a bit as a pre-ChatGPT publication. Although there are some interesting examples of AI systems discussed in Chapter 3, the next chapter disconnects enough from the user experience that I did not find it valuable as a UX text. The focus of Chapter 4 is the data that AI runs on. The authors are correct that without quality data, the user experience of any AI product will suffer, but since current AI systems are such black boxes when it comes to their training data, this is somewhat of a moot point for me right now.  

I will say that I perked up a bit reading about voice assistants and Grice’s four maxims for communication (67). Anyone studying generative AI could benefit from using those maxims as a starting point for evaluating what our machines are capable of. Current LLMs and systems based on LLMs seem to handle the three of the maxims with relative ease much of time (quantity, relevance, and clarity), but the truthfulness of LLM’s communication is where many people are finding the most problems. One could argue that truthfulness is the most important of the four, but it is obvious that advances in the other three areas have come quickly and impressively. I think it is entirely possible that AI systems make progress on that fourth maxim in the near future. And if things in the AI world are not interesting enough for someone yet, they will be once the programs are more reliably accurate purveyors of information.

Chapter 5: Applying a UX Framework

This final chapter is still relevant in the post-ChatGPT world. It ties the idea of UX and AI back together (whereas they diverged a bit in the previous two chapters). This quote at the beginning of the chapter seems especially relevant:

“For many people, there’s still a hesitance, a resistance, to adopt AI. Perhaps it is because of the influence of sci-fi movies that have planted images of Skynet and the Terminator in our minds, or simply fear of those things that we don’t understand. AI has an image problem. Risks remain that people will get disillusioned with AI again” (109).

I think the authors are correct that people could become disillusioned with AI again, but this will probably be less about the UX dimension and more about the existential threats, security concerns, and intellectual property issues that accompany 21st century AI. Either way, since AI is becoming so ubiquitous, I would not predict another AI winter like authors detail in Chapter 2.

One of the most interesting points in Chapter 5 regards the purpose of a product. As they lay out the case for applying a UX framework to AI, the authors pose the following questions:

“Probably the most important thing that defines any application is what it does—we call this “utility” or “functionality” or “usefulness.” Basically, is there a perceived functional benefit? In more formal terms, does the application (tool) fit for the purpose it was designed for?”

The reason this is interesting is because I am not sure the creators of ChatGPT and the other generative AI systems (or any of the precursors dating back to the 1960s) really had a specific end user functions in mind—at least not as the driving motivation for their creations. It seems like the systems have all been designed just to see if the creators could make a machine that could communicate like a human and display some level of “intelligence.” Along the way, clever people have figured out how to leverage this technology for different purposes, and profit-driven people have too, but I really don’t think that thoughts of the usefulness of LLMs weighed heavily on the creator’s minds. Evidence for this exists within the current user experience of ChatGPT. When users first access this application they see an interface with suggestions for how the app could be used. That is weird.

When we buy tools or access technologies, we typically already have the function in mind; that’s why we sought the tool to begin with. Generative AI companies are almost saying to the user, “Here it is. Figure out for yourself what purpose it has for you.” For the time being, that is the user experience for many users of generative AI.

As for the user experience of Lew and Schumacher’s book, I think they did a decent job of connecting two fields that need to be connected. A reader with a good grasp on AI could probably skip chapters 3 and 4, but there is plenty of helpful information and background in Chapters 1,2, and 5 that still holds up well in this four-year-old title from Springer/Apress.

Final Project Outline and Schedule: ChatGPT Student Users Case Study

Project Goal: This project seeks to investigate and analyze how students in a first-year college writing course interact with the ChatGPT website/app in the context of educational assignment. From a UX perspective, the researcher wants to understand how the generative AI tool is used, whether students show variations in how they use the tool, and if there might be changes that could improve the experience for student users.

 

Schedule:

(Completed 3/4/2024) Create and post project outline/schedule

(Completed 3/4/2024) Create questions for focus group discussion

By 3/12/2024: Update assignment instructions

3/12/2024: Provide instructions for student writing assignment that will incorporate ChatGPT usage

3/12/2024:  Students create OpenAI accounts

3/12 and 3/14: Students complete the pre-writing/invention assignment

3/14/2024: Students submit links to their ChatGPT conversations

3/19/2024: Conduct focus group discussion

3/20-4/15/2024: Synthesize data and findings; write report (Organizational Structure of Report: Explore, Define, Ideate)

By 4/22/2024: Post final report

 

Target Research Sample: ~12 ENGL 1101 students at GGC. 

GGC student demographic profile:

Application of student demographic information: The researcher believes the institutional data for the college is representative of the student sample he is working with. The researcher has deemed that the benefits of having more accurate demographic data for the research sample is not worth the costs (time and effort) of collecting the data. An attempt to collect this data would be an additional burden and privacy encroachment for the research participants. The risks of not collecting the data are minimal but could result in some problems with the interpretation of certain data that is collected and the accurate representation of the student population of the institution. Generally, the researcher should take into account the diversity of the student population (specifically racial diversity and students with disabilities that impact learning) when designing the case study. Information on student socioeconomic status would be helpful as well because of correlations with access to technology. To that end, 2021 data shows that 62% of GGC students received Pell grants (compared to 47% at comparison group institutions). In sum, GGC serves a student body that is far more diverse and financially in need than other institutions in a comparison group of similar institutions.

https://nces.ed.gov/ipeds/use-the-data

Methods: Students will use ChatGPT for a prewriting activity that will help them explore their own tentative thesis statement or topic.

Prewriting Activity Prompts:

Please help me prepare for writing my persuasive essay about a significant problem for Generation Z. Ask me questions–one at a time–that will help me come to a thesis, understand the strengths and weaknesses of my position and any opposing view(s). The problem I want to focus on is…

–or—

Please help me prepare for writing my persuasive essay about a controversial issue in my field of study. Ask me questions–one at a time–that will help me narrow down my topic to a specific debatable question, come to a thesis, understand the strengths and weaknesses of my position, and understand any opposing view(s).

–or—

Please help me prepare for writing my persuasive essay about a social justice issue. Ask me questions–one at a time–that will help me come to a thesis, understand the strengths and weaknesses of my position and any opposing view(s). The problem I want to focus on is…

–or–

Let’s engage in a dialectic exercise. I will present my tentative thesis, and then you will take the role of Socrates and ask me yes or no questions (one at a time) to expose any potential flaws or contradictions in my position.

Data Collected: Students will be asked to submit their ChatGPT conversations in the form of sharable links provided by ChatGPT. The researcher can use this data for qualitative and quantitative analysis (number of replies and length of replies). Qualitative data will also be collected from a focus group discussion.

Focus Group Discussion Questions:

(Researcher will start with background info on research purpose, discussion of consent, and info about groupthink)

How would you describe your prior use of generative AI (like ChatGPT, claude.ai, etc.)?

              Never used it/Used once/Used a few times/Use regularly

              For what purposes?

 

Did you use the mobile app or desktop website?     Mobile / Desktop

What were your initial impressions of the appearance of the tool? Design? Functionality? Features?

              Notes:

 

How easy was it to…

find and access ChatGPT and create your account?  

Very Easy / Somewhat Easy / Somewhat Difficult / Very Difficult

Comments:

use ChatGPT to….

get responses (any responses—helpful or unhelpful)

              Very Easy / Somewhat Easy / Somewhat Difficult / Very Difficult

                             Comments:

share the link to your ChatGPT conversation?

              Very Easy / Somewhat Easy / Somewhat Difficult / Very Difficult

                             Comments:

 

What about…

Speed of ChatGPT responses?  

Scroll speed of the screen as ChatGPT responds?

Appearance of the interface? Colors, font size? Layout?

              Comments:

 

Additional features

              Did you notice the thumbs down button? Did you use it?

                             Did not notice /Noticed but didn’t use / Noticed and used

              Did you notice the message that says, “ChatGPT can make mistakes. Consider checking

important information”?

              Noticed / Did not notice

              Comments:

Other features you wished the interface had?

              Comments:

Features that you wished the interface would change or remove?

              Comments:

Accessibility questions

How could the tool be better designed to improve the experience for people with disabilities?

              Comments: