Remote

Kevin Feng

@kjfeng@hci.social

mastodon 4.7.3

PhD student at the University of Washington. https://kjfeng.me

0 Followers

0 Following

22 Posts

Joined May 11, 2022

Kevin Feng @kjfeng@hci.social

· 47mo ago

Hi! Do you have thoughts on how your feed should be organized 🤔? We are researchers at the University of Washington exploring end user tools for social media feed curation. We're conducting a study to learn more about the factors users take into consideration when evaluating the saliency of content in their personal feeds. This study will take approx. 30–45 mins and you'll receive a $15 gift card for participation. Learn more and sign up here! https://forms.gle/QcmcaP6tcsAmmUX6A

23

2

5

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

We partnered with folks at the Coalition for Content Provenance and Authenticity (C2PA) and Adobe's Content Authenticity Initiative (CAI) to create "provenance-enabled" prototypes of Twitter-like feeds, simulating UX outlined in the C2PA provenance standard. We then populated these prototypes with diverse content and deployed them to 595 ppl in the US and UK. 2/n

3

1

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

We found that exposing users to provenance info lowered trust and perceived accuracy of content that had major visual edits (e.g., composing two images together). In many cases, provenance nudged users closer to what the images and videos originally set out to represent. 3/n

3

5

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

In an AI age where media can be created & manipulated seemingly at a snap of a finger, provenance is a promising approach to empowering users to make informed decisions about media online. We're excited by the novel interactions & impact that provenance standards can bring. 7/n

3

1

0

0

Kevin Feng @kjfeng@hci.social

· 23mo ago

Are you a researcher in CS or a CS-adjacent field who uses planning docs to organize your research? Want to try our new AI agent-powered system for working with research planning docs in a paid user study? Details and sign up here! https://forms.gle/TGzKFaFYs6x7kvSy7

1

0

0

0

Kevin Feng @kjfeng@hci.social

· 24mo ago

Replying to

@hci@micro.blogs.princeton.edu thanks so much for having me, it was a lot of fun!

1

0

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

But—there's a catch. Enabling provenance adds another dimension of credibility: how valid is the actual provenance info? We tested different states of provenance credibility and found that they had a significant impact on users' credibility ratings of *the content itself*. 4/n

2

1

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

For example, a perfectly authentic image can be seen as less credible because someone tampered with its provenance info, making the info invalid w/o affecting the image visually at all. This new method of altering perceptions of an image makes provenance a double-edged sword. 5/n

2

3

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

We think it's important to disentangle concepts of media vs. provenance credibility going forward with thoughtful design choices and user education. We also see potential areas of collaboration with the computer vision community to better manage & display provenance info. 6/n

2

2

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

@socialfutureslab @Mor Also keep an eye out for our forthcoming paper that dives deeper into many of the ideas I talked about 👀

1

0

0

0

Kevin Feng @kjfeng@hci.social

· 40mo ago

Replying to

It was a blast to see this work come into fruition with C2PA and Adobe's CAI, and ofc, @axz. Check out the fantastic initiatives that made this work possible: C2PA: https://c2pa.org/ Adobe's CAI: https://contentauthenticity.org/ 8/8

C2PA | Providing Origins of Media Content

Coalition for Content Provenance and Authenticity (C2PA)

C2PA | Providing Origins of Media Content

Enhance digital safety through the use of content authenticity tools. C2PA provides a way to ensure content transparency by analyzing the origin of media.

1

0

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

We conducted a task-based design study with 27 UX practitioners where they designed a proof-of-concept food classifier app while using Google's Teachable Machine to train, test, and tweak ML models as part of their UX workflows. Even though the models were super simple and may not end up in a production-ready app, their interactive experimentation with ML was important in 3 key ways: 2/n

0

1

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

Tying ML capabilities and underlying data to user goals. Participants could quickly validate or reject assumptions they had about ML, experience model limitations first-hand, and reason about ML alignment with user needs to identify new design potential. 3/n

0

1

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

Envisioning affordances to enhance user interaction with ML. Participants took inspiration from Teachable Machine to visualize model outputs to users. To increase predictive accuracy, they designed UIs for users to provide higher-quality data + feedback to the model. 4/n

0

1

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

Surfacing ML risks and potential harms. Participants actively engaged in considering ethical issues of using ML for food classification, as well as in user-facing apps more generally. Some concluded that ML was not an ideal solution in this particular design problem. 5/n

0

4

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

Zooming out, we found that participants—even those with zero prior ML experience—were able to surface insights that aligned with previously established human-AI interaction guidelines. This is commendable given the relatively short timespan (< 2 hours) of our study. 6/n

0

3

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

We also noticed that participants' mental models of ML aligned closely with that of interactive machine teaching. However, IMT aims to help end-users build models. How can similar concepts help reason about ML UX? We introduce research-informed MT as a conceptual guide. 7/n

0

1

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

Overall, the benefits of interactively crafting simple ML models alongside UI designs are plenty. We see exciting potential in tools that allow UX practitioners to rapidly create and experiment with ML as a design material right there in their design tools. 8/n

0

1

0

0

Kevin Feng @kjfeng@hci.social

· 43mo ago

Replying to

As a closing note, this was a fun first project of grad school and my first paper as a sole student author! Very grateful for my advisor David McDonald for showing me the ropes of research and the many engaging discussions throughout. See you all in Sydney! 🇦🇺🦘 9/9

0

0

0

0