The Problem Behind the Problem: What Service Design Can Teach Us About Building Better AI

As AI becomes increasingly accessible, discussions often focus on models, datasets, and technical performance. Yet many of the decisions that ultimately shape user experience, adoption, and impact are made much earlier, when teams define the problem they are trying to solve.

Drawing on service design principles and experiences designing AI-enabled solutions in agriculture and education, this talk explores how problem framing influences technical decisions, the role AI plays within a larger system, and the value it can realistically create for users. Through examples from an AI-powered grain quality assessment tool for soybean farmers and a foundational numeracy solution for government schools in Uttar Pradesh, I will share how understanding users, workflows, incentives, constraints, and surrounding ecosystems can reshape both technical and design decisions.

The session will reflect on how decisions about what data to collect, what outputs a model should generate, and what assumptions are built into a system ultimately shape user experience, adoption, and impact. It will also explore how a clearer understanding of the problem, its constraints, and its intended outcomes can help teams build more robust solutions and communicate them more effectively to implementers and adopters.

In this talk I share my reflections from designing AI systems intended for social impact to share insights about how technical decisions are indeed early and most crucial user experience and value proposition decisions, that need to be iterated on as the context of the problem and the constraints if the technology become clearer.

How a deeper understanding of real-world contexts, workflows, incentives, and constraints can help teams make technical decisions that are better aligned with how people actually work and make decisions.

• Why defining an AI problem is also a decision about who a solution serves, what value it creates, and what outcomes it can realistically influence.

• How choices about data, model outputs, and system architecture shape the information users receive, the actions they can take, and ultimately the adoption and impact of a solution.

• Practical lessons from designing AI-enabled solutions for soybean farmers and government school teachers, including how field insights led to changes in both technical and design decisions.

• Why clearly communicating a system's assumptions, constraints, intended users, and intended outcomes is critical for successful implementation, adoption, and open-source reuse.