Life’s Information Hierarchy

The explanation for the complex, multi-scale structure of biological and social systems lies in their manipulation of space and time to reduce uncertainty about the future.

By: Jessica C. Flack

Date: April 2014

In: Santa Fe Institute Bulletin and From Matter to Life – Chapter 12

  • Central premises:
    • Biological systems (order in space and time) form by processing information about regularities in the environment.
    • Individual components of biological systems perceive environmental regularities.
    • Individual perceptions may converge and a collective consensus about the “average value” of a regularity over time may emerge.
    • The consensus value may better reflect the regularity than the (noisy) fluctuations perceived by any one component.
    • Individual components may use the collective consensus view (based on the  past) to estimate the appropriate course of action (in the present) or make predictions (about the future).
    • This type of “information processing” by biological systems and their components leads to:
      • More efficient survival (maximum energy extraction and minimum uncertainty).
      • Faster adaptation (fine tuning of individual strategies, learning).
      • Increased complexity (energy is freed up to explore new paths).

Key Takeaways

  • Priors, derived from experience of regularities over time, drive better prediction.
  • Better prediction drives more efficient survival, faster adaptation and increased complexity.
  • No regularities to “exploit” = no complex biological systems.
  • Longer time scale of regularities, slower feedback: = higher required level of complexity to understand and adapt behavior (elaborate model needed for encoding).

Worth Reading

  • The framework for collective cognition is helpful in exploring the formation of priors, hierarchy in the brain, learning, chunking, etc.
  • Helps to understand our bias for using averages as our “slow variables” for better prediction (or more broadly, formation of most perceptual biases).
  • Also points to difficulty of dealing with irregular, longer time scale, slow feedback regularities. It is more difficult for a consensus view to emerge when dealing with non-intuitive, non-linear, less predictable outcomes.
  • Along the same lines, macro-properties (for instance, political institutions) deteriorate when they are no longer (or have never been) supported by cohesive consensus views. Hierarchical order can only be maintained when the consensus provides a “good enough” estimate for a sufficient number of individuals / components.

Key Concepts:

Biological systems: networks that manipulate space and time.

  • Biological systems are networks.
    • Systems …
      • Composed of basic (heterogenous) elements, components.
    • … with inputs in the form of components that pursue strategies …
      • Rule-based interactions among components.
    • … and outputs in the form of collective behavior …
      • Functional properties that emerge at aggregate system levels.
    • … where feedback …
      • System (macro) properties feed back into component (micro) decision making and behavioral strategies.
    • … allows for adaptation.
      • Components (and the system) learn.
  • Biological systems process information to improve prediction:
    • More efficient survival:
      • Maximize energy extraction.
      • Minimize uncertainty (order, low variance).
    • Faster adaptation:
      • Learn: fine-tuning behavioral strategies.
    • Increased complexity:
      • Free-up energy

Course-graining (subjective interpretation of regularity in the environment) leads to better prediction

  • Biological systems exploit regularities in the environment …
    • Evolution is essentially the process of learning from and adapting to regularities in the environment.
    • Adaption requires regularities that can be estimated (and/or manipulated).
    • The goal of adaptation is fine-tuning strategies to maximize energy generation and minimize uncertainty.
  • … as the perceptions of its individual components converge …
    • Each component perceives noisy, short time-scale, fluctuating regularities.
    • Over time, individual perceptions of components may converge.
  • … a collective consensus may emerge …
    • About the average value of the perceived regularity (the “slow variable”).
  • … that improves prediction …
    • The slow variable may be a better input for prediction than any noisy individual perception.
    • The slow variable operates on a slower time-scale than the noisy underlying interactions.
  • … and leads to the formation of priors.
    • Slow variables become the subjective, inferential basis for the formation of priors: models of the environment.
    • Priors are hypotheses about the present and future environment that are induced from the past environmental states (collective perceived averages).
    • Priors are updated with observed regularities.
    • [In other words, “beliefs” are formed against which any new incoming data will be tested – see also “How to Change your Mind” and “REBUS“.]

Better prediction leads to survival, adaptation, complexity

  • More efficient survival.
    • Minimizing uncertainty through more accurate prediction.
    • Less energy wasted on inefficient individual perception and estimation.
  • Faster adaptation.
    • Fine-tune the strategies and behavior of components.
    • Lower costs for components to explore a broader range of strategies.
  • Increased complexity.
    • Freeing up of energy allows for more complexity.
    • Emergence of hierarchical organization.
      • As components start using the consensus estimate, hierarchical organization emerges.
      • Slow variables that were derived from regularities occurring at a lower level become encoded in distinct properties at a higher level.
    • This happens when:
      • Components rely more on these (macroscopic) slow variables thans on (microscopic) local fluctuations.
      • Components estimates are largely in agreement (convergence on “good enough” estimates of underlying correlated behavior).
      • The value of the slow variable has functional consequences (at the system or component level).
      • The value of the slow variable is driven by component interactions in pursuit of a strategy.
      • Is sufficiently stable over a biologically relevant period of time.
    • Nested organizational levels.
      • In terms of space and time scales.
    • Each level is associated with a new, emergent function.
      • Consequences, pay-offs typical of that particular level.
      • For the system as a whole or its components.

Finding the “slow variables” that drive biological systems to better prediction.

  • Need to understand the flow of continuous collective computation.
    • Collective estimation and compression of environmental regularities over time.
    • Inputs:
      • Multiple components interacting and implementing rules or strategies (microscopic behavior).
    • Algorithms:
      • Connecting the inputs and the outputs.
      • Mapping micro and macro.
      • Understanding the manner in which micro strategies combine to produce macro outputs.
    • Outputs:
      • Measured macroscopic behavior.
    • Flow:
      • Regularities in the environment -> summing of prior experiences -> predictions about strategy that improves fit -> behaviors.
  • Which slow variables are biologically fundamental?
    • Derived from the microscopic level:
      • From data on interactions among components that are important to the system.
    • Feed back into the microscopic level:
      • The resulting slow variable needs to be read by and influence the behavior of components individually or collectively.
  • Using strategically statistical mechanics to calculate emergent properties.
    • Discovering law-like behavior at the aggregate level.
    • Providing the microscopic basis for the macroscopic variables.
    • Exploring chains of probabilistic events that generate and respond efficiently to average features of the world.
      • Similar to physics: thermodynamics.
  • Understand how underlying regularities are subjectively processed by the system.
    • Start at the bottom and work upward from the data: simulations.

Example: distribution of social power in animal groups.

  • Slow variable: distribution of power.
    • Power = degree of consensus in the group that an individual can win fights.
    • Provides order, hierarchy.
    • Institutionalized: hard to change, slow time scale (many opinions need to change).
  • Better prediction.
    • Power structure provides information about the future cost of interactions, the conflicts that can or can’t be afforded.
  • Derived from micro-level interactions.
    • Bridging differing time scales.
    • Micro level: distribution of fighting ability (shorter time scale, individuals, changes faster).
    • Macro level: distribution of social power (longer time scale, collective, changes more slowly).
  • Computation of slow variable (compression).
    • Summing up of the outcomes of many fights and conflicts over time encodes a slowly changing power structure.
    • Collective assessments of conflicts converge on a consensus about who has power.
  • Slow variable leads to better prediction.
    • Power structure is a better predictor than the outcome of individual interactions.
    • Individual interactions can randomly / contextually fluctuate.
  • Leading to increased complexity, hierarchy.
    • Fine tuning of strategies:
      • “Policing” becomes an affordable strategy.
      • Before the emergence of order, it was too costly a strategy.
      • With an established order, those high in status can “police” without being challenged.


Leave a Reply