Finding the Problem to Solve: Human-centered AI
summary
Advocates a UX-first claim: start by validating a user problem before building AI, then choose whether to augment decisions or automate predictable tasks. Shows a concrete case: Planta: plant identification + care schedules that uses computer vision for IDs, scheduled watering reminders and a snooze flow tied to measured plant health improvements. Recommends weekly user tests, Google PAIR worksheets and Wizard of Oz minimal-fidelity testing to measure value, and defines reward functions to monitor false positives/negatives.
AI-written summary · about
chapters (12)
- 0:00Welcome and introduction
- 1:04Speaker introduction
- 2:24Why human-centered AI matters
- 4:10Design thinking for AI products
- 7:41Key UX questions for AI
- 9:08Does the problem need AI?
- 12:17Behavior patterns and success
- 15:28Automation vs augmentation
- 21:54Choosing the right AI approach
- 23:21Handling AI failures
- 26:14Key lessons and takeaways
- 27:38Q&A session
related talks
28mHuman Computer Interaction: The UX of AI
24mA Call for Human-Centered AI
25mArtificial intelligence & design: imagining the future of UX
54mEthical AI Development: Leveraging Human Centered Design for Inclusive and Trustworthy Solutions
12mDesigning AI To Scale Human Thought
37mThe Augmented Engineer: Evolving Sensible Defaults through Human-Centric AI
23mUser Experience Now Includes Machine Experience
36mThe UX of AI: Making AI-Powered Apps Your Users Don't Hate