Design UX for AI
•
0 likes•130 views
Debapriya Basu is a Principal Product Manager at Zillow who specializes in AI products and technology. The presentation covers designing user experiences for AI-infused systems, including principles of design such as embracing customer needs and building trust. Guidelines are provided for user interaction stages like before first interaction, during interaction, and when things go wrong. A framework is outlined for the design process, including determining how AI can help, identifying user needs, collecting and analyzing data, designing prototypes, and incorporating feedback.
Report
Share
More Related Content
What's hot
What's hot (20)
Similar to Design UX for AI
Similar to Design UX for AI (20)
Recently uploaded
Recently uploaded (20)
Design UX for AI
- 1. DESIGN UX FOR AI DEBAPRIYA BASU PRINCIPAL PRODUCT MANAGER @ ZILLOW
- 2. HELLO! I am Debapriya Basu I specialize in AI products and technology You can find me at https://www.linkedin.com/in/debapriyabasu/ Current: PM@Zillow Group Former: PM@Microsoft Previous Product School talk: Lead AI (ML) incubations as a PM 2
- 3. WHAT I WILL COVER Design for AI infused systems ◎ Principles of design ◎ Guidelines for user interaction stages ◎ Framework of design process 3
- 5. EMBRACE CUSTOMER NEED ◎ Need : long (sometimes painful) process to ◎ learn how much is a home worth in current local markets ◎ Zestimate ◎ Started as home estimate for next one year ◎ Updated to reflect quality of home (using computer vision) 5 ZESTIMATE
- 6. BUILD TRUST ◎ Is it well intentioned ◎ Does it perform the way we envisioned ◎ Is it safe to use. Is it correctly built 6 ref
- 7. ETHICAL AI Privacy AI models should be built within guardrails to ensure users cannot be identified or their details inferred from model output Researchers were able to identify Netflix users by correlating anonymized test data provided in the Netflix Prize competition with publicly available IMDB movie review database. 7 Fairness AI/ML models used for making decisions or predictions should not be biased with respect to protected attributes (latent bias) such as race, gender and sexuality. It should be aware and counter impacts of interaction and selection bias also. A MIT study Project Gender Shades uncovered the bias that facial analysis technologies have a heavy bias towards white males. JoyAdowaa Buolamwini foundedAlgorithmic Justice League, an organisation that looks to challenge bias in decision making software
- 8. AMPLIFY HUMAN CAPABILITIES WITH AI ◎ Google Pixel Buds integrated with Google assistant ◎ Personalized recommendations ◎ Automated personal communication ■ (Smart Reply) ◎ offloads monotonous tasks ■ Set timers ■ Control IoT devices 8 GOOGLE ASSISTANT here
- 10. STAGES OF USER INTERACTION ◎ Before first interaction ◎ During course of interaction ◎ When things go wrong ◎ Over continued repeated interaction 10
- 11. BEFORE THE FIRST INTERACTION 11 Clarify what the system can do Clarify how well the system can do what it promises to do ref ref
- 12. DURING THE INTERACTION 12 Show contextually relevant insights Match relevant social norms Mitigate social bias Time services based on context
- 13. WHEN THINGS GO WRONG 13 Scope services when in doubt Clarify what the system did Allow efficient correction Allow efficient dismissal Allow efficient invocation UX has a critical role to play here
- 14. OVER CONTINUED REPEATED INTERACTION 14 Remember recent interactions Update and adapt cautiously Encourage granular feedback Learn from user behavior
- 15. OVER CONTINUED REPEATED INTERACTION 15 Notify users about changes Provide global controls Convey consequences of user action
- 17. STAGES 17 Determine how AI can help Identify user needs Train ModelsCollect Data Test prototypesDesign to meet user’s mental model Build Trustworthy AI Explain AI Handle failures gracefully Incorporate Feedback
- 18. IDENTIFY USER NEED + HOW CAN AI HELP RESEARCH Understand the user need through user research, opportunity analysis, competitive analysis OBJECTIVE OR REWARD FUNCTION What kind ofAI technology will help solve. What is to be prediction or concluded 18 EXPERIENCES Identify hero experience to ideate SUCCESS DEFINITION Define success in terms of online metrics ALIGN Align stakeholders to a common understanding of the user problem and proposedAI infused experience
- 19. COLLECT DATA + TRAIN MODELS DATA ◎ 1st Party vs 3rd Party ◎ Privacy and Security ◎ Fresh ◎ Unbiased ◎ Representative ◎ Relevant ◎ Missing data ◎ Noisy data ◎ Inconsistent 19 AI TECH ◎ Machine learned ◎ Supervised ◎ Unsupervised ◎ Reinforced ◎ Modeling Techniques ◎ Deep learning ◎ NLP ◎ Ensemble Learning ◎ Pre-trained Models DESIGN FOR LABEL GENERATORS Accommodate diversity of raters and their context and incentives
- 20. DESIGN TO MEET USER MENTAL MODELS UNDERSTAND MENTAL MODEL ◎ User group and their primary goal, ◎ novice user’s preferred flow, ◎ expert user’s preferred flow, ◎ current mental model, 20 INFLUENCE MENTAL MODEL ◎ Clarify future mental model, ◎ Create bridge from current to future ◎ explain cause and effect ofAI model
- 21. TEST MENTAL MODEL ◎ Onboard users ◎ Ask users what they thought the product will do ◎ Study their reactions to wireframes/prototypes ◎ Learn about their expectations about using theAI experience 21 PRODUCT ◎ Guerilla testing ◎ Wizard of Oz testing
- 22. 22 TRANSPARENT + EXPLAINABLE + TRUSTWORTHY AI MODEL CONFIDENCE DATA SOURCES MODEL FEATURES FAIRNESS
- 23. INCORPORATE FEEDBACK IMPLICIT (critical forAI) ◎ Subscribed to channel (+) ◎ Add to playlist (+) ◎ Add to watch later (+) ◎ Shared video (+) ◎ Watched till the end (+) ◎ Bounced off after few secs (-) ◎ Clicked on a related recommendation after few secs on video (-) ◎ Watch videos in bits and pieces (-) 23 EXPLICIT
- 24. DEGRADE GRACEFULLY IN CASE OF NON HAPPY PATH RESULTS OR SYSTEM FAILURES ◎ Category suggestions ◎ Alternate queries ◎ Personalized recommendation ◎ Helpful links ◎ Popular products ◎ Popular categories ◎ Contextual ads 24
- 25. WHAT I COVERED Takeaways for AI powered product design ◎ What are the cornerstone principles ◎ How to design for user interaction stage ◎ What is the framework for the design process Please reach out with your comments and thoughts at tutulpriya@gmail.com 25
- 26. REFERENCES ◎ https://pair.withgoogle.com/guidebook/ ◎ https://www.microsoft.com/en-us/research/uploads/prod/2019/01/Guidelines-for-Human- AI-Interaction-camera-ready.pdf ◎ https://www.media.mit.edu/projects/gender-shades/overview/ ◎ https://medium.com/myplanet-musings/10-ui-patterns-for-a-human-friendly-ai-e86baa2a4 471 ◎ https://uxdesign.cc/guidelines-for-infusing-artificial-intelligence-to-products-bbbcf7b928f 7 ◎ https://becominghuman.ai/ux-design-for-implicit-and-explicit-feedback-in-an-ai-product- 9497dce737ea ◎ https://techcrunch.com/2018/12/07/google-translate-gets-rid-of-some-gender-biases/ ◎ https://www.nngroup.com/articles/recommendation-guidelines/ ◎ https://www.pcworld.com/article/3393186/microsoft-ideas-words.html ◎ https://baymard.com/blog/no-results-page ◎ https://www.blog.google/products/assistant/stay-top-your-day-proactive-help-your-assistant/ 26