A new era of UX: evolving your design approach for AI products

Before ChatGPT rolled on THE scene A year There is, artificial intelligence (AI) And machine learning (ML) were THE mysterious tools of experts And data scientists – teams with A plot of niche experience And specialized domain awareness. NOW, things are different.

You are probably while reading This because your business has decided has to use OpenAI Google Tag Or another LLM (big language model) has build generative AI features In your product. If It is THE case, You could be feeling excited ("It is SO easy has TO DO A great new functionality!") Or overwhelmed ("For what TO DO I get different outputs each time And how TO DO I TO DO he TO DO What I do you want?”) Or maybe You are feeling both!

Functioning with AI could be A new challenge but he doesn't need has be intimidating. This job distills My experience Since years spent design For "traditional" M.L. approaches In A simple together of questions has help You move Before with trust as You to start design For AI.

A different kind of UX design

First of all, a few background on how AI UX design East different Since What You are used has TO DO. (Note: I will be using AI And M.L. interchangeably In This job.) You could be familiar with Jesse James Garrett 5 layers model of UX design.

Jesse James Garrett Elements of User Experience diagram

Garrett model works GOOD For determinist systems, but doesn't capture THE additional elements of machine learning projects which will affect THE UX considerations downstream. Functioning with M.L. means add A number of additional layers In THE model, In And around THE strategy layer. NOW, In addition has What You are used has design, You Also need A Deeper understanding from:

How THE system East built. What data East available has your functionality, What he understand, how GOOD And reliable he East. THE M.L. models you go to use, as GOOD as their strengths And weaknesses. THE outputs your functionality will generate, how they will vary, And When they will fail. How humans could react differently has This functionality that you would have to wait for Or to want.

Instead of ask yourself "How could We TO DO This?" In answer has A known, scope issue, You could find yourself asking, "Can We TO DO that?"

Especially if You are using LLM, you go likely be functioning backward Since A technology that unlocks entirely new abilities, And You to have has determine if they are appropriate For solve problems You know about, Or even problems you have Never considered soluble Before. You could need has think has A upper level that usual – instead that display units of information, You could to want has synthesize big the amounts of information And here tendencies, patterns, And predictions instead.

"You are design A probabilistic system that East dynamic And that reacts has contributions In real time"

Most important, instead of design A determinist system that do What You say he has TO DO, You are design A probabilistic system that East dynamic And that reacts has contributions In real time – with results And behaviours that will be unexpected Or inexplicable has times, And Or weighing compromise could be A dark exercise. This East Or My together of five key questions come In play – not has provide You with answers, but has help You take THE following stage In THE confront of uncertainty. Let's go dive in.

1. How will You ensure GOOD data?

Data scientists love has say "Garbage In, garbage out." If You to start with bad data, there East in general No path You will END up with A GOOD AI functionality.

For example, if You are building A chatbot that generates answers base on A collection of information sources, as articles In A online help center, low quality articles will ensure A low quality chatbot.

When THE team has Intercom spear