First dataset lets researchers measure how emotions show up differently in aging brains and bodies

Researchers created the first multimodal dataset designed specifically for older adults: 42 people, 38 hours of video, audio, brain waves, and heart rhythm data, plus annotations of emotional and cognitive states. This means scientists can now build AI models that actually recognize emotions in aging populations instead of assuming models built on young, healthy people will work the same way.

For decades, emotion recognition AI was trained almost entirely on young, cognitively healthy people. That's a structural problem because aging changes how emotions appear in the face, voice, and body — and cognitive decline changes them further. This dataset breaks that bottleneck. Now researchers can actually see whether the aging brain processes emotional signals differently, and whether early cognitive decline shows up in emotional expression before it shows up in memory tests. The first downstream use case is obvious: emotion recognition that works for people in nursing homes or assisted living, which is where you'd actually need it.