Day 1- Navigating the Ethical Frontier: DesignOps Strategies for Responsible AI Innovation

ea1

— Thank you for the intro.  I’m Jay and we will jump straight in

 

ea2

— I’ll break some news
  • AI is kind of a big deal and all we can talk about in DesignOps
— When CEOs mention AI, company stock soars, and when they fail to mention it, it suffers

 

ea3

— Nvidia made AI related announcement and it’s stock increased

 

ea4

— Meanwhile, Siemens slowness in adopting AI has contributed in stock dropping 36% since June
  • Companies are under tremendous pressure to get AI out and quickly

 

ea5

— We are familiar with equation of pressure and speed leading to shortcuts
  • Add to that tech staff being cut everywhere and us having to do more with less
  • Dangerous recipes for shortcuts and mistakes when we work at breakneck speeds

 

ea6

— Not a new problem, but while tech debt is as old as software development, AI hits differently

 

ea7

— The old way had technical debt, but AI has compounding interest
  • Rushing AI to market, has bad effects on product and brand
    • If we release an AI with unfair biases to people in disabilities the damage to brand and reputation will far outlast any fixes we patch in after the fact

 

ea8

— So what do we do? There are many directions to take
  • So how to keep implementation ethical?

 

 

ea9

— So I’ll tell a story about Robert
  • He is married, with two young daughters, has an ordinary life in suburbs, works at an automotive supply company
  • But you should know he has secret that not even his mother knew about

 

ea10

— On the day before his 42nd birthday, Robert was arrested for larceny
  • Why?
    • Six months earlier, while driving home from work, he created an Instagram reel, and had dinner with wife and kids
  • So what happened?
    • For Robert, a burglary happened 25 miles away and captured video surveillance was analyzed by police with an AI software called DataWorks+, which explicitly doesn’t measure its system for accuracy or bias
      • Police ran images through DataWorks+ and Robert’s driver license photo was a match
    • Robert then got a call from unidentified number and picked up the phone
      • The caller identified himself as police and asked Robert to run himself in
      • Robert refused to do it, and thought the whole call was a hoax
    • Later in the day the police came with a warrant to his home
      • Robert’s wife called Robert up, and Robert pulled in to driveway, only to be arrested by the police in broad daylight while his kids, wife, and neighbors watched
  • Finally, 18 hours later, police pulled Robert into an interrogation room and ask him to confess to the robbery
    • Robert held the photo up to his face and said the photo was not  of him — police realized this, and said it looked like the computer got it wrong
  • It took 12 hours and $1,000 for Robert to be released.  But the consequences were severe
    • Neighbors treat him differently
    • Kids don’t trust police
    • His boss advised Robert to keep events to himself
  • Shame of being wrongfully arrested, left Robert so shook-up he wouldn’t tell his mother about it
— This was a consequence of poor police work, yes, but also the AI system refusing to measure its system for accuracy or bias

 

ea11

— I believe we should insert ourselves into the story, so that these things don’t happen as we push out AI out the door as quickly as possible
  • AI is even now discriminating against women in hiring practices and people with disabilities
— We are the solution to prevent this from happening further

 

ea12

— I’ll move onto the solution now

 

— Parties are fun
  1. I don’t like attending parties, but appreciate being invited and what to bring to the party
  2. DesignOps role is to ensure the right people are at the party

 

ea13

— We need to ask if we have a team in AI process to provide variety of perspectives,
  • If no, send out those invites
— Disclaimers
  • I vew these roles as hats people wear, as there is overlap with goals
  • Make sure roles are all represented

 

ea15

— On this first slide
  • We need legal expert to protect people from turmoil and who is familiar with fine print

 

ea16

— To bring the AI to life we need the Machine Learning Engineer

 

ea17

— This slide comes with a story
  • In my interaction with ChatGPT, I figured out how to eventually talk with the AI
  • Saw potential for business applications and asked ChatGPT to ask for risk factors. It’s answer:
    • The number one answer to find use case or use cases to address
  • I followed up with: Who is needed for use cases for AI?
    • The number one answer is UX research

 

ea18

— Domain experts have SMEs, and this is area where AI can make a difference
  • BA takes info from UXR and domain experts and turn into business requirements and objectives for the AI

 

ea19

— For ethical AI integration, you don’t just need an ethicist
  • You need a data scientist to make data the clean

 

ea20

— You also need a data engineer to provide all the data
  • A mystery card remains

ea21

— That mystery person is the DesignOps leader
  • Focus on superpowers brought to conversation
—  A quick story on AI bias
  • I used MidJourney to create the images in this deck, but noticed the following when I asked the AI to generate the image of a CEO
    • I got the image of a CEO sitting at a desk and only four white men were provided
  • I refreshed the prompt for ‘men and women’ but they were all white
  • I then asked the AI to give me pictures of high status jobs like ‘doctor’, ‘nurse’, ‘teacher’
    • Combination of white men and women
  • For low-status job like fast fastworker though
    • Found minority representation under fast-food worker
  • Midjourney’s response was as follows:
    • Dataset was based online, which was stereotyped, and Midjourney was working on it
      • Dataset problem applies to the Internet as a whole

 

ea22

— So let’s ask questions that you should ask when launching a product like a) what problem we are trying to solve and b) what questions are essential

 

ea23

— Ask the following (we don’t need to know answer to questions, but need to ask them)
  1. How is data we are using supporting problem we are solving?
  2. Where is data coming from?
  3. Is the data tested for biases?
  4. How will we address these biases?

 

ea24

— What measures do we have to monitor and evaluate data (i.e. Midjourney)?

 

ea25

— And how will we incorporate user feedback and real-world insights into our AIs ongoing development?

 

es26

— So we’ve coralled people identified problem, asked ethical AI questions, and now time to build

 

ea27

— A rock solid foundation lets you create ethical prototypes
  • Prototype any new product features and how it holds up against user personas and questions for biases and plan for anything that comes up
  • Proactive approach will save you time, technical debt, and back-end work

 

ea28

— Then take the prototype for a test drive
  • Similar to how auto manufacturers test cars for crash speeds

 

ea29

— Need to ask what is ethical stress testing
  • The AI is subject to simulated scenarios that are ethically challenging
    • i.e. Autonomous vehicles needing to choose between passengers and pedestrians
  • Ethical stress testing will make sure results align with norms

 

ea30

— User-centric testing to focus on AI
  • How does AI respond to different accents, genders and cultural contexts?
  • Cater specifically to what you are developing and think of broad range of people that can encounter AI

 

ea31

— Finally iterate ethically

 

ea32

— Remember being in grade school and learning things from trusted sources, and learning things from untrusted sources like your friend Taylor?

ea33

— This is where ethical iteration comes
  • AI is tech debt with compounding interest, as AI doesn’t just exist, but continually learns how to be more useful and helpful with users
  • Technology and society are evolving quickly
— As you refine products, keep gathering data, user feedback, and real-world insights for improving accuracy and aligning with ethical goals
  • Create processes for cycle of improvement
  • Fair inclusive and ethical considerations

ea35

— So  let’s go full circle, going forward
  • You are then doing your part in making sure what happened to Robert Williams doesn’t happen to others
    • Your expertise , passion, and commitment matter here

 

ea36

— DesignOps thanks you as we map our critical role for ethical AI innovation
  • Ethical AI is not a buzzword, but Ethical AI won’t happen on its own
  • We are the solution for a responsible technological future

 

Q&A
  1. How have you navigated pushback your received from unethical datasets or outputs from AI?
    1. Sharing examples of unethical AI, makes the case for why to do this
  1. What about the intentionality of people invited along with disabilities?
    1. Yes, include diversity as part of invites, and classify who will encounter AI, under the ethicist card
    2. Lot of people in the world, but have access to only a few– so need to think how we can avoid missing out
  1. What to do in org that considers AI as important, but leadership set-up tiger team and you are not included?
    1. I feel engineering led orgs will try to take control of AI, so shoehorn self into conversation and show impacts of what happens to brand
    2. Ramifications of damaging AI , is that you can’t put toothpaste back in tube, and damaging impacts now part of brand
      1. Escalate conversation until you reach someone who cares about it