Insights from the brain:
The road towards Machine Intelligence

Matthieu Thiboust - 2020


This ebook is a personal view and synthesis of some selected experimental findings and theoretical ideas from neuroscience research, that are currently – or could be soon – used in neuroscience-grounded artificial intelligence (AI) approaches.

Even if the inconceivable complexity of our brains makes it near-impossible to perfectly understand its inner workings, we can still get valuable insights from an incomplete and modest approach.

The humble goal of this ebook is to provide AI researchers with neuroscience chunks of information related to AI. Some chunks are solid ground truths with large scientific consensus, others are general principles on which there is no complete consensus, and the remaining ones are just informed speculations that fit with theorical and experimental results.

Download the ebook (PDF, 14 Mb)

Table of contents



  • Artificial Intelligence needs a new momentum. Why not look at the brain?

Brains and cognitive abilities

  • The primary function of a brain is not to think but to efficiently control complex behavior

  • This control is supported by abilities that were progressively acquired and refined through evolution

  • Biological intelligence gradually emerged with active perception and cognition

Brain general machinery

  • Neurons are sophisticated elementary components of the neural “hardware”

  • Neuron plasticity allows to retain memories of previous neural activity

  • Interconnected brain structures group neurons into organized network architectures

  • Brain activity continuously loops across those structures through parallel pathways

Focus on the neocortex

  • The neocortex is divided into hundreds of functionally specialized but anatomically similar cortical areas

  • Cortical areas receive and send information in a laminar-specific way

  • A majority of long-distance projecting pyramidal neurons cohabits with a minority of local inhibitory cells

  • Functional neocortical circuits rely on laminar-specific lateral and radial interactions

  • Sensory stimuli, motor actions and spatial navigation offer a window into the cortical code

  • The dynamics of cortical activity can only be analyzed in relation to brain oscillations

Back to code

  • Next-level artificial neural networks model more realistic neurons, architectures and learning rules

  • The transition from artificial networks to artificial agents is a necessary step towards machine intelligence

  • The potential emergence of machine intelligence already raises existential questions


  • The road towards machine intelligence is inseparable from a mixed AI & neuroscience approach