Thread by @theiaconf

#IAC19 Keynote by Ha Phan — Signal vs. Noise @hpdailyrant

.@hpdailyrant begins by talking about a photo of her father. The photo was taken by a manual. A manual trigger is a distinct intent of the user. Now photos are automatically generated and decided by a computer. What signals the computer about our photos? #IAC19

Computers (ie our phones) get signals from us all the time. All data starts out at noise. How can we measure intent if there’s no button to push? “Signals begin as the tallest leprechaun.” @hpdailyrant #IAC19

A signal is a piece of data significantly strong, rising above the noise of the system. The strong a signal, the more accurate and relevant it is. @hpdailyrant #IAC19

.@hpdailyrant defines intelligence as the sensemaking of many signals. Interactions are receptors that tunes the signals to train AI. Humans have to trigger the true/false signal for AI to learn from. #IAC19

On the problem of over capture in handling AI models, what comes first? Capability or experience? How does UX work in emerging tech? @hpdailyrant #IAC19

Engineers are capability-focused, “what can the tech do?” Designers are user-focused, “what are users motivated by?” @hpdailyrant #IAC19

GoPro encourages videographers to “set it and forget it” but you get huge amount of footage. How do you curate that? @hpdailyrant #IAC19

How do you create a story out of 3 hours of footage, in 5 minutes? @hpdailyrant did it at GoPro. They bet they could use AI to choose the best footage. #IAC19

Here’s how AI did it, but sensing changes in audio, changes in background image, and faces. @hpdailyrant #IAC19

UX focused on emotional storytelling. “How might we make storytelling quick, easy, and fun?” @hpdailyrant #IAC19

GoPro had a storytelling team to help users create awesome videos without needing advanced video editing skills. Instead of focusing on the video, they focus on music as the timeline. “Music is the current that carries the story.” @hpdailyrant #IAC19

For the GoPro storytelling team, they used music highlights to cut the video automatically. You just choose your music, the clips and highlights, and it would cut it for you. (Less than 5 seconds per shot is the interest sweet spot.) @hpdailyrant #IAC19

The music syncing tech was highly technical, but was user-driven. Humans had to label hundreds of tracks, to set a roadmap for the AI to analyze the music they used. @hpdailyrant #IAC19

There was a balance between engineering and design - a balance between satisfaction and effort when it came to the music synching tech. @hpdailyrant #IAC19

Everyone sees the tech, but @hpdailyrant saw the architecture of what a highlight was. Humans still have to identify what it is so computers can use it. The UI is the trigger and receptor. Be intentional about the trigger and the data, the signals it produces. #IAC19

When @hpdailyrant joined Pluralsight, she started with research to understand the user motivations. There are the follow-throughs and the cherry-pickers. Pluralsight catered to follow-throughs, while almost half were cherry-pickers. #IAC19

There was a dichotomy of behaviors with Pluralsight users. By scoring behaviors, they could predict what a user would do. @hpdailyrant #IAC19

But Pluralsight missed behaviors of users who grew within the system, so they didn’t exist. If behaviors weren’t standardized, it didn’t exist. By missing data, AI is inherently biased. @hpdailyrant #IAC19

Human behaviors fuel everything. Humans have to curate, the AI is just helping. There is no full automation for anything yet. @hpdailyrant #IAC19

Value and meaning is defined by behavior. (We have to measure that accurately!) The human is the sensor. @hpdailyrant #IAC19