Design is [Smart] – Human-Centered Machine Learning

Google Design2017-10-17

machine learning#ml#artificial intelligence#ai#automation#human centered#human centered design#ai design#ml design#inclusive#inclusivity#inclusive design#material#design#google#google design#google ml#google ai#designers#design is#design series#design event#design events#jess holbrook#jessscon#josh lovejoy#jdlovejoy#material design#product: design#Location: MTV#Team: Scalable Advocacy#Type: Other#GDS: Yes;

3K views|6 years ago

💫 Short Summary

The video discusses the integration of machine learning and AI in UX design, emphasizing the importance of understanding how these technologies work. It explains the differences between AI and machine learning, highlighting the need for repetitive training to improve models. The importance of human biases and preferences in training algorithms is also emphasized. Josh Lovejoy stresses the significance of ethnography and face-to-face interactions in understanding individuals, advocating for a more inclusive approach in the UX community that celebrates diversity and uniqueness.

✨ Highlights

📊 Transcript

✦

Integration of machine learning and AI in UX design.

01:03

AI involves automated decision-making, while machine learning involves learned decisions.

Machine learning is likened to teaching someone, requiring repetitive training to improve models.

Feedback loop plays a crucial role in enhancing AI capabilities.

Caution against solely relying on AI to solve all problems, as human biases and preferences influence training algorithms.

✦

Importance of Ethnography and Face-to-Face Interactions in Understanding Individuals.

03:07

Lovejoy challenges categorizing people into neat groups and stresses the significance of recognizing both differences and similarities.

He advocates for a multifaceted and inclusive approach in the UX community, promoting diversity and uniqueness over homogeneity.

Embracing and respecting differences can lead to more inclusive and impactful designs in human-centered practices.

00:00[MUSIC PLAYING]

00:07JESS HOLBROOK: Our talk is entitled "Design is Smart."

00:10And so we're going to really kind of hopefully

00:12communicate to you some of the things we've learned as UXers

00:16working with machine learning and AI over

00:18about the past three years.

00:19This is really important and can be one of most daunting parts

00:22because if you, as a UXer, don't understand actually

00:24how some of this really works, you're

00:26never going to get in there into the deep conversations

00:28with engineering and product teams.

00:31JOSH LOVEJOY: When we think about AI,

00:33often it is these large, mechanistic systems--

00:36you know, an assembly line of many robots

00:38doing many things seemingly magically.

00:41Artificial intelligence is really

00:43anything where there is an automated decision being made.

00:47And what's distinct about machine

00:49learning as a subset of AI is that these decisions are

00:53learned.

00:53And really, when you talk about decisions,

00:55it's one more layer we have to unpack.

00:57Its predictions.

00:59So like the one thing to be cognizant of really

01:01of anything in the process of machine

01:03learning is to ask what prediction is being made here?

01:08If you've ever taught anyone anything ever,

01:11you already know how to do machine learning.

01:13So supervised learning with my son

01:16looks like pointing to a strawberry

01:18and saying, strawberry, then pointing to another strawberry,

01:22and saying, strawberry.

01:24It's machine learning in child format.

01:27But to train a model, you need to do that like 100,000 times.

01:31JESS HOLBROOK: One way that's really useful

01:32is we just ask people, describe the way

01:34a theoretical human expert might perform this task today?

01:38The feedback to these models as well

01:40is basically how they get better,

01:41so if they don't have good data, it's

01:43really to slow them down to actually give you better

01:45predictions in the future.

01:46There's sometimes this fantasy of like, oh, well,

01:48if we apply AI to this system, it

01:50will actually remove a bunch of the problems

01:52with human judgment, or human errors, or things like that.

01:55The problem with that though is algorithms

01:57are trained on data that's collected by and about humans.

02:00They have in them, our preferences, our opinions,

02:03and our biases are embedded into this data.

02:05JOSH LOVEJOY: Like we choose where all of the data

02:07come from.

02:08We get to choose what good looks like.

02:11And we determine those success criteria and what truth is.

02:14When raters go through and label stuff,

02:16they're doing so based on protocols

02:18that have been designed by a person with success criteria

02:21and evaluations that have been deemed correct by people,

02:25using data that has been sourced by people

02:27and originally captured by people,

02:30and then we get affected.

02:33And it's not a pipeline.

02:34It's a feedback loop.

02:36JESS HOLBROOK: A lot of times, there's

02:36kind of this notion of like, we'll just

02:38apply machine learning to it, we'll apply AI to it.

02:41None of that replaces all of the user research you need to do.

02:44None of that replaces ethnography,

02:46or interviewing people, or like god forbid talking

02:49to some people face to face.

02:51JOSH LOVEJOY: We do this little process--

02:53I find many people do this process of saying,

02:55OK, well, I'm this complex entity,

02:57but then there's everybody else and they

02:58can fit into some sort of a neat bucket.

03:01And that's not true.

03:02By calling to attention these characteristics

03:04and these differences and recognizing that everyone

03:07is that complex, my hope is that as a UX community,

03:12we can help everyone understand that the world is a lot more

03:15like this--

03:16much more multifaceted and interrelated,

03:18where we have similarities and we have differences.

03:21And being a human-centered practitioner

03:25doesn't mean hiding from these differences

03:27or trying to believe that inclusion or inclusive

03:30design is about believing everyone's the same

03:33and homogeneity is the goal.

03:35It's actually highlighting and celebrating our differences,

03:38calling them out as things to be respected and cherished.

03:40[MUSIC PLAYING]

🎥 Related Videos

What vaccinating vampire bats can teach us about pandemics | Daniel Streicker

a16z Podcast | Things Come Together -- Truths about Tech in Africa

2024 TSCRS Applications of anterior segments diagnostic instruments in cataract surgery

a16z Podcast | The Infrastructure of Total Health

The Robot Lawyer Resistance with Joshua Browder of DoNotPay

NES Controllers Explained

🔥 Recently Summarized Examples

Uncovering Corruption From Health "Experts" | Scott Carney

The Forgotten Geometry: A New Path to Unification

From Tesla to DNA: The Science of Scalar Waves - Dr. Sandra Rose Michael - Think Tank E44

Bitcoin Holders...Watch Out for Sept

The World of Sasquatch, Ep. #28: The Forest People of Spirit Lake, Part 2

View original video