Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
SlideShare a Scribd company logo

UX STRAT Online 2021 Presentation by Gideon Simons, Zinier

•
•

These slides are for the following session presented at the UX STRAT Online 2021 Conference: "Progressive Design with AI" Gideon Simons Zinier: Senior Director of Product Design & User Research

1 of 29
Download to read offline
Progressive Design with AI GIDEON SIMONS - DEC 2021 @ UXSTRAT
Gideon Simons INSPIRATIONAL QUOTE: SR. DIRECTOR OF PRODUCT DESIGN AND UX RESEARCH @ ZINIER PREVIOUSLY UXR Agency, Innovation Lab, Mediacorp, IBM, SeeSaw, CTO x2 “I'd rather have a bottle in front of me than a frontal lobotomy” - Tom Waits EMOJI 🦄
It’s 2018, robots haven’t taken my job.. phew! my mirror is still dumb, drones are not delivering my packages and the taxi I took had a chatty human driver. From a talk I gave 3 years ago at UXSEA conference!
Designers Don’t be afraid of AI!
AI has a lot of fancy terminology such as ‘confusion matrix’ which can be intimidating and also… a bit confusing..
In this talk you will learn an easy framework for designing amazing experiences with both dumb and smart AI
Progressive design with AI basically means that the user experience grows together with AI getting better. At the heart of the framework are these two important factors. AI Performance How and when to use the AI AI Confidence What do you do with the prediction THE FRAMEWORK
Let’s start with AI Performance
AI Performance is how well can the AI detect True Positives and False Negatives (on your data)
It’s a Lion False - Positive It’s a Lion True - Positive Isn’t a Lion True - Negative Isn’t a Lion False - Negative Confusion Matrix explained with cats and lions
As good as flipping a coin! 50% - 85% below 50% 75% and Above Let’s do it above 85% Unreliable Acceptable Performance Score (F1) Value Scale * Note that these are rule of thumb numbers and may vary between use cases
Performance Score is 89% The AI Model can predict task demand and manpower capacity with 89% accuracy!
Performance Score is 70% Which isn’t very reliable but maybe still useful as a softer insight
Performance Score is 44% AI is completely useless but we still want to give something useful!
Next is AI Confidence AI Performance How and when to use the AI
AI Confidence Is how confident AI is about a given prediction (on a new input)
Confidence Score explained with a lion cub App AI App AI How confident are you about that? Is this a Lion? I am about 86% confident that this is a Lion.. And about 35% that it’s a cat.. And about 5% that it’s a coffee mug.. Hmm… yes! * Tip - If your AI model doesn’t have this out of the box then work with devs and data scientists to create one together
Predicted No 25% - 85% below 25% 75% and Above Let’s do it above 85% Not sure! Predicted Yes Confidence Score Value Scale * Note that these are rule of thumb numbers and may vary between use cases
Confidence Score is 89% AI is confident that this is an Antenna and so we can help nudge the reviewer to Pass!
Confidence Score is 52% AI is not very confident, lets ask the user to help teach the AI
Confidence Score is 12% for Cabinet AI is doing a good job telling you that the thing in the photo isn’t a Cabinet!
And Lastly - Some more helpful guidelines AI Performance How and when to use the AI AI Confidence What do you do with the prediction
Gain trust from users Progressively! Not immediately..
Don’t Overfit solutions Like what 80% of all blockchain startups do 😁
Be Ethical! No life-death use cases Avoid putting some groups of people at a disadvantage to others
Acknowledge biases.. Don’t be afraid to let your users know about your weaknesses and ask users for help!
“When AI solves the wrong thing, is built on faulty data, operates with lack of purpose, is misaligned with people’s needs, creates no room for feedback, doesn’t consider context, oversimplifies nuance, it is likely to fail, and worse, to harm.” - Jennifer Comiskey From: The Future of AI is People-Centered - which you can find on Medium Super 3d illustrations by Alzea Arafat!
Questions? 󰢧󰢨🙋
Thank You!

More Related Content

UX STRAT Online 2021 Presentation by Gideon Simons, Zinier

  • 1. Progressive Design with AI GIDEON SIMONS - DEC 2021 @ UXSTRAT
  • 2. Gideon Simons INSPIRATIONAL QUOTE: SR. DIRECTOR OF PRODUCT DESIGN AND UX RESEARCH @ ZINIER PREVIOUSLY UXR Agency, Innovation Lab, Mediacorp, IBM, SeeSaw, CTO x2 “I'd rather have a bottle in front of me than a frontal lobotomy” - Tom Waits EMOJI 🦄
  • 3. It’s 2018, robots haven’t taken my job.. phew! my mirror is still dumb, drones are not delivering my packages and the taxi I took had a chatty human driver. From a talk I gave 3 years ago at UXSEA conference!
  • 5. AI has a lot of fancy terminology such as ‘confusion matrix’ which can be intimidating and also… a bit confusing..
  • 6. In this talk you will learn an easy framework for designing amazing experiences with both dumb and smart AI
  • 7. Progressive design with AI basically means that the user experience grows together with AI getting better. At the heart of the framework are these two important factors. AI Performance How and when to use the AI AI Confidence What do you do with the prediction THE FRAMEWORK
  • 8. Let’s start with AI Performance
  • 9. AI Performance is how well can the AI detect True Positives and False Negatives (on your data)
  • 10. It’s a Lion False - Positive It’s a Lion True - Positive Isn’t a Lion True - Negative Isn’t a Lion False - Negative Confusion Matrix explained with cats and lions
  • 11. As good as flipping a coin! 50% - 85% below 50% 75% and Above Let’s do it above 85% Unreliable Acceptable Performance Score (F1) Value Scale * Note that these are rule of thumb numbers and may vary between use cases
  • 12. Performance Score is 89% The AI Model can predict task demand and manpower capacity with 89% accuracy!
  • 13. Performance Score is 70% Which isn’t very reliable but maybe still useful as a softer insight
  • 14. Performance Score is 44% AI is completely useless but we still want to give something useful!
  • 15. Next is AI Confidence AI Performance How and when to use the AI
  • 16. AI Confidence Is how confident AI is about a given prediction (on a new input)
  • 17. Confidence Score explained with a lion cub App AI App AI How confident are you about that? Is this a Lion? I am about 86% confident that this is a Lion.. And about 35% that it’s a cat.. And about 5% that it’s a coffee mug.. Hmm… yes! * Tip - If your AI model doesn’t have this out of the box then work with devs and data scientists to create one together
  • 18. Predicted No 25% - 85% below 25% 75% and Above Let’s do it above 85% Not sure! Predicted Yes Confidence Score Value Scale * Note that these are rule of thumb numbers and may vary between use cases
  • 19. Confidence Score is 89% AI is confident that this is an Antenna and so we can help nudge the reviewer to Pass!
  • 20. Confidence Score is 52% AI is not very confident, lets ask the user to help teach the AI
  • 21. Confidence Score is 12% for Cabinet AI is doing a good job telling you that the thing in the photo isn’t a Cabinet!
  • 22. And Lastly - Some more helpful guidelines AI Performance How and when to use the AI AI Confidence What do you do with the prediction
  • 23. Gain trust from users Progressively! Not immediately..
  • 24. Don’t Overfit solutions Like what 80% of all blockchain startups do 😁
  • 25. Be Ethical! No life-death use cases Avoid putting some groups of people at a disadvantage to others
  • 26. Acknowledge biases.. Don’t be afraid to let your users know about your weaknesses and ask users for help!
  • 27. “When AI solves the wrong thing, is built on faulty data, operates with lack of purpose, is misaligned with people’s needs, creates no room for feedback, doesn’t consider context, oversimplifies nuance, it is likely to fail, and worse, to harm.” - Jennifer Comiskey From: The Future of AI is People-Centered - which you can find on Medium Super 3d illustrations by Alzea Arafat!