Day 2-AI as a Design Partner: How to Get the Most Out of AI Tools to Scale Your Process

ap1

— I’m excited to be here and talking about AI as design partner and how to get the most out of tools to scale processes
  • Sr. Product Designer at Google

 

ap2

—  I think about designing systems at scale, and have worked at health companies and Google products like Nest

 

ap3

— This will be a deep dive into designing for all at scale, and systematizing user journeys for all users through AI sprint methods

 

–Will also discuss difference between augmentation and automation of process

 

ap4

— Deep dive on using generative AI to design inclusively at scale

 

ap5

— I’ll give a brief overview about generative AI systems
  • Trained on mass amounts of data to learn patterns of human language to generate new and coherent meaningful content

 

ap6

— Why should we care though?

 

ap7

— As a product creator you can be product manager, researcher, designer, or anyone involved in digital product creation

 

ap8

— You are often balancing many priorities from features to users, business, and partners

 

ap9

— Things fall through the cracks though with manual efforts, whether features or users not accounted for

 

ap10

— As a creator, you can leverage AI to design for all to be more systematic in process

 

ap11

— So what is product inclusion and equity?
  • You are designing for everyone and centering the most marginalized voices
  • Building at every phase of product creation process

 

ap12

— You are building on the right to belong to be more systematic, as opposed to more manual process

 

ap13

— This sounds great, but how to apply systematic product design process?

 

ap14

— Three pillars exist:
  • Documentation through user flows and journeys
  • Ideation through workshops or design sprints
  • Synthesis, with organizing and prioritizing
— Can be more systematic in speed and execution

 

ap15

— Can design more intentionally to avoid exclusion

 

ap16

— Product inclusion will be more important as well, and can actively promote diversity and inclusion

 

ap17

— So let’s review some data:
  • 1/7 of people have disability in world population

 

ap18

— 1/5 of world population will be people older than 50

 

ap19

— There can be up to 700 million with disabling hearing loss by 2050 and we need to design for these users
  • Typically design with average users in mind
  • But people with many abilities to design for, and could be big part of population

 

ap20

— So as creators know we need to design for everyone

 

ap21

— Situational disabilities can be things like cooking, speaking different languages

 

ap22

— Can also be permanent can be color blindness  or hand tremors

 

ap23

— We also design for multi-modality to use products on TVs, mobile devices, and having scalable experience across all devices

 

ap24

— So what happens when we don’t have these considerations?

 

ap25

— See recent news headlines for consequences of not accounting for women or minorities, or big misses in accessibility

 

ap26

— So how to systematize critical user journeys?

 

ap27

— We can use dimensions of identity as set of hidden or visible that define who we are, how we think, and how we interact with others

 

ap28

— These identities intersect and create more complex identities
  • Intersectionality is what this is called

 

ap29

— We can use AI to do the following
  • Write inclusive user flows
  • Consider intersectionality with personas
  • Define complex identities for our users

 

ap30

— So let’s beak down inclusive user journeys
  • For example: As a low vision users, I can complete device set-up in < 30 seconds
— This can be written across dimensions of identity

 

ap31

— A critical journey can be broken down to “As a…task”

 

ap32

— Will use sample PawPaw dog-walking app to model example going forward

 

ap33

— Follow-along by opening up conversational AI tool for your choice and break down good prompt design

 

ap34

— Prompt design
  1. Tasks/Steps: Describing the guide to mapping out the experience
  2. Persona: Who do we want model to be?
  3. Provide Examples: Tell model what you are looking for
  4. Constraints: Prime model to understand constraints of ask
  5. Chain: Refining original ask for more detail and original ideas through more constraints or examples

 

ap35

— Will show some examples using Bard, but you can use other tools

 

ap36

— I’ll ask the large language model (LLM) to list out several dimensions of identity to use in the next exercise, with one word answers

 

ap37

— Use first prompt to frame your request
  • i.e. Description of PawPaw app
— We can use the prompt as reference rewrite user journey and write user journey for each dimension
  • PawPaw app for people near me, and search for available dog walkers nearby

 

ap38

— Can then generate better prompt that is incredibly detailed and specifies the nature of the ask

 

ap39

— Dimension of identity can pull out a list, and use list with five different examples or 100s of examples

 

ap40

— Ask the LLM to generate creative tasks based on a chosen identity

 

ap41

— You can repeat  this with other dimensions of identity, like age and attributes, to create a relevant user stories for it

 

ap42

— In example, I asked  the model to create dimensions, and  provided coaching at the end
  • Put info in table to read and primed product with user journey
— Results of user journeys were helpful and others missed the mark, so don’t be afraid to refine original ask

 

ap43

— This approach can be used to scale other types of prompt framing for other users modalities or user types

 

ap44

— Can frame multi-modality, temporary disabilities, situational disabilities, or permanent disabilities

 

ap45

— Showing example of permanent disabilities, temporary disabilities, for product prompt and using exact product for reference and replacing dimensions of identity with things like broken bones, sprains and strains

 

ap46

— Can also leverage AI for HMW statements

 

ap47

— HMW work to reframe insights to opportunity areas

 

ap48

— HMW  move from clicks or taps

 

ap49

— Using LLM to focus on specific business priorities like inclusive design, fewer clicks to find a dog walker, multi-modal design and ideas that are refined

 

ap50

— Another example is affinity clustering to take large amounts of ideas and finding common themes, through a design sprint process
  • This workshop can take hours
  • Generative AI can take minutes
— Able to define to five different themes on the left

 

ap51

— Can try abstraction laddering to move from concrete details to abstract concepts, to solve a problem
  • Using ‘why’ to get more abstract

 

ap52

— Can use LLM and example of abstraction laddering, and priming it for easy scheduling on PawPaw app, and can prime the process for refining it later

 

ap53

— Let’s talk through augmenting versus automating

 

ap54

— LLMs are not 100% perfect and we need to stress-test the output

 

ap55

— There can be bias in the LLM such as generating images that are based on white people and gender bias and occupational bias
  • So we need to augment not automate

 

ap56

— Finally, putting it together
  • We enter with intent to scale user journeys using dimensions of identity and have scaled output through a LLM

 

ap57

— This approach also works with Abstract Laddering or leveraging How Might We (HMW) statements for scaled output

 

ap58

— Talked through systematizing user journeys to scale all users
  • This is augmenting, but not automating that process
— Can scale to other design sprint methods

 

— Remember good prompt structure and adjust
  • If something doesn’t work initially, it doesn’t mean that it doesn’t work at all
  • Continue to refine prompt and outputs

 

ap59

Q&A
  1. How are critical user journeys with such a vast user base?
    1. When looking for user journey and usage, frame it out around user types and who we are designing for.
    2. Can be overwhelming, but GenAI tool can write user journey for users in minutes
      1. Try method for designing at scale
  1. Will ChatGPT learn from itself for questions on DEI?
    1. LLM is learning from itself, but datasets are trained on biased data– so make sure you are testing it.
    2. You can approve or disapprove of an output, so use that to train the models
      1. In process of getting better,  so keep as augmentation not automation
  1. If trying to scale in other languages, what to keep in mind?
    1. Depends on the language and some verb structure is different
    2. Put in brackets and model replacements
  1. Library of collected prompts to use?
    1. Don’t have anything published, but will let you know
  1. Downsides to approach outside of existing biased data?
    1. Will get massive amount of info to sift through
    2. Next, what to do with information and how to design for a 100 things
      1. Need to define it and tools to synthesize and shorter list to design for
  1. How to marry techniques with UXR?
    1. Depends on UXR, and more inclusive UXR can be done to take user journeys and recruit those users to study with and account for them
    2. Design for people who are blind or neuro-diverse
  1. Should managing prompt libraries be DesignOps responsibility?
    1. Can be good one
    2. Don’t need to manage library but list of ideas on how to do collaborative design sprints and can create own list to share with team and own design sprint methods
      1. If it works once for you, you can keep using it