The 'wiki' section of the site is the progress of being migrated from an old site. There are some rendering issues and many of the links don't work yet.

Focusing on the why for Animal Kingdom

This page forms part of the [[Focusing on the Why]] series. It presents a comparison to the Animal Kingdom.

Introduction

Note: In deciding which stage a particular animal / insect fits, I’m trying to answer a very specific question:

  • What’s the simplest architecture that could emulate the behaviours? Thus, a fly, for example, may employ attention control, but it may not. So I will assume for now that it does not, until I find a strong reason to believe that it needs attention control and for what reason.

Background

To discuss:

  • sensor-adjustor-effector biologies
    • See https://journals.biologists.com/jeb/article/214/8/1215/10743/Do-jellyfish-have-central-nervous-systems, and https://www.sciencedaily.com/releases/1999/04/990415064510.htm, in note: “Biol. Summ. - Animals and Evolution”

A Progression through the Biological Kingdom

Plants

These use hormonal (ie: chemical) signalling. While there is a level of macro-scale feedback, it is via a sense-act-outcome cycle, rather than a pre-action cognitive feedback. Thus, it is of the form of a feedforward-only network. In other words, our current “state of the art” deep neural networks, are just scaled-out versions of plant-cognition.

Plankton and single-celled organisms

The simplest animals (eg: bacteria and simple plankton) have no neural connections. They employ evolutionarily-trained pre-configured simple chemical and electrical signalling. The sensor-adjustor-effector triad are typically collapsed into cilia (micro appendages on the surface of cells) that merely react to their immediate environment, without any global coordination.

Without any means for accurate long-range signalling between parts, such organisms typically repeat the same sensory capability across its surface, in order to control the span of cilia. This is inefficient. Some zooplankton species have been found to employ simple neural connections that would appear to improve the cellular efficiency - ie: less unnecessary duplication (Jékely, 2011; Conzelmann et al, 2011). These neural connections a likely merely pass-through, providing no computational benefit.

Jellyfish

Background:

  • See note “Biology Summary - Animals” Classification:
  • Box jellyfish hunt prey, which requires more complex behaviour. They also have multiple distinct overlapping networks, which could be examples of centralised control and/or predictive coding feedback loops.
  • All other jellyfish seem to be pretty much very simple feedforward reactive machines.

Flies

Background:

  • See note “Biology Summary - Animals”

Flies nervous system could be a very simple combination of the following main features:

  • Primarily genetically hard-coded, with minimal post-birth learning.
  • Examples of genetically hard-coded behaviours:
    • Olfactory sense triggering flight response towards smell for specific smells. The smell that draw flies are a fairly straightforward chemical signal that are consistent across time.
    • Vision of fast moving object comings towards fly triggering flight away (threat response).
  • Main learning occurs via predictive-coding processes.
  • Predictive-coding + afference copy to identify the difference between self-caused visual change versus external environmental changes.

The above can even cope with simple associative learning, eg: learning that one side of a chamber is too hot, and thus avoiding it. With reference to this study;

  • Williams-Simon, P.A., Posey, C., Mitchell, S., Ng’oma, E., Mrkvicka, J.A., Zars, T., King, E.G. Multiple genetic loci affect place learning and memory performance in Drosophila melanogaster (2019) Genes, Brain and Behavior, 18 (7), art. no. e12581. https://doi.org/10.1111/gbb.12581

The next “why” question receives some answers from the following question:

  • If a flies behaviour was entirely genetically hard-coded, with no learning even from birth, what would it be missing?
  • What learning is required to support adaptation?
  • What learning is required because it would be more efficient / accurate to learn post-birth rather than being genetically hard-coded?

However, I think I saw an article that found that flies have attention too.

Plus, while the behaviours of a fly may seem simple, the environment that they operate in is almost as complex as it is for humans. So in practice they may need many of the same systems.

References

Jékely, G. (2011). Origin and early evolution of neural circuits for the control of ciliary locomotion. Proceedings. Biological sciences, 278(1707), 914–922. https://doi.org/10.1098/rspb.2010.2027. [Full Text]

Conzelmann, M., Offenburger, S., Asadulina, A., Keller, T., Münch, T. A., and Jékely, G. (2011). Neuropeptides regulate swimming depth of Platynereis larvae. PNAS, doi: 10.1073/pnas.1109085108. [Website]