From: aidotengineer
Alex Les, VP of Data Science and AI at Huge design and technology company, discusses how teams can utilize AI simulation to accelerate design, focusing on “invisible users” and “invisible interfaces” [00:00:09]. The discussion covers the current state of UX and AI, potential future states, and a proposed method to achieve them [00:00:19].
The Current State of AI and UX
Currently, AI faces a “trust gap” [00:00:35]. Research from December 2024 by Edelman indicated that only 32% of US adults trust AI, and 44% of adults globally are comfortable with businesses using AI [00:00:38]. This gap is attributed to “AI slop,” where generative AI fails, such as incorrect web search results or unrealistic product pricing on websites [00:00:56]. Many website creators are “stuffing AI chatbots” everywhere, presenting it as magic [00:01:25].
However, the true “magic” of Generative AI (GenAI) is its capacity as a UX revolution, allowing users to interact with machine learning models using natural language, a previously impossible feat [00:01:31].
The Opportunity: Invisible Interfaces and AI Accelerated Need Finding
Revisiting UX first principles, the concept of “invisible interfaces” by Don Norman (author of The Invisible Computer) describes software so seamless and intuitive that users forget they are using it [00:02:18]. The opportunity for AI is to design interfaces that genuinely feel like magic, not through poorly integrated chatbots, but by accelerating the “need finding” phase of design [00:02:31]. While the presenter is a data scientist and not a designer, a process for delivering AI accelerated need finding is proposed [00:02:46].
A New Approach to the Design Life Cycle
The current design process, pre-ChatGPT, is data-driven, relying on qualitative, quantitative, and ethnographic observations to guide prototyping [00:03:33]. The proposed new approach for design is analogous to pilot training, which heavily uses simulation for operating in complex environments [00:03:13].
This evolution involves empowering designers to work with “invisible users” in the form of AI simulation [00:03:54]. This turns traditionally collected data artifacts from need finding into active participants in the design process, providing designers with a mini feedback cycle to refine designs [00:04:01].
Accelerated Need Finding Process with AI Simulation
The new process for need finding in the era of AI simulation shares many similarities with existing methods, including:
- Defining the audience [00:04:39].
- Mapping their intentions [00:04:42].
- Identifying specific tasks to accomplish intentions [00:04:45].
- Conducting analysis [00:04:49].
- Refining insights [00:04:50].
- Developing design alternatives [00:04:52].
However, key differences arise in the components and workflow with AI simulation [00:04:57].
Defining Audiences with Data
The process starts by utilizing data that represents target audiences [00:05:06]. Huge’s internal data platform, “Live,” contains demographic, psychographic, and contextual data to simulate audience behaviors, forming a critical foundation [00:05:14].
Intent Mapping with Intelligent Twins
The data is then transformed into an active simulation called an “intelligent twin” [00:05:34]. An intelligent twin represents user behaviors, desired outcomes, needs, and motivations, becoming an active participant in the design simulation process [00:05:41].
Briefing Intelligent Twins
Intelligent twins can be briefed to evaluate interfaces by focusing on specific tasks, similar to how a human designer might conduct a heuristic analysis [00:05:56]. This allows for the evaluation of specific interfaces or websites [00:06:09].
Example Project: Global Audit of Sports Websites
To demonstrate the advantages of this simulated methodology, particularly its scale and speed, a sample project was conducted: a global audit of sports websites [00:06:23]. This scenario imagined a business wanting to partner with global sports leagues and understand how to engage audiences across different countries and cultures [00:06:41].
Project Setup
Two personas were created: a casual fan new to sports and a lifelong, savvy super fan [00:07:06]. Across three different sports websites (basketball, Olympics, English Premier League), a series of tasks were briefed for intelligent twins [00:07:18]. These tasks spanned categories like navigation, information architecture, and fan engagement, with four tasks per category, resulting in 72 AI-simulated actions [00:07:23].
Findings
The speed and scale of intelligent twins allow for high-level insights across an entire category or granular dives into specific areas [00:07:49]. The audit revealed that navigation on all sites performed well in terms of task completion [00:08:09]. However, as fans went deeper into browsing content, information architecture, and engagement pathways, initial successes dropped off [00:08:29]. This highlights user pain points that, when addressed, can help repair the AI trust gap [00:08:53].
The methodology can also scale to surface friction in specific, nuanced areas of the experience, enabling the creation of focused, broad, and deep design briefs [00:09:06]. Utilizing computer vision models and human-in-the-loop observation, the process can generate design briefs that allow human teams to focus on critical user pain points, thereby accelerating and improving the design process [00:09:33].
Future Outlook and Limitations
Significant progress is anticipated, with tools like Anthropic’s MCP protocol potentially integrating prototype components from Figma directly into code components for frameworks like React or Node.js [00:10:06]. This acceleration in design tool capabilities emphasizes the importance of focusing on “the why”—the strategy and the problem users need solved—as “the how” becomes increasingly easier [00:10:31].
This methodology is currently experimental. Limitations and areas for improvement include:
- Reproducibility: Standardizing parameters like briefing instructions, simulated audience dimensions, audit runs, and task completion/failure parameters in a code repository [00:11:00].
- Test and Control: Applying a test and control methodology to isolate the strengths of intelligent twins for AI accelerated need finding and determine how they can complement human teams across various industries, geographies, and domains [00:11:24].
Conclusion
To repair the current AI trust gap, the focus should shift from force-feeding users ineffective GenAI chatbots to providing better websites, mobile apps, and interfaces that offer clarity and simplicity [00:11:57]. AI simulation can be a powerful tool to empower design teams to gather insights smarter, faster, and better, ultimately creating interfaces that restore and repair user trust [00:12:17].