Recent analyses of the Planck data and quasars at high redshifts have
suggested possible deviations from the flat $\Lambda$ cold dark matter model
($\Lambda$CDM), where $\Lambda$ is the cosmological constant. Here, we use
machine learning methods to investigate any possible deviations from
$\Lambda$CDM at both low and high redshifts by using the latest cosmological
data. Specifically, we apply the genetic algorithms to explore the nature of
dark energy (DE) in a model independent fashion by reconstructing its equation
of state $w(z)$, the growth index of matter density perturbations $\gamma(z)$,
the linear DE anisotropic stress $\eta_{DE}(z)$ and the adiabatic sound speed
$c_{s,DE}^2(z)$ of DE perturbations. We find a $\sim2\sigma$ deviation of
$w(z)$ from -1 at high redshifts, the adiabatic sound speed is negative at the
$\sim2\sigma$ level and a $\sim3\sigma$ deviation of the anisotropic stress
from unity at low redshifts and $\sim3.5 \sigma$ at high redshifts. These
results suggest either the presence of a strong non-adiabatic component in the
DE sound speed or the presence of DE anisotropic stress, thus hinting at
possible deviations from the $\Lambda$CDM model.