In her qualitative research, Google Research Scientist Rida Qadri routinely puts technology under the microscope as she examines how social context shape and impact it.
And her biggest conclusion? End users seldom use technology in the way we expect them to.
Read on to discover her findings on how we can:
- Bridge the disparity between a designers’ intent and the practical use of a product
- Build more customer-centric designs
- Rethink AI models to represent the needs and intricacies of the global population
- Use qualitative research as a foundation to unravel complex, unstructured data
- Amplify the value of high-quality research within an organization
Rida speaks passionately about how qualitative research is the only way to build more representative and inclusive AI.
Dive into the power of qualitative user research with Rida. She joined us to reflect on how technology permeates cultures but doesn’t translate across geographies quite so seamlessly.
User Research in the Global South
Rida visited Jakarta, Indonesia to study mobility platforms for her Ph.D. thesis at MIT.
The 30 million residents of Jakarta primarily used bikes or scooters to move around. Gojek and Grab, two large mobility platforms, tried to disrupt age-old practices by integrating ride hailing, food and package delivery, and task handling within their applications.
The Diverse Mobility Landscape
In the U.S., Uber transformed the way people move. It also alienated workers from each other because they had no central meet-up location.
Rida was acutely aware that transit options aren’t as regulated in Jakarta, and the mobility industry was less formal and organized. She spent time on the ground, speaking with drivers and understanding how they integrate new technology into their lives. Rida discovered that drivers found interesting ways to continue their existing practices and bring them into the new tech moment.
To her surprise, mobility platform drivers created a collective sense of identity around their employment. They designed logos, wore uniforms, held events and pooled insurance to take care of their communities. They established ‘base camps’ across the city — second homes where they had access to seating, food, drink and electricity. More important, it was a place of community building, socialization, rest and relaxation.
“You could not miss these groups. As a result, drivers developed a lot of latent power. Platforms were forced to sit up and take notice,” Rida said.
A Shift to Customer-Centric Design
Gojek and Grab introduced base camp-like stations at high-traffic areas, such as malls and train stations. Rida discovered through her in-person interactions with drivers a mismatch between tech offerings and the on-ground reality. Assumptions of tech designers and researchers sitting in the U.S. didn’t translate into an effective approach for the Indonesian market.
Mobility apps used proximity-based matching to link drivers with customers. The closer they were, the likelier they received tasks. Unfortunately, drivers couldn’t wait more than a couple minutes outside busy malls and stations.
As a workaround, drivers exploited a security loophole. They created a grey market app that masked their GPS location. They’d wait a mile away, at a location where parking wasn’t a problem, and head over to busy areas once they received customer alerts. Indonesian drivers’ local knowledge was vital in how they interact with mobility apps.
“Local context matters and practices users are used to in that context, matters. Users will always have the power and understanding to shape and bend technological tools to their will,” Rida said.
Treating users (the drivers) as experts shaped the way Rida now thinks about UX design.
“As tech designers, we define the research question, the protocol (and) the kind of insights we want. ‘Here’s the technology that we’ve made, and here’s how it’s great. Now just tell us how you’re going to use it.‘ I think we need to approach it in a more open-ended way, where we go to users and ask — ‘what are your pain points? What are your solutions’?”
User Research and Generative AI: Studying South Asian Representation in Images
User research needs a big orientation shift. Companies must carefully consider who is the purveyor of design solutions and who has the technical know-how to deploy them.
“Instead of trying to control the user, can we co-create and co-design with users?” Rida asked.
Qualitative Research Brings Rich Insights
In examining how certain cultures are disregarded by AI, Rida’s team at Google explored the limitations of text-to-image models in representing the South Asian context. A text-to-image model requires you to enter a prompt; and the model generates an image for you based on your criteria. Application or use cases of this AI software include storyboarding stock, photography, animation and illustrations.
“We already know that technologies developed in the Western world don’t work as well for non-Western users. Generative AI is being trained on the internet as the collective archive of our digital world. But it is a limited archive,” Rida said.
Ordinarily, annotators ask participants to evaluate images on a large scale, using a thumbs up or thumbs down or ranking it on a 5-point Likert scale. This felt very limiting to Rida and her colleagues.
“If you’re going to get a human being to evaluate these images, then take advantage of their humanity, how much cultural knowledge they have implicitly,” said Rida. They conducted in-depth focus groups with participants from India, Pakistan and Bangladesh. Rida and team asked interviewees what they would like the model to generate, and asked them to reflect on the output generated.
Overlooked by AI: Shedding Stereotypes
Some participants said AI depicted their culture as narrow-minded, stereotypical and not necessarily representative.
“Generative AI is developed to give you the majoritarian, most empirically obvious and abundant representation,” Rida reminded us. That’s why the output feels like stereotyping – it doesn’t produce anything novel.
“Fundamentally not seeing yourself represented in technologies continues this long line of what people felt was stereotyping and erasure of their cultures in the media. If our new technologies continue that, then what are we disrupting?” Rida asked.
This isn’t the first time we’ve heard that. Microsoft’s Mary Gray spoke extensively about how AI is failing certain demographics.
To fix this misrepresentation, Rida recommends going back to the source:
“If you want to create something new with these models, you have to go back to the creators and ask, can we retrain (or) customize this model to make sure that we uplift more niche and personalized representations?” she said.
For more details, read the entire study at AI’s Regimes of Representation: A Community-centered Study of Text-to-Image Models in South Asia.
Implications for User Researchers & UX Designers
With all the hype surrounding generative AI, it’s increasingly important to evaluate how companies have implemented systems. Are you deploying technology into an atypical structure?
It’s important for companies to continually ask UX designers:
- How does this work?
- What process does it follow?
- What does it do structurally from A to B?
- Does this process make sense for the task specification that I have in mind?
It’s vital to understand the use cases and limitations of AI models.
“I think we focus so much on what new technologies can do, that we forget to ask what they can’t do,” Rida said.
Rida shared an example of the job market to illustrate her point:
For years, companies have used resume or CV-screening technologies to determine top candidates from the applicant pool. Many applicants discovered that companies were only looking at top-tier Ivy league universities. So they inserted text (such as “University of Oxford” or “Harvard”) in white ink to pass the first stage of the resume-screening process.
Loophole: Humans can’t see the fib, but the computers can.
“If we are to build evaluation systems at scale that are sensitive to, and useful for globally launched technologies, then we have to understand what users globally care about, how they evaluate technologies and what they want to use technologies for. That’s going to change across the globe. So how do you do this without deeper engagement with people? Personally, I feel it’s qualitative research. That’s why qualitative research and social scientists are such an important asset for companies as they build AI technologies,” Rida said.
The Case for Qualitative Research in AI
Machine learning experts and computer scientists train in quantitative methodology. Theirs is a quest for scale and generalizability:
Give us a sample, and we’ll tell you characteristics that apply to the larger population.
A rather close-minded approach.
“Recognize that the world is a very complex and messy place. Pick up one part of it and then try to make sense of that complexity. As opposed to trying to impose this external, superficial order that just doesn’t work at all,” Rida said. “Qualitative research is a great tool not just to uncover the messiness, but to try to make some sense of it and try to use that messiness to your advantage and gain insights from it.”
At Rosenfeld Media’s Advancing Research’ 2023, Rida echoed the value of user research in conversation with Marvin CEO Prayag Narula.
Why Choose Quantitative Research over Qualitative Research?
One question we hear all time:
If AI and qualitative research are the linchpin of making more customer-centric decisions, why do companies continue to prioritize hiring more quantitative researchers?
Well, it’s a grassroots problem.
Rida said intervention is necessary in the education system — curriculum designers and educationists must incorporate interdisciplinarity in their coursework.
“A lot of work needs to be done to build respect for interdisciplinary expertise within the computer science pedagogy,” she said.
Rida cited an example from MIT, where students can now enroll in a computer science(CS) and economics or CS and urban studies major.
The intent behind it? It enables an understanding of what other people bring to the table. It brings respect for each other’s expertise and acknowledgment of the limits of your own. Expertise can come in many forms — cultural, lived and disciplinary expertise. “This is a culture shift that companies need to make. It’s fundamentally a pedagogical shift we need to make in how we train computer scientists,” she said.
Advice for Qualitative Researchers
In an environment densely populated with computer scientists and engineers, how do Rida and her colleagues encourage people to place value on their work? She offered some advice to budding researchers on how to communicate the influence of qualitative research:
- Establish a foundation. How do you integrate your insights? Pitch your findings as foundational knowledge which you can build off. “You’re not going to know how to measure something you don’t know exists. Qualitative research is a good way of figuring out what exists, how it exists, and then you can think about building metrics,” said Rida.
- Strength is in the method. Qualitative research is a rigorous process. Understand the capabilities and limitations of your methodology. “Be very convinced about the strengths of your method — emphasize what this method got us that we wouldn’t have gotten through surveys, rater tasks or observational data,” she said.
- Build a narrative, tell the story. Create a deck to showcase the real-life value of research work. “Marshal a good corpus of examples, whether from yours or someone else’s research, where qualitative data brought out insights that were important and pivotal and otherwise would have been missed. Human beings respond to stories, (it’s) ingrained in our DNA evolutionary imperative,” she said.
Rida addressed one and all with her final piece of advice:
“As qualitative researchers, own your expertise. As non-qualitative researchers, I hope you now have an understanding of how incredibly powerful and essential qualitative research is to build any technology that wants to engage with users.”
Let this be your rallying cry, user researchers. We need your qualitative research skills to improve AI of the future!
Photo by Google DeepMind on Unsplash
