This book is comprised of three different sections, more details on each section below:

1. How the brain works
   1. The brain is composed of over 10,000 neurons. Neurons are specialized cells in the brain that consist of synapses and dendrites, which enable them to sense and transmit signals.
   2. The neocortex is the conscious part of the brain responsible for thinking and observation, while the “old brain” controls basic life functions like eating and breathing.
      • The old brain might urge you to eat a piece of cake, while the neocortex advises against it due to health concerns.
   3. The neocortex is a folded organ located in the skull, giving it a wrinkled appearance. Think of it like a napkin crumpled up and placed inside a wine glass.
   4. The brain works by observing the world through the five senses and creating models of the world based on these observations. It then uses these models to predict what is likely to happen next. This process applies to vision, hearing, smell, taste, and touch.
      • For example, your brain forms a model of the room you’re in, knowing where chairs and windows should be. If something in the room changes, your brain detects the inconsistency with the model.
      • The neocortex also builds associations between these models, creating visual maps that help us navigate. One room is connected to another, and one building on a street is next to another, enabling movement.
      • Dendrites make predictions about what you should observe next, while neurons observe the sensory input from your body.
      • The brain stores information using these models, often linking abstract concepts to concrete images or sounds. The memory palace technique is an example, where you associate a list of things with specific locations in your house or childhood street.
   5. The “thousand brains theory” suggests that each neuron in the neocortex functions as an independent brain. To make sense of observations, there is a voting mechanism within the brain that reaches a consensus on the interpretation.
      • For instance, when you see a car, your brain has already constructed a model of what a car is and how it works. This is why when presented with a 2D drawing of stairs, your brain maps it to 3D since it recognizes the model. Another example is the vase or two faces drawing, where your brain cannot perceive both simultaneously.

1. Machine Intelligence:
   1. On generative AI (AGI) and  intelligence. The Author defines intelligence as the ability for something to model the world which it is operating and make a predictive model. Today’s AI like chat GPT cannot do this, it can process information and compare it to other information. Below are a few examples:
      1. Deep learning models must be trained before they can be deployed. This is how humans learn any skill like playing golf or driving a car.
      2. As of right now AIs have to be retrained to recognize a new object. An AI cannot be taught the model of a flower and then use that context to learn a new model of a vegetable. Just because an AI can beat any human at a game of chess does not mean it can also drive a car.
      3. This idea that machines can learn on a model based concept in the way that our brains do would be what is commonly referred to as generative AI.
   2. Consciousness
      1. Consciousness is the awareness of internal and external existence. Once machines can start building models of the world and create moment to moment memories of those models.
      2. An example of when this process does not work well is when we enter a room and wonder why we came in there in the first place. When our brains work well the neurons form a continuous memory and recall our previous thoughts.
      3. Qualia is the name for how sensory inputs are perceived and how they feel. They are subjective which is why some people like a given food and others don’t.
         1. Another example of this was the famous image of the dress that was either perceived to be white and gold or blue and black.
      4. Lex Friedman gave the example of a computer receiving consciousness from the movie ex machina. At the end of the film the AI robot smiles to herself which shows she wasn’t doing it for anybody else.
   3. Future of machine intelligence
      1. The main way that we learn is by moving around. We grab a tool and touch it, we watch a sporting event, we listen to music or a podcast, etc. In order for robots to have this type of learning model they will need to be able to move around and observe too.
      2. We would need to program machine intellegency with an equivilent of an old brain. Instead of breathe, eat, sleep it would be something like don’t hurt humans, you must listen to human commands unless it breaks rule #1, etc.
      3. Machines can process and action tasks much faster than we can. This is where you get the idea of AGI developing 100s of thousands of years in human progress overnight. The issue with this is learning typically requires interaction with the physical world which will require more time. For example to build. rocket you need to fasten equipment together, this can be done faster than humans, but not instantly. To cure cancer we will need to observe cells reacting to treatment, this does not happen instantly.
      4. An advantage machines have over humans is that they can upload new information versus with humans we have to learn by observation.