Making (today’s) AI work for human-centered design
Cover image generated using Canva AI Image Generator. Prompt: Woman walking on a path in an urban park.
Our relationship with AI existed before we noticed it. The digital tools we’ve been using for years started offering us new options and features. They were helpful so we used them; it was as simple as that. The conversations about AI as an industry with standards, ethics, leaders and losers seemed disconnected from the daily work of scheduling research interviews and facilitating client workshops. When we finally started talking internally about how we wanted to use AI, we realized we were already well down that road.
In this post, we look at AI’s role in our projects, while also noting the requirement that we help AI across the process. We hope this serve as a guide for leaders trying to determine when it makes sense to delegate certain tasks to AI. These thoughts are initial, as we are still exploring our use of AI iteratively, and we continue learning best applications for the technology. We also still have a lot of big, unanswered - or inadequately answered - questions about ethical use and content ownership that will likely be resolved through a series of lawsuits, so take our thoughts here with that in mind.
AI can accomplish many of the tasks that cost us a lot in time and money.
We’ve found that AI is well adapted to undertaking much of our grunt design research work. It can perform rote tasks and save us time to work on tasks where the human element is indispensable. Look for ways AI can save you time and money. These are often repetitive tasks that don’t require much nuanced thinking. What are the ways that AI can free up team member time to do the parts of a job that only your team can do?
As a human-centered design practice we are conscious of the interplay between people and technology. Design research requires our very human ability to empathize as we seek to understand people– their beliefs, values and behavior– in cultural context. Our roles continue to shift and evolve. Our guideposts - what’s best for our team, our world, our work and our clients - are even more essential as we navigate this new relationship.
We primarily use AI as a supporter in our roles and offer two examples below. Hopefully this gives you an idea of the kind of tasks AI performs as you and your team consider your way forward.
AI is best at performing necessary, but mundane, tasks. Look for ways that you can offload some of the tasks that can be automated by AI.
Quire case study - audio transcription: We use a platform called Rev to transcribe the audio of our interviews into a text form. Rev has incorporated an AI tool that saves us a lot of time and money on the frontend in transcribing interviews and other recordings of our primary research. We use Rev to turn audio recordings of our interviews and events into a text form. Rev’s tool automatically transcribes audio recordings, converting audio to text, and cutting down the time it would take for us to listen and transcribe a full audio segment. And by cutting down time, we save money by spending our time on higher value tasks. AI also saves us money since we don’t have to pay for a costlier human transcription service. This is a basic task for which AI is well suited. But even with transcriptions, we still have to review the work before beginning analysis. When we first began using Rev’s AI transcription services, we found that it often mis-transcribed our research subjects’ words due to their southern accents. Over the years, we’ve seen AI improve on this front, but we still have to verify the transcripts’ accuracy– some things still slip through the cracks. When we identify a discrepancy, we review the audio to hear what the research subject actually said and make sure the text is understandable.
AI summarizes long texts, which saves us a lot of reading time. This is a function with broad applications for most organizations.
At its foundation, AI is designed to be skilled at pattern recognition. It can quickly identify patterns in text and convert these into summaries of the main ideas. AI can be helpful when you have a large document and need to conserve time that would be spent on close reading. AI can analyze the document in seconds and send back to you a brief summary. AI can read board meeting minutes, annual reports, handbooks – anything really – and deliver a succinct summary. You can often choose how long you want the summary to be. A paragraph or a page?
Quire case study - text summaries: Beyond transcription, Rev’s AI tool is also able to summarize our transcripts. Likewise, the qualitative analysis software we use, MAXQDA, can create summaries of secondary literature that we use in early stages of a project. With the summaries as a guide, we can then review documents to gain a deeper level of understanding. This helps us more quickly develop our own ideas, questions and strategies as we plan our research. MAXQDA’s AI tool helps us when we’re trying to go through a mass of background literature while working within the time constraints of the project. MAXQDA also has an AI coding tool which is able to do a first pass at assigning codes to the text. Codes are the categories of themes that we assign to words and sentences in a document. If we’re doing a strategic planning project and trying to understand an organization by analyzing interview transcripts in MAXQDA, we might have a code for “values,” one for “personal experience,” another for “challenges/barriers,” and many, many more. MAXQDA’s AI tool can cut down on our time by doing some higher-level, categorical coding before we get into the data. AI has a harder time teasing out subtleties, and we have to do this analytical work on top of the foundation that AI has laid. That’s where our follow up work is required.
Look to AI for performing low-skill tasks that are time intensive. These are tasks that while necessary, keep you from performing other higher value work that can only be done by a member of your team. These mundane tasks still require your review, to make sure AI is effectively serving the goals of the team and the project.
AI has a harder time scaling up to discursive, ambiguous tasks like considering means for an organization and the people within it. What time and space does your organization want to protect for human deliberation?
While AI is helpful in doing simple tasks like transcribing and summarizing, it has a hard time moving from that lower level to the level of understanding what it all means and where to go from here. We do that work as a team through discussion, deliberation and debate. We lean into AI where information can be fact checked. AI’s literal approach to data is incredibly useful there. It can’t yet identify the ‘so what’ about all the information. Contextual research also gives data that gets lost in a transcript. We hear a speaker’s tone of voice, can see when an unfinished thought communicates a lack of trust or simply the speaker stumbling upon a better example. All of those cues factor into decisions in a way that AI can’t yet handle.
Mitigating risk with AI’s unknowns
The AI landscape is changing rapidly. Currently, we use AI features within purchased, digital tools that are not ad supported. The tools we purchase have privacy and use policies that speak specifically to sensitive data protections. We upload content that has been privately provided, or is publicly available only for our internal use. One of our highest goals is to create a context of safety and trust where clients and research participants know that the information they provide us in the course of our work will be safeguarded to the best of our ability throughout the life of the project and beyond. Our use of AI has to align with that goal.
Learning through doing
By experimenting with AI capabilities, we’ve found the technology to be incredibly helpful in accomplishing much of the frontend research tasks in human-centered design. These easily automated tasks, like transcribing interviews and summarizing text, serve as foundations for the work that only we can do. Through improvisation in our practice, we’ve found that our team remains irreplaceable by AI when it comes to higher level tasks, like making sense of all the data, brainstorming future directions, and making hard and nuanced decisions. Further, much of our work necessarily entails empathetic relationship building with research subjects so that trust can be established and conversations can be had. Organizations get to decide when and where AI might be appropriate and which situations still require more robust human input. With the AI landscape changing so rapidly, it’s hard to predict what comes next. We’ll continue to explore what’s possible, beneficial and ethical as our understanding evolves.
