top of page

Learning #6: Redefining Human-Centered AI Design: Why Traditional Design Approaches Fall Short in the New Context of AI

Updated: Jan 20

By: Sarah Tan, Founder and CEO of Formatif.

AI product design requires a new design approach. With uniquely complex AI tech challenges, traditional user-centered design thinking is no longer good enough. AI is an ever-evolving field with limitless potential, as we haven’t unlocked all its capabilities. How do we design for the unknown and ensure responsible AI solutions? In this blog post, I introduce the concept of 'Human-Centered AI'—a revolutionary, pan-disciplinary design thinking approach tailored for AI.


Traditional design thinking is dead.


With AI, there needs to more than just design thinking for approaching AI product design.

With uniquely complex AI tech challenges, traditional user-centered design thinking is no longer enough.

Here’s why.

So quick context.

I’ve spent over 1.5 years researching Human-Centered AI which stemmed from my design thesis project. I’ve interviewed industry AI designers across Meta, Google, IBM and reading (virtually) ever AI design research paper available on the web.

This is what I uncovered —

Traditional user-centered thinking focuses on user-centricity and idea prototyping.

However, when it comes to AI products, the traditional design thinking process falls short due to the new complexity of AI.

Unique Human-Centered AI Interaction Challenges

This is due to 2 distinctive complexities of AI.

Unique Human-Centered AI Interaction Challenges

1. AI Capability Uncertainty:

AI is an ever-evolving field with limitless potential, as we haven’t unlocked all its capabilities.

So, even what may seem like ‘blue-sky’ ideas today could become reality with new data and overnight advancements.

How do we then design for the unknown?

When working with AI models, focusing solely on user needs isn’t sufficient from a user-centered viewpoint to spark design ideas.

Instead, we need to consider the constraints of AI capabilities and the technical feasibility of meeting user needs.

Some questions to ask:

  • What is the new value add with AI capabilities

  • How can we translate user and business requirements into feasible AI outputs

2. AI Output Complexity:

Addressing AI Output Complexities

AI solutions often present unpredictable outcomes or answers, especially in complex situations like natural language conversations with virtual assistants.

When we make AI solutions, it’s difficult to predict and get ready for all the outcomes or answers, especially in complex situations like Siri’s talks using natural language.

Designing for AI becomes challenging because we cannot anticipate its actions or creations.

Additionally, the evolving adaptive nature of AI systems further complicates sketching and prototyping.

Most AI systems are now evolving adaptive systems, which produce many possible outputs, making sketching and prototyping more complex and unpredictable.

Rapid prototyping works when we know exactly how users will use something, but with complex AI like Siri, how do we predict every single potential user flow?

It is simply impossible.

Instead, it’s crucial to take into account the intricacies of extreme scenarios, both positive and negative, while also assessing the potential unforeseen consequences.

What exactly does this mean?

Let’s break it down.

Take an ecommerce app using conventional technology for example.

The outputs can be defined quite straightforwardly — etc possible user flows, user journeys (checkout cart, make payment, add to order, choose not to purchase)

Rapid prototyping works when we know exactly how users will use something, but with complex AI like Siri, how do we predict every single potential user flow?

For more complex AI features and solutions, we cannot anticipate every action.

This becomes a problem in more consequential AI solutions that can have larger societal impact — etc Tesla autonomous driving, there can be infinite contextual scenarios that lead to numerous moral and ethical dilemmas.

Rapid prototyping now comes with new contexts like Wizard of Oz simulation.

Beyond just prototyping, Designers now need to consider the intricacies of extreme scenarios to assess potential unforeseen consequences.

After all, AI is always evolving and never 100% accurate.

And it learns from user inputs and data over time.

To build trust and manage expectations, we must focus on how to explain AI to users and it’s potential impact.

This guides UX development to account for these scenarios to anticipate “extreme, unintended” consequences.

And helps us to design scenarios that promote responsible usage.

…How do we overcome these challenges then?

We must comprehend AI’s capabilities, learning processes, and user interactions.

This involves an understanding of technology, business, societal and humanistic factors.

In essence, a pan-disciplinary design approach of:

  • Humanistic: User needs and human society impact

  • Judical: Ethical AI, Responsible AI policies and guidelines

  • Technological understanding of AI Capabilities

Note: It’s not about in-depth technical knowledge of AI.

While helpful, you don’t have to know about the technicalities of how a neural network works for example.

It’s more about familiarizing ourselves with common AI capabilities.

And identifying user design opportunities from there.

Source: Human-centered AI, The role of human-centered design research in the development of AI

Human-Centered AI: The Pan-Disciplinary Design Thinking for AI

Designing for AI requires a broader understanding beyond user-centricity alone.

Yes, UX research and user design is still, and always relevant.

But to effectively design with AI, we need more beyond Design.

The new AI design process requires a fresh approach that integrates design thinking, technology, data, and societal impact.

This includes:

  • Considering ethics and societal impact

  • Connecting user needs with data

  • Designing with AI complexities and understanding its capabilities

Navigating the complexities of the emerging tech landscape necessitates a fresh design approach beyond traditional Design Thinking.

We need to adopt a new approach that combines design thinking and data + humanistic factors.

I talk more about the new AI Design skills set Designers need in my recent article on the AI Designer Blueprint (read more)

I’ve spent over 1.5 years researching Human-Centered AI for my design thesis, and I’ve developed a unique Human-Centered AI Methodology approach to navigate these unique AI challenges.

Human-Centered AI Methodology: Here’s a sneak peek

Want to learn how to apply each step of the HCAI framework?

I share actionable steps in my new “How to build a Human-Centered AI product” article series.

Check it out below:

👇🏻 How to build a Human-Centered AI product

👇🏻 Learn more about Human-Centered AI principles here

👇🏻 Learn how to become a better AI Designer:

About Sarah Tan

Sarah is the Founder and CEO of Formatif, an AI-enabled venture design studio for emerging technology startups. Through AI design workshops and initiatives, she advocates Human-Centered AI Innovation for emerging technology startups.

From designing AI-driven products for Silicon Valley startups, venture investments in B2B startups, and crafting organizational design strategies for MNCs – Sarah’s unconventional design entrepreneurship journey has always revolved around blending business, tech, and design to lead innovative transformations in complex tech sectors. Her experiences include working in Venture Capital investments in the Bay Area, driving innovation consulting in corporates like Philips and Gensler, to working with emerging technology startups in Web 3.0 and AI to design and launch human-centered products. A significant highlight was her yearlong collaboration with AI Singapore, where she developed the Human-Centered AI (HCAI) design methodology and toolkit, revolutionizing AI product innovation with an ethical focus.

She writes about Human-Centered AI and actionable AI-UX tips for startup on her Medium and Linkedin. 

48 views0 comments
bottom of page