Josh Clark speaking at An Event Apart Orlando 2018 on October 10, 2018
Just as mobile defined the last decade of digital products, machine learning is set to define the next. Learn to use machine-generated content, insight, and interaction as design material in your everyday work. Refit familiar design and UX process to work with the grain of the algorithm, to help the machines solve real problems without creating new ones. This lively and inspiring talk explores the technologies and practical techniques that you can use today—like right now—not only to make existing products better but to imagine surprising new services. The challenges and opportunities of machine learning are plenty; learn to handle this powerful new design material with care and respect.
Notes
- “It’s fun to do the impossible.” — Walt Disney
- No one cares what you think is fun
- How do we use technology the way that it wants to be used?
- How do we use technology in a way that adds meaning?
- How do we use technology to amplify human potential?
- Machine learning
- 60% of companies implementing ML now
- 18% planning to implement in 12-24 months
- No one knows what they are doing — we’re all still figuring out what this new tech is good for
- Machine learning is a design material
- What is it for? — Why team up with machine learning?
- What if we could detect patterns in anything and act on them?
- 5 ways machine learning can bring value
- Recommendation
- Prediction
- Classification
- Clustering
- Generation
- Mobile defined the past decade of design, machine learning could define the next decade
- Get cozy with casual uses of machine learning
- Clustering — Organize and group things (people, concepts, transactions) according to their affinities and differences
- Identify invisible patterns
- Our jobs will change — machines will do the tasks we don’t like to do: time-consuming, repetitive, detail-oriented, error-prone, joyless
- Minimize input, maximize value
- Let people do what they do best; let machines do what they do best
- Be smarter with questions we already ask
- Ask new kinds of questions
- Identify patterns you didn’t even know about
- The job of UX: point AI at problems worth solving
- How do we work with it? — What tools and techniques do we use?
- Match a human need to a machine skill
- Do this today:
- Use a public API
- Extend these APIs
- Use a public data set
- Build your own model
- Learn the texture of your model
- Find the right data set
- Understand the uncertainty and constraints of your system
- What is its grain? — What is machine learning’s sweet spot?
- The “grain” of machine learning:
- Weirdness
- The machines are weird
- Design for failure and uncertainty
- Our job is the set expectations and channel behavior
- BASAAP (be as smart as a puppy)
- The way we preset ourselves sets expectations
- Match language and manner to system ability
- Narrow domains
- Solve narrow problems
- Narrow problems don’t have to be small problems
- Opaque logic
- Help people form a useful mental model
- Signal your intention. Reveal your action.
- If you think people won’t like it, you probably shouldn’t do it
- Make transparency a default design principle
- Probabilistic results
- Machine learning is probablistic
- How do we convey confidence?
- Present information as signals, not as absolutes
- What’s the interface that’s true to the bot, and to the data?
- Pegged to “normal”
- The machines reinforce normal
- What if normal is garbage?
- Let’s not codify the past
- Surfacing bias allows us to act on it
- Help users know when to be critical
- Amplify human judgement, don’t replace it
- Weirdness
- The “grain” of machine learning:
- How does it change us? — How will values and behaviors shift?
- Kranzberg’s first law — technology is neither good nor bad; nor is it neutral
- If we don’t decide for ourselves, the technology will decide for us
- The more we talk to robots, the more we talk like robots
- We’re inviting the future together
- This is a time for wild ideas
Speaker Links and Resources
More on this topic from Josh
- Design in the era of the algorithm
- Systems smart enough to know they’re not smart enough
- The Juvet Agenda
- The state of UX in UI (User Defenders podcast)
- Designing with artificial intelligence: the dinosaur on a surfboard (Presentable podcast)
- The rise of artificial intelligence: how AI will affect UX design (Adobe interview)
- Making software with casual intelligence
- Stop pretending you really know what AI is
- AI first—with UX
- In a few years, no investors are going to be looking for AI startups
- “Algorithms aren’t racist. Your skin is just too dark.”
Context
- Benedict Evans: Ways to think about machine learning
- MIT Technology Review: Machine Learning: The New Proving Ground for Competitive Advantage
- Harvard Business Review: In the AI age, “being smart” will mean something completely different
- Kranzberg’s laws (PDF)
Example applications
- Slack topic search: A new way to discover and connect with just the right teammates
- The Washington Post’s robot reporter has published 850 articles in the past year
- Automated Journalism – AI Applications at New York Times, Reuters, and Other Media Giants
- Google Forms: No Machine Learning in your product? Start here
- Post-It Plus app captures Post-Its in digital form
- CycleGAN transforms photos of apples into oranges, horses into zebras
Robot designers: AI for design and code
Machine learning APIs and data sets
- Microsoft cognitive services
- Amazon Web Services (AWS) machine learning
- Google Cloud AI building blocks
- IBM Watson products and services
- Topbots: Chihuahua or muffin? Searching for the best computer vision API
Customize or construct your own model
- Lobe (user-friendly visual interface)
- Google Cloud AutoML
- Microsoft Custom Vision
Open-source data sets
- Open Images
- Mozilla’s Common Voice data set
- Google’s “Quick, Draw!” data set
- Large self-annotated corpus for sarcasm
Designing for speech interfaces
- Awni Hannun: Speech recognition is not solved
- Benedict Evans: Voice and the uncanny valley of AI
- Cooper: Talking to machines is primitive and frustrating, but it’s gonna be huge
Mental models and opaque logic
- MIT Technology Review: The dark secret at the heart of AI
- Ralph Ammer: Make me think! The design of complexity
Surveillance capitalism
- Vicki Boykis: What should you think about when using Facebook?
- Digital Content Next: Google data collection research
- Gizmodo: Facebook is giving advertisers access to your shadow contact information
Designing to counter misinformation, error and bias
- Facebook: Designing against misinformation
- Book by Cathy O’Neil: Weapons of math destruction
- ACLU: Amazon’s face recognition falsely matched 28 members of congress with mugshots
- Google’s speech recognition has a gender bias
- AI programs are learning to exclude some African-American voices
- Passport system rejects this dude’s photo for a pretty racist reason
- Google mistakenly tags black people as ‘gorillas,’ showing limits of algorithms
- Machine Bias: There’s software used across the country to predict future criminals, and it’s biased against blacks.
Manner and language as cues for confidence and interaction
- Wikipedia: Paralanguage
- Amazon: Alexa speechcons
- Google: Google Duplex
- Jeremy Keith: Google duplicitous
Machine ethics
- The Juvet Agenda
- Book by Cennydd Bowles: Future Ethics
- Mike Loukides: The ethics of artificial intelligence
- Genevieve Bell: From human-computer interactions to human-computer relationships
- The European Union’s General Data Protection Regulation