Ergodicity and human decision making on the impact of path dependencies on decision-makers

Expected values are at the foundation of scientific models across diverse fields dealing with uncertainty. Unfortu-
nately, the reliability of the expected value approach relies heavily on the ergodic hypothesis, which is the assertion
that a system’s ensemble average equals its average over time, i.e. time average. In our research, we question the
validity of the ergodic hypothesis and propose a paradigm shift, arguing that the implicit assumption of ergodicity
is at the root of many discrepancies between theoretical models and the real world. We explore the mechanics of
ergodicity breaking, examining its manifestations in self-reinforcing systems, the presence of absorbing barriers,
and its overlooked role in science. In particular, we will focus on economics, decision-making, and reinforcement
learning, which all rely heavily on expected values in their computations and predictions.
In this thesis, we aim to bridge the gap between theoretical understanding and practical applications of ergodic-
ity by adapting theoretical foundations to better reflect discrete processes and real-world dynamics. Additionally,
we will use theoretical models and (thought) experiments to test and apply this new framework to the decision-
making challenges humans and machines encounter.
Finally, we demonstrate through stated preference experiments that time averages serve as a more suitable
null model for modelling human decision-makers than ensemble averages. Additionally, we reveal that many
robust findings in behavioural economics are not primarily rooted in psychology. Instead, the behaviour exhibited
by human decision-makers can be understood as a logical consequence and potential adaptation to living in a
non-ergodic world, where optimizing the time average leads to robust strategies that are largely independent of
personal biases and heuristics.