AI-designed balance exercises improved walking measures, but the human instructor still mattered

The most interesting finding here is not that a chatbot “trained” older adults. It is that a human instructor delivered an exercise plan generated with an AI model, and the small trial compared it against two other structured approaches.

Technology-assisted groups showed some larger gait gains. But sixteen people per arm cannot settle safety, fall prevention, durability or how much of the benefit comes from program design, instructor enthusiasm or the amount of practice. Every group improved during repeated exercise sessions, which is itself worth noticing.

The participants trained under supervision. Improvements on walking measures do not establish that an autonomous exercise program would be equally effective or safe.

The evidence: 48 retirement-home residents aged 60–75 randomized equally to human-delivered AI-generated exercise, virtual-reality exergaming or traditional balance training; 60-minute sessions three times weekly for eight weeks, with two-month follow-up. All groups improved balance and gait measures. AIand VR-supported programs improved gait speed more than traditional exercise (both P<0.001 in reported comparisons), without a reported AI–VR difference. Virtual reality had the largest cadence gain and AI the largest gait-stability ratio reduction; no adverse events were recorded.

plans prevent falls or extend life.