Episode 7

We show how the normal software development cycle does not work with AI and how the modified dev model needs attention from UX at every step

Music: The Pirate And The Dancer by Rolemusic

Transcripts

Here is the scenario for this episode: The boss gives you access to the companies data and asks you to come up with a model that uses it. With all this data it’s got to be good for making something the users will use right? You buckle down, work with data scientists and make a lot of tweaks to the data come up with something, but no matter how much you advertise it no one wants to use it. Back to the drawing board. This time you find out what the users do want, more tweaks to the data and get a model that is accurate. People love it, tons of users flood in and flood the server. The servers crash from too large of a model. The IT guys say they can fix it and bring in a bunch of new hardware. It all seems to be going fine until you notice every review of your app laughs at how inaccurate it is. This can’t be, it’s the same model, just running on different hardware, right? Lets make sure this doesn’t ever happen. Today we are covering the development cycle for AI This podcast is called design for AI It is here to help define the space where Machine learning intersects with UX. Where we talk to experts and discuss topics around designing a better AI. music is by Rolemusic Im your host Mark Bailey Lets get started music Machine learning up to this point has been more on the research side. So much so that it really doesn’t fit in to the normal software development cycle. There are all these gotchas that won’t let you fit into the normal cyclic agile sprints that most people are used to. This affects getting in good design. A big part of UX design not slowing down the software development cycle is to have a regular process so UX can run in parallel to development. It is possible with machine learning development, the cycle just looks a little different. The normal software development process is building a machine. It’s a really complicated machine, but in development terms it is still stateful, so development is done to by writing to the test case. For the updated process, instead of a machine, think of it like you are hiring an employee. There are 5 stages to hiring an employee.

• Plan
• This is laying the groundwork
• lay out the job listing – what are the requirements?
• Find Objectives, why are you hiring them?
• Job posting
• What is the purpose & design
• Set your goals
• Define benchmarks
• hire
• Build On Expertise
• Collect representative data
• Build the model
• Data scientists train the model
• Train – The model is watching how you do things
• Reinforce Education
• Subject matter experts train the model
• Shadow – You are standing over their shoulder.
• Build Trust
• Lead
• Mentorship
• AI leads task
• Subject matter expert manages AI

Step 1: “Plan” Let’s start with the plan. Before even thinking of Machine learning – collect data. Not just analytic data, user data. This is normal UX research. Is machine learning necessary? Remember AI is not a fortune teller. Aim for problems that are possible now but would take many hours for many people to solve. If a person can’t perform the task, then neither can an AI. For the people side of UX research, visit in location, in car or lunch to watch real tasks. Bring artifacts if they can’t be visited. Do not talk down to user, ask them to explain things. Write quotes instead of opinion, Take pictures, Ask open ended questions. Do not ask them to design. Do not ask for them to predict the future. People are bad at that. Do not write solutions or bug fixes and do not teach no matter how much you think you can help. Instead, Can you tell be more? Can you explain x to me? Do they have questions for you? All of these are important to learn the user journeys and to find the user’s true goal. You’ll use these as part of the data design. As part of the UX, this is also the data to use to build the personas, and map out the user journeys Step 2: “Job Posting” Purpose & Design of the model

• Set your goal
• Define benchmarks

Take your user’s user journeys and goals and work with the data scientist to line them up to the data points you have available. What data do you have available? Don’t look at the data you have; then design a product around it. this leads to a product that management wants instead of what the users want. Design for what is needed; then f...