[Exclusive Deep Dive] SuperHero Training Academy: Crafting the Ultimate Centaurian Curriculum

In superhero narratives, the journey from ordinary to extraordinary is often marked by a transformation normally depicted by a training montage. Last week, we explored how to train our brains to adapt to superpowers, extra limbs, and enhanced senses, drawing from the work of David Eagleman and his concept of Livewired -- the plug-and-play aspect of our brains and how they can create new capabilities. We showed through the work of Steve Swink and his framework to create and understand 'Game Feel', which immerses you in an experience to learn complex skills. This goes beyond mere gamification; it's about immersing oneself in the experience, feeling every aspect of the task. 

"The tactile sensation of manipulating a digital agent. The thing that makes your mom lean in her chair as she plays Rad Racer. Proxied embodiment. Game feel."

Steve Swink: https://www.gamedeveloper.com/design/game-feel-the-secret-ingredient

But how do we elevate this concept into an ultimate training curriculum for our aspiring superheroes?

To answer this, we turn to the insights of Max Bennett and his seminal work on the evolution of human intelligence. Bennett identifies five major breakthroughs in our cognitive development, providing a blueprint for understanding learning and intelligence. 

"Let's start with Artificial Rat–level intelligence (ARI), then move on to Artificial Cat–level intelligence (ACI), and so on to Artificial Human–level Intelligence (AHI)." —YANN LECUN, HEAD OF AI AT META

"...the entirety of the human brain's evolution can be reasonably summarized as the culmination of only five breakthroughs, starting from the very first brains and going all the way to human brains.

Bennett, Max. A Brief History of Intelligence (p. 10). HarperCollins. Kindle Edition. 

While his analysis stops short of exploring non-human intellects (octopus)—a topic for future discussion—it lays the groundwork for our superhero training academy's curriculum.

Breakthrough #1: Steering and Associative Learning

"These bilateral body plans simplified navigational decisions into binary turning choices; nerve nets consolidated into the first brain to enable opposing valence signals to be integrated into a single steering decision. Neuromodulators like dopamine and serotonin enabled persistent states to more efficiently relocate and locally search specific areas. Associative learning enabled these ancient worms to tweak the relative valence of various stimuli. In this very first brain came the early affective template of animals: pleasure, pain, satiation, and stress..."

Bennett, Max. A Brief History of Intelligence (pp. 359-362). HarperCollins. Kindle Edition. 

The inception of intelligence began with steering—a fundamental ability to navigate. This basic navigation system, driven by simple valence states, enabled early creatures to move towards desirable outcomes and away from adverse ones. It marked the beginning of affective emotions and associative learning. The ability to navigate the physical world is the start of intelligent behavior and learning.

Training Principle 1: Learning to navigate with basic guidance signals -- pain, pleasure, satiation, and stress.

Breakthrough #2: Reinforcement Learning

"Five hundred million years ago, one lineage of ancient bilaterians grew a backbone, eyes, gills, and a heart, becoming the first vertebrates, animals most similar to modern fish. And their brains formed into the basic template of all modern vertebrates: the cortex to recognize patterns and build spatial maps and the basal ganglia to learn by trial and error. And both were built on top of the more ancient vestiges of valence machinery housed in the hypothalamus. This model-free reinforcement learning came with a suite of familiar intellectual and affective features: omission learning, time perception, curiosity, fear, excitement, disappointment, and relief. "

Bennett, Max. A Brief History of Intelligence (pp. 359-362). HarperCollins. Kindle Edition. 

Next, we delve into reinforcement learning, the process of repeating behaviors that yield pleasure and avoiding those that cause pain. This model-free self-reinforcement underpins much of AI research today, offering a framework for training AIs to accomplish tasks. 

The Credit Assignment Problem:

What is the credit assignment problem? Recognizing which actions lead to success necessitates a more sophisticated approach. In complex games like Chess and Go, deciding which moves finally lead to winning is hard to determine. Bennett used a simple example to explain it:

"Imagine you are playing checkers. For the first nine moves, it is mostly neck and neck between you and your opponent. And then on the tenth move you pull off some clever maneuver that turns the tide of the game; suddenly you realize you are in a far better position than your opponent."

Bennett, Max. A Brief History of Intelligence (p. 107). HarperCollins. Kindle Edition. 

Figure 6.1 Original art by Rebecca Gelernter

Bennett, Max. A Brief History of Intelligence (p. 107). HarperCollins. Kindle Edition.