Ppl can't seem to stop talking about LLMs having their iPhone moment. But what if the LLM isn't the iPhone? More like the touch screen?
Remote
Niloufar Salehi
@Niloufar@hci.social
mastodon 4.7.3Human Computer Interaction, Human Centered AI, health communication, restorative justice
Assistant professor at UC, Berkeley School of Information
0 Followers
0 Following
19 Posts
Joined October 28, 2022
Website:
So the whole rich people buying social media the second they don't like what is on it is foreshadowing what's going to happen to all the big models isn't it?
Replying to
We interviewed Muslim Americans who are public figures and have a large following on social media. We found that their work is severely limited by the violent harassment that they face which stretches over time (e.g. weeks, months, repeatedly every 10 seconds) and space (mentions, tags, emails, dms). We argue that this harassment is not accidental or an anomaly, but a coordinated practice of gaining and maintaining racial and spatial ownership by dominant (white) publics.
Design is the art of being wrong safely
https://spavel.medium.com/design-is-the-art-of-being-wrong-safely-7575b0c395c2
Replying to
We end with a discussion of the role that platform design and policy -- disembodied, perceived as white, enabling frictionless sharing of text and images to large audiences for all users -- plays in enabling this harm and what alternatives based on restorative and transformative justice might look like.
Many models produce outputs that are hard to verify for an end user.
🏆 Our new #emnlp2023 paper won an outstanding paper award for showing that a secondary quality estimation model can help users decide when to rely on the model output.
We ran a controlled experiment showing that a calibrated quality estimation model can make physicians twice better at correctly deciding when to rely on a translation model output.
Replying to
Defense teams face challenges assessing and contesting reliability due to difficulties (a) navigating how these tools are developed and used, (b) overcoming judges and jurors' non-critical perceptions of the technology, and (c) gathering technical expertise. Technical and policy interventions are urgently needed to help public defenders overcome these challenges.
(Beyond) Reasonable Doubt: Challenges that Public Defenders Face in Scrutinizing AI in Court https://arxiv.org/abs/2403.13004 #chi2024
Replying to
Surprisingly (to us) none of the people we interviewed used platform moderation tools. In fact they considered content moderation mostly useless, sometimes harmful. This was partly because the sheer amount of the abuse meant that it would have to be someone's "full time job" just to click report all day. But also because their past experiences with platform moderation had led them to not trust that it would actually do anything for them.
Replying to
Overall, our participants work had them constantly dealing with their content being mass reported and taken down, their hashtags being taken over, and their mentions being filled with violent imagery and threats, or as one participant put it: "constantly dealing with the possibility that there's just so much more keyboard power on the other side"
@katestarbird@mstdn.social really interesting, we make a similar argument here about harassment interventions: https://niloufar.org/wp-content/uploads/2023/03/sustainedonlineharmmuslims-CSCW2023.pdf
Replying to
We ran a controlled experiment in which 65 physicians were shown sentences from a real emergency discharge instruction and their Google translations. The group that saw a QE model calibrated to medical communication was almost two times better at deciding when to rely on the translation than a baseline (measure is confidence weighted accuracy).
Replying to
Replying to
@msbernst great article thank you for sharing. I feel like HCI is a bit hard to teach so it's great to read about your approach and how it's worked. For "getting students to reflect on their own ideas," I have a week toward the end where they pick three of the Tarot cards of tech and apply the questions to their own final project. It's one if the most engaging assignments. They write their answers and also how they would continue the design process to tackle the new challenges they discovered.
Replying to
@evanmpeck @msbernst Evan I really liked how you structured your course and use it in my own (with attribution of course).