After reading this book I have to say it is currently the number 1 recommendation I give if somebody wants to try to understand how the human brain works. Starting from eukaryotes Max Bennett tried to reason about what it means for a living organism to be intelligent. Not only did he summarize the current literature he also synthesized the main ideas into 5 breakthroughs:

  • Steering
  • Reinforcing
  • Simulating
  • Mentalizing
  • Speaking

One important note about the whole story is, that it is about the human lineage in the evolutional history, and therefore about animals, vertebrates, mammals and primates.

Steering (neurons, bilateral, valence, brain, emotions, continuous learning)

Because animals differ from plants and fungi among many differences also in the important aspect that they have internal digestion, they evolved neurons for contracting their stomach muscels for very fast and specific movement. So the first very intriguing fact is that the unit of intelligence, the neuron is the same in all animals, from anemones till humans. Furthermore moving around and hunting prey is much easier when you have bilateral body layout (left-right symmetry with clear front direction) instead of being radially symmetric without clear front direction. This reduced the requirements of skills and choices from moving in all directions to be able to go forward or to turn. Also evolution invented valence, which means nothing more than good or bad response.

Reinforcing (td learning, credit assignment problem, causality, time)

Simulating (generative model of world, self-supervised)

Mentalizing (generative model of myself and other agents, imitation, supervised learning)

Speaking (labelling, unsupervised learning, )

Questions

  • Has anybody tried to build such a layered system with many fallback options. With this I mean for example a reinforcement learning system which tries to be model based, however if forced to take a fast deciscion falls back to a model free system.

  • Do we already have all parts built separately in AI or ML? If not, which are missing and how are they built in biological systems?

  • Which of the parts we already have re-built in AI are the most valuable from an AI point of view and which from a biological point of view?

  • What is intelligence after reading this book? What is consciousness?

  • Why did evolution evolve intelligence/consciousness? What evolutional benefit does it have?