The UX landscape is rapidly evolving as artificial intelligence (AI) and machine learning (ML) create new tools and reshape user expectations of technology. However, as with all new technology, ML today is still evolving and can produce inaccuracies or unexpected outcomes during user interactions.
Currently, AI struggles with:
- Retaining Context: often initially generating relevant outputs only to lose focus mid-interaction
- Hallucination: confidently providing inaccurate information
- Optimized for Specialization: AI models excel at optimizing for a specific variable or metric, but struggle when pivoting or balancing multiple priorities
To mitigate these impediments, defensive UX can help companies anticipate and manage potential user issues in advance by guiding user behavior, preventing misuse, setting expectations, and gracefully handling errors.
Demystifying defensive UX
Defensive UX helps bridge the gap between users’ expectations and the limitations of any new technology, not just AI. From mobile phones to cloud-computing, defensive UX has helped companies and brands deliver gratifying customer experiences, buying time for new technology to catch up to expectations.
How defensive UX mitigates challenges
Increasing accessibility
Defensive UX helps users understand the workings of ML and LLM features and their limitations, making them more accessible and user-friendly.
Increasing trust
When users observe a feature avoiding harmful outputs and handling challenging scenarios gracefully, they are more likely to trust it.
Improving the user experience
By designing the model and UX to handle ambiguous situations and errors, Defensive UX ensures a smoother and more enjoyable user experience.
Exploring defensive UX guidelines
Google, Apple, and Microsoft have used this time honored strategy to deliver exceptional user experiences with the advent of new AI technologies. What can we learn from these leaders?
Set expectations
All interactions must include subtle reminders of what users can and can not expect from AI
Users are still learning how to interact with AI technology. All too often their expectations are defined by how a human would approach the problem – an expectation that AI will consistently fail to meet. It’s not enough to set expectations during the initial onboarding experience (although that is important). All interactions must include subtle reminders of what users can and can not expect from AI. We can see a great example of this in ChatGPT.
Even users making their thousandth query still see inspirational use cases to prompt new ideas and a key reminder.
E.g “ChatGPT can make mistakes. Consider checking important information.”
Apologize with grace
AI apologies go beyond escalating from a chatbot to a human
AI can and will be wrong sometimes. Your company has to respectfully and gracefully own your mistakes and provide a path of escalation. AI apologies go beyond escalating from a chatbot to a human. There are many different ways that a company can provide a sincere apology and provide the user with a path forward:
-
Bids for the user to try again:
Alexa’s “I’m sorry, I’m having trouble hearing right now.”
-
Suggesting a human instead:
Zendesk’s chatbot escalation macros
-
An opportunity to learn:
Google’s “Tell us why. Let us know what you didn’t like about our suggestions so we can improve our content
Transparent attribution
To build forgiveness for misunderstandings, AI needs to communicate why it made certain choices
AI often behaves in ways that are unintuitive to humans. While humans pick up on social and contextual cues when formulating a response, AI has no such conventions. Consequently, to build forgiveness for misunderstandings, AI needs to communicate why it made certain choices.
E.g Bing Chat interface takes attribution a step further by using it to link to sources, helping not just create understanding with the bot, but helping to mitigate hallucination by allowing users to check the AI’s sources.
Get started applying defensive UX with Algolia Toolkits
Implementing defensive UX for the first time can be a daunting experience, especially while society is adjusting to the new normal of an AI-driven world. While building a defensive UX strategy might be unfamiliar territory, ultimately it utilizes a familiar toolkit of design and UI components. Looking to implement defensive UX into your own AI-driven search and discovery experiences? Satellite components can help you get started. One of the top components to start building defensive UX into your experience include:
- Alerts
Both to provide expectation setting warnings throughout the experience and to surface error information when AI makes a misstep
- Overlays
To educate users about the AI and set realistic expectations
- Forms
Create an escalation path for users when AI gives them inaccurate or unusable results so your team can learn from the AI’s mistakes
It’s difficult to believe it’s only been a couple of years since AI technology abruptly dropped into the consumer market. Though AI has been eagerly received, implementation has had its challenges. To overcome current and future obstacles developers will need to continue developing innovative solutions both on the back and front end. Proactive strategies, such as defensive UX, will ensure the pace of AI adoption will continue to accelerate at a healthy clip.
%20(2).svg)
