Ruihan Gao - 2026 Three Minute Thesis (3MT) Championship Presentation at CMU — Transcript
Full transcript
- 0:01Our sixth presenter is Ruihan Gao from
- 0:03the School of Computer Science. Ruihan's
- 0:06title is Teaching AI the Sense of Touch.
- 0:11Raise your hand if you ever bought
- 0:12something online that looked great on
- 0:14the screen, but feels still totally
- 0:16different in your hands.
- 0:19Now, keep your hands up if that was the
- 0:21reason you sent it back.
- 0:23Exactly. Thank you. That happens because
- 0:26screens only show us how things look,
- 0:28but not how they feel like.
- 0:31Today, generative AI can create images,
- 0:34videos, 3D worlds in games, but they are
- 0:37missing something fundamental. Touch.
- 0:40As humans, we not only see the world, we
- 0:44feel the softness of fabric, the
- 0:46friction, the roughness of the wood, and
- 0:48also the coolness of glass.
- 0:51Touch can tell us something vision
- 0:53cannot.
- 0:55But, modern AI normally ignores it for
- 0:58two hard reasons.
- 1:00First, the scale gap. The camera sees
- 1:03the whole sweater, but the fingertip
- 1:05feels it thread by thread, almost 100
- 1:08times smaller.
- 1:10Second, unlike language, touch doesn't
- 1:13have a vocabulary. How do you quantify
- 1:15how a velvet feels like?
- 1:18My research solves both.
- 1:20We use a digital fingertip, a
- 1:22high-resolution sensor, and press it
- 1:24against object surface, and record the
- 1:27invisible heels and valleys into a
- 1:30texture map, and train the AI model to
- 1:33binds the big picture we see and the
- 1:35macroscopic details we feel
- 1:38into an integrated shared
- 1:40representation.
- 1:42The results, when you type smooth marble
- 1:45or rough linen, with the AI model can
- 1:48generate both it feels like and how it
- 1:51looks like.
- 1:53For you, that means an immersive life.
- 1:56Imagine feeling the weave of a sweater
- 1:58on your phone when you buy it.
- 2:00Or run your fingers across the stone
- 2:03wall in a video game.
- 2:05Beyond convenience, that's about
- 2:07accessibility.
- 2:09For blind and low vision individuals,
- 2:11touch isn't just a feature.
- 2:14It's their primary way of understanding
- 2:17the world.
- 2:18My work allows AI to generate tactile
- 2:21graphics, the 3D printed surfaces that
- 2:24translate images into the objects people
- 2:27can explore with their hands.
- 2:30Blind students, for the very first time,
- 2:32can see how a biological cell structured
- 2:35or touch a painting in the museum.
- 2:39We have demonstrated for rigid surfaces
- 2:41and fabric, and next we're modeling
- 2:43softness, so that medical medical
- 2:45students can play with the simulators of
- 2:48the soft skin without a
- 2:51you know, a simulator that finally feels
- 2:53real.
- 2:55The future of AI should not be limited
- 2:58to a flat glass screen. With us teaching
- 3:02the AI the sense of touch, we want the
- 3:04technology to be more sensory, more
- 3:07inclusive, and ultimately more human.
- 3:10Thank you.
- 3:11>> [applause]
- 3:13>> Thank you.
About this transcript
This page contains the full transcript of Ruihan Gao - 2026 Three Minute Thesis (3MT) Championship Presentation at CMU by Carnegie Mellon University Libraries, generated from the public captions YouTube serves with the video. The transcript has 437 words across 77 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.