Vikram Mohanty

Bright orange graphic from Duolingo featuring the animated owl mascot with glowing eyes and outstretched wings, celebrating a “800 day learning streak!” in bold white text.

This trip also marks 🇯🇵 800 days of Japanese on Duolingo! 😂Can confidently order 🥤 & 🍚

Looking forward to meeting y’all at #CHI2025! 👋

Would love to chat about HAI collaboration, startups, new career adventures, or just swap chaotic ✈️ stories and explore Yokohama 🗺️🍜

Excited to present this work done with Jude Lim &
@kurtluther.bsky.social

🗓 Talk: Tues, April 29 — Human-Agent Interaction @ #CHI2025
📍 Annex Hall F204

What Lies Beneath? Exploring the Impact of Underlying AI Model Updates in AI-Infused Systems AI models are constantly evolving, with new versions released frequently. Human-AI interaction guidelines encourage notifying users about changes in model capabilities, ideally supported by thorough b...

Across both studies, users struggled to notice model updates — even when the changes shaped their behavior and perception!

People formed folk theories, misattributed differences, and often preferred models based on feel over facts.

📄 Learn more here: arxiv.org/abs/2311.10652

Screenshot comparing facial recognition results from the old and new models in Civil War Photo Sleuth. Both models return Alfred Pleasonton as a match, but the old model shows 289 results while the new model shows only 2. Visual side-by-side highlights differences in result volume and ranking.

📓 Study 2:
A 2-week diary study with active Photo Sleuth users.

This time, users could toggle between the two models.

Which model did they prefer? Why?
How did they make sense of the differences?
And what folk theories did they build to explain what the AI was doing? 🧠🤖

Diagram showing Study 1 setup: Each of 8 trials asks participants to find a match for a query photo using a facial recognition model. Trials show different silhouettes with question marks, and icons of a blue robot (old model) or green robot (new model) are shown beneath. The model used in each trial is randomly assigned.

👥 Study 1:
An online experiment with 252 participants.

Across 8 trials, users interacted with two different facial recognition models—but were NOT told if (or when) the model had changed.

Could they tell the difference? Did it affect their behavior or performance?

Promotional image for Civil War Photo Sleuth showing historical portraits of men and women from the Civil War era, overlaid with facial recognition landmarks. The Civil War Photo Sleuth logo with a star inside a magnifying glass appears prominently in the center.

To explore this question, we ran two studies using Civil War Photo Sleuth (civilwarphotosleuth.com), a real-world historical photo ID platform, as our testbed.

The system uses facial recognition + human expertise to ID Civil War-era portraits. ⚔️📸

Screenshot of the CHI 2025 paper titled “What Lies Beneath? Exploring the Impact of Underlying AI Model Updates in AI-Infused Systems,” by Vikram Mohanty, Jude Lim, and Kurt Luther, showing the abstract and introduction.

Heading over to @chi.acm.org means a break from online inactivity!

🚨 AI systems update constantly—but can users even tell?

In our #CHI2025 paper, we studied what happens when AI systems silently swap models under the hood. Do users notice? Do they perform better?

Spoiler: it’s complicated. 🤷

Call for Participation - ACM Collective Intelligence 2025 ⏰ Submission Deadlines and Templates Collective Intelligence 2025 offers several opportunities to share your latest research and insights. All submissions should be submitted via EasyChair by the list...

The ACM Collective Intelligence 2025 Conference is coming up. Dive into "Diverse Intelligence" & explore collective problem-solving across disciplines. Time to prepare your submission for opportunities due on May 1st: Doctoral Colloquium, CrowdCamp, Demos or Posters
ci.acm.org/2025/submit/
#ci25

Thanks Mia! :) We’re really excited to bring some more cool stuff out in the coming months!

I am recruiting a postdoc via the CBI Fellowship Program (carnegiebosch.cmu.edu/fellowships/...). Interested in designing & evaluating interactive systems for participatory AI, augmented and collective intelligence, responsible AI, or related areas? Please apply and reach out to me via email!

The course website is soooo cool!! 😍

8 years ago, I made an iOS app to just show me a cheat sheet. I don't need to type an exact currency number, I just need to know ¥20,000 is $129.80. Thus CurrentSees was created. Now it's free & public! No ads, no tracking, and open source! Works great on WatchOS too! apps.apple.com/us/app/curre...

CHI 2026 Open Call for Volunteers Take this survey powered by surveymonkey.com. Create your own surveys for free.

CHI 2026 Barcelona Open Call for volunteers is Open! Please sign up to join the team! This is for organizing and technical committees (excluding subcommittee chairs and area chairs for the papers program…that comes later). Details in the submission form. www.surveymonkey.com/r/GGPYYPD