Neuroscientists
study the brain at multiple scales, ranging from the detailed structure of ion
channels (those proteins in the neuronal membrane that let ions in and out of
the cell) all the way to inferring brain function by examining behavior. There
is also a similarly wide range of temporal domains from the sub-millisecond
processes that govern the movement of ions up to the consequences of aging over
the years.
We
all hope that one day we will be able to navigate smoothly across all of these
scales, in the same way that we can transition from subatomic particles all the
way to galaxies (or can we?). A term that often arises when discussing
different scales is the concept of emergence.
Sometimes, it is useful to consider macroscopic properties of a system where “…
the whole is more than the sum of its parts…” It is useful to use temperature
to describe the property of an object even when there is no direct temperature
in each of its atoms. There is a whole arena of Physics where we relate
macroscopic thermodynamic properties such as temperature to the statistical
properties of the microscopic components and their interactions.
We
may expect the evolution of a similar niche in Neuroscience where we can relate
the macro and micro worlds. The notion of emergence in brain circuits is not
new. Ultimately, every aspect of brain function emerges from the interactions
of a large number of interconnected neurons. Yet, the problem of how to
rigorously define and study emergence in neural circuits has been quite
elusive. Recently, Giulio Tononi’s group made a significant leap forward in an
elegant manuscript that was published before the close of 2013.
The
authors started by quantifying how well one could predict future states from
previous states, a measure they refer to as effect information. In a nutshell,
the effectiveness by which causal interactions in a system can be described
depends on the degree of determinism and degeneracy. Determinism is a measure
of how much noise there is. If the system is so noisy that the future is
completely random and unrelated to the present, then predictability will be
low. Degeneracy refers to the number of ways in which a future state can be
reached from the present.
Skipping
a few technical details, for any spatial or temporal level of analysis, it is
possible to go through all possible states and evaluate how well future states
can be predicted from the present state. In principle, it is possible to
perform this analysis at multiple different coarse graining scales. We can ask
about how well we can predict future states based on the flow of ions, or upon
considering action potentials, or studying local field potentials, or even at
the behavioral level. It turns out that it is not always the case that the most
reductionist version of the system is necessarily the most informative one in
terms of future states. This observation holds even though the system is
perfectly “supervenient”, that is, the macroscopic states are perfectly
dictated by the microstates and the mechanisms that group microelements into
macro ones. In other words, there is no new magic in the macroscopic states.
The
manuscript provides a series of elegant and simple examples where causal
emergence arises due to an increase in determinism and/or a reduction in
degeneracy upon spatial and/or temporal coarse graining at the macro scale. In
this way, the authors provide a series of definitions to rigorously evaluate
the optimal level to characterize and predict the behavior of the system, the
level that “carves nature at its joints”.
Brining
these definitions to bear on real brains is far from trivial. One of the
difficulties resides in the exponential growth in the number of states and
possible interactions. Even for simple organisms with a relatively small number
of neurons, the number of possible microscopic and macroscopic states is so
large that we cannot really compute these quantities in any simple way. Yet,
not all hope is lost. The formalism defined in this study may help look for
simplifications and feasible comparisons among macroscopic states or network
motifs that can be characterized.
The
simple and well-thought theoretical foundation established in this study opens
the doors to a serious discussion and quantification of causality and emergence
in neural circuits. In turn, the quantification of causal emergence may find
implications for deciding upon what kind of measurements to make to study
neural systems, for deciding how to best make use of neural data for practical
applications such as prosthetic devices and, most importantly, for bridging
different levels of analyses that can transform our understanding of brain
circuits.
Hoel EP, Albantakis
L, Tononi G (2013) Quantifying causal emergence shows that macro can beat
micro. Proceedings of the National Academy of Sciences of the United States of
America 110:19790-19795.
