How computers are learning to predict human gravity adaptation

Living without an anchor

Every step and catch you make quietly relies on Earth pulling down at a steady rate. Remove that downward pull, and your inner senses lose their familiar anchor. Learning about gravity adaptation is vital for missions to the Moon or Mars. However, testing real astronauts under those exact conditions is rare, risky, and costly.

To explore this problem without leaving Earth, scientists built a framework that models our changing biology. Think of your balance system like a phone screen. It uses inner sensors to check which way is down to keep the screen upright. If you swap the physics rules entirely, the software miscalculates until it adjusts to the new environment.

Mapping nerves and numbers

The system uses two math tools alongside a large language model. The first part, named CorticalG, is a multilayer perceptron, which is a basic neural network built from connected layers of virtual nodes. It takes a single gravity value and predicts shifts across brain wave bands, such as alpha and beta waves.

To help the model learn complex, wavy patterns from simple numbers, the team used Fourier features. These are wave conversions that turn simple inputs into sine waves. The second tool, PhysioG, uses eleven Gaussian Process models to estimate heart, motor, and involuntary body changes. A Gaussian Process is a statistical method that makes predictions while giving clear mathematical uncertainty bounds for each estimate.

These numbers then feed into Claude 3.5 Sonnet. The language model acts as an in silico participant, which means a subject created entirely inside a computer simulation. It turns the physical predictions into written, first-person accounts of alertness and bodily awareness.

Where the body struggles

To train the system, the authors built an artificial dataset of 94,000 observations based on data from parabolic research flights. These planes fly in high arcs to briefly mimic weightlessness or partial gravity.

The findings showed that gravity levels below roughly 0.4 g are not strong enough to keep muscles, bones, and hearts healthy on their own. Both the Moon, at 0.16 g, and Mars, at 0.38 g, fall below this safety line. This means future space crews will need active exercise machines and medical tools to survive long stays.

What we still do not know

Because these findings come from computer runs rather than deep space telemetry, key questions remain. The training data came from brief flight arcs. High-gravity phases near 1.8 g were grouped together, rather than tracking continuous months in deep space.

In addition, the subjective sensations were generated by an AI assistant rather than spoken by living people facing actual space hazards. While a phone sensor adjusts instantly when rotated, human biology changes in messy, slow ways that simple code cannot fully capture. Scientists must still test these virtual outputs against real-time data collected from crews walking on the Moon or Mars.