Invisible interfaces

From: aidotengineer

The State of UX and AI Today [00:00:21]

Currently, AI faces a significant “trust gap.” Research from December 2024 by Edelman indicated that only 32% of US adults trust AI, and merely 44% of adults globally are comfortable with how businesses use AI [00:00:35]. This skepticism is largely due to “AI slop” or “GenAI fails,” where AI-powered interfaces provide incorrect information, such as suggesting eating rocks daily or buying a car for $1 [00:00:56]. Many website creators are integrating AI chatbots haphazardly, presenting them as “magic” without ensuring their effectiveness [00:01:25].

However, the true “magic” of Generative AI, as Cassie Kosakov points out, lies in its potential for a UX revolution: enabling users to communicate with machine learning models using natural language, a capability previously impossible [00:01:31].

The Concept of Invisible Interfaces [00:02:20]

Drawing from UX first principles, specifically Don Norman’s concept from his book The Invisible Computer, an invisible interface is software that is so seamless and intuitive that users practically forget they are using it [00:02:05]. The opportunity presented by AI is to design interfaces that genuinely feel like magic, not by forcing chatbots onto websites, but by accelerating the “need finding” process in design [00:02:31].

AI Simulation for Accelerated Design [00:02:56]

The complexities of modern digital environments, akin to a pilot’s cockpit with numerous screens and information, necessitate new approaches to design [00:03:07]. Just as pilots use extensive simulation for practice, designers can leverage AI simulation [00:03:19].

Traditional design processes, pre-ChatGPT, relied on designers collecting qualitative observations, quantitative data, and ethnographic insights to guide prototyping [00:03:33]. The proposed new approach empowers designers to work with “invisible users” in the form of AI simulation, transforming static data artifacts into active participants in the design process [00:03:54]. This creates a mini feedback cycle for designers, leading to better design [00:04:11].

A New Approach to the Design Life Cycle [00:03:03]

The accelerated need-finding process using AI simulation shares many similarities with existing design methodologies:

• Defining audiences [00:04:39]
• Mapping intentions [00:04:42]
• Identifying specific tasks [00:04:45]
• Conducting analysis [00:04:50]
• Refining insights [00:04:50]
• Developing design alternatives [00:04:52]

However, key differences arise in the components and workflow [00:04:57]:

1. Audience Representation: Start with data representing target audiences. Companies like Huge use data platforms, such as “Live,” which blend demographic, psychographic, and contextual data to simulate real-world audience behaviors [00:05:06].
2. Intent Mapping with Intelligent Twins: This data is transformed into active simulations called “intelligent twins.” These twins represent user behaviors, desired outcomes, needs, and motivations, becoming active participants in the design simulation process [00:05:34]. Intelligent twins can be briefed to evaluate interfaces based on specific tasks, similar to how human designers might conduct a heuristic analysis [00:05:56].

Example Project: Global Sports Website Audit [00:06:17]

To demonstrate the advantages of this simulated methodology (scale and speed), a sample project conducted a global audit of sports websites [00:06:23]. This imagined scenario involved a business seeking to partner with global sports leagues and understand how to engage diverse audiences across different countries and cultures <a class=“yt=“yt-timestamp” data-t=“00:06:41”>[00:06:41].

Methodology [00:06:58]

• Personas: Intelligent twins were categorized into two personas: a casual fan new to the sport and a lifelong, savvy super fan [00:07:08].
• Websites: Three different sports league websites were audited: Basketball, Olympics, and the English Premier League [00:07:18].
• Tasks: A series of tasks were briefed across categories like navigation, information architecture, and fan engagement, with four tasks per category [00:07:20].
• Simulation: This setup allowed for simulating 72 AI-simulated actions [00:07:31].

Findings [00:07:46]

The speed and scale of intelligent twins enable high-level insights or granular dives into data [00:07:50]. The audit revealed that navigation (the starting point for fans) performed well across all examined leagues [00:08:09]. However, as fans delved deeper into browsing content, information architecture, and engagement pathways, initial successes dropped off [00:08:26]. This highlights the AI trust gap and the need to understand and solve user pain points to repair that trust [00:08:42].

This methodology can scale to surface friction in specific, nuanced areas of the user experience, allowing for the creation of focused, broad, and deep design briefs [00:09:06]. By using AI acceleration, computer use models, computer vision models, and human-in-the-loop observation, this process can generate design briefs that enable human design teams to concentrate on and solve user pain points [00:09:33].

Future Outlook and Limitations [00:10:03]

Significant progress is being made in AI tools. For example, Anthropic’s MCP protocol is showing potential to integrate prototype components from Figma directly into code components in React, Node.js, and other frameworks [00:10:09]. This acceleration in design creation tools emphasizes the importance of focusing on the “why”—the strategy and the problem to solve for users—rather than just the “how” [00:10:31].

Limitations and Improvements [00:10:53]

This methodology is still experimental. Key areas for improvement include:

• Reproducibility: Standardizing parameters such as briefing instructions, simulated audience dimensions, number of 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 design need-finding [00:11:24].
• Complementary Use: Understanding how intelligent twins can be used alongside or in a complementary manner with human teams, especially across different industries, geographies, and domains [00:11:35].

Conclusion [00:11:46]

To repair the current AI trust gap, the focus should not be on more ineffective GenAI chatbots on websites [00:11:52]. Instead, users benefit from better websites, mobile apps, and other digital surfaces that provide clarity and simplicity [00:12:06]. AI simulation can be a powerful tool to empower design teams to gather insights more smartly, quickly, and effectively, ultimately creating interfaces that restore and repair user trust [00:12:17].