A review of Max Bennetts book: A Brief History of Intelligence
This story delves into the origins of the brain, providing a comprehensive understanding of its functioning, purpose, and interactions with modern AI systems. By synthesizing ideas from neuroscience and AI, it offers a unified perspective on these two fields.
Introduction
While overly optimistic predictions about AI may have accurately predicted much of todays existing technology, they could not have anticipated that the yardstick used to measure how far AI has progressed from achieving human-level intelligence has not changed since its inception. This test remains: Can an AI system load a dishwasher?
Large language models (LLMs) represent the latest surge in AI development. LLMs are trained on vast data sets like the entire Internet, and essentially perform mathematical pattern matching to generate the most probable response to a given prompt.
As a result, their development has been a rollercoaster of both brilliance and foolishness: todays AI can defeat a Grand Master at chess but struggle to answer common sense questions and, of course, cannot load a dishwasher. So, what should we do?
Natures hint.
Since the human brain excels at common sense and finds loading a dishwasher easy, perhaps nature offers better clues about how intelligence works than simply matching patterns in vast data sets.
At least, that is what this author thought until he ventured headfirst into the complexities of the brain.
With 80 billion neurons and 100 trillion connections, he realized that using nature as a guide was equally daunting. Even if it was possible for him to map the wiring of each neuron to all its connections, he would still be far from comprehending the complexity of brain functions.
The reason this is so lies in the fundamental difference between the wiring of a computer and the neuronal connections in the brain. Unlike a computer, where the wiring transmits a single uniform electrical signal across its entire network, neuronal connections employ diverse electrochemical signals from one neuron to the next to transmit its messages. This unique nature of signal transmission makes reverse engineering the brain virtually impossible.
Perhaps he thought, a more promising starting point would be the fossil record and the brains of both extant and extinct animals.
The Missing Museum of brains
The remarkable similarity between the brain structures of fish and humans provides valuable insights into the nature of our intelligence.
It suggests that brains were initially simple but gradually evolved into great complexity. This complexity emerged gradually over time through small random variations that were either selected for or eliminated by evolution.
In fact, the brain in our heads evolved 600 million years ago from a worm the size of a grain of rice. This worms brain contained the first neurons and emerged as a result of the pressures of the Cambrian explosion.
Therefore, every brain, in some way, is a time capsule containing hidden clues about the intelligence of the minds that came before it. By studying the intellectual achievements of animals with such brains, we can, not only reconstruct the brains of our ancestors but also determine the intellectual capabilities possessed by them. By gradation , we can then trace the evolution of each mental power over time. However, the author faced a another challenge: he had to overcome the prevailing myth of layers.
The Myth of Layers
In 1960, Paul McLean proposed that the human brain consists of three distinct layers, each built upon the previous one: The neocortex, often referred to as the center of cognition, language, abstraction, planning, and perception, according to McLean, evolved most recently. He also hypothesized that the neocortex was constructed atop the limbic system, the center of emotions, fears, parental attachment, sexual desires, and hunger. Building on this, McLean assumed that the limbic system evolved on top of the reptile brain, which served as the primary center of our basic survival instincts, such as aggression and territoriality.
However, McLeans theory proved to be wrong and led to erroneous conclusions about the evolution and functioning of the brain. Notably, we do not possess a reptile brain, and evolution does not simply overlay one system on top of another without first modifying the existing system.
Drawing from McLeans longstanding error, this author concluded that understanding the workings of the brain necessitates a solid foundation in our understanding of intelligence, including its applications in artificial intelligence.
The Five Breakthroughs
The author got back on the right track by discovering five breakthroughs that summarize the real story of how intelligence works. It is a theory that served as an organizing map for a journey back though cognitive time.
Each of the authors breakthroughs equipped animals in the evolutionary record with a new portfolio of intellectual abilities, resulting from structural adaptations dictated mostly by extreme environmental pressures.
The world before brains: A theory of how cells scaled the wall of entropy into life
Since intelligence predates brains, we cannot understand why and how brains evolved without first examining the evolution of intelligence itself. Although we still do not know the exact process, we know that even the earliest animals must have possessed some rudimentary form of intelligence to overcome the thermodynamic barrier of entropy, giving rise to life.
The authors theory proposes the existence of self-replicating strains of DNA-like nucleotides that emerged in the thermal hypervents of the ocean floor. These nucleotides produced their own set of ribosome molecules, which began translating specific sequences of amino acids into proteins.
These proteins then initiated their own synthesis, taking on various shapes and performing diverse functions within the cells that contained them.
One of these functions exhibited characteristics looking suspiciously like intelligence: they could move and respond to stimuli, and they possessed receptors that monitored and responded to their environment.
As these ancient cells moved, reacted, and monitored their surroundings, they indeed possessed a primitive form of intelligence, implemented not by neurons but by a complex network of proteins in cascading chemical reactions.
Therefore, with the development of protein synthesis, even though we still don't fully understand how it began, the seeds of intelligence were sown.
This development also led to the transformation of DNA from mere matter into to a medium for storing information. Instead of being merely self-replicating, DNA was transformed into the informational foundation upon which life itself is constructed. DNA officially became life's blueprint, ribosomes its factory, and proteins its products.
With these foundational elements in place, the process of evolution could continue unabated. Variations in DNA led to variations in proteins, which in turn led to the evolutionary exploration of new cellular machinery. This new cellular machinery was then selected for based on whether it further supported survival.
At this point in the story of life, we have reached the threshold between life and non-life, between abiogenesis and genesis. All that remains now is to categorize the overall intellectual abilities cells needed to employ from this point onwards to avoid slipping back over the wall of entropy and returning to non-life.
The authors five evolutionary mental breakthroughs serve as the central theme of this narrative and the centerpiece of this book.
Intellectual Breakthrough #1: Steering
Noting that the neuron, dopamine, and serotonin emerged around the same time, it logically follows that during the newly competitive Cambrian Explosion, valence-guided reactive attention could have alerted small bilateral worms to when to go left or right, depending on which direction enhanced or diminished their survival chances. This valence-guided steering ability constituted the first emotion-based system for decision-making. With it came associative learning, as steering decisions soon became influenced by previous experience. Valence-guided steering solidified our ancestors place as the first large multicellular animal who survived by navigating with muscles and neurons rather than with hair-like cellular propellers.
Intellectual Breakthrough # 2: Reinforcing
You try it, and if it works, you try it again. This is the essence of reinforcement learning, or learning by trial and error.
This method could not have existed without first knowing how to steer right or left. Trial and error learning thus bootstrapped its way off the backs of neurotransmitter-assisted valence signaling and associative learning. Once trial and error was established, it ushered in a cascading suite of related intellectual developments, including morphological changes that resulted in the development of the basal ganglia as the brains structure for critiquing its own decisions.
Curiosity emerged as well, as a way of solving the exploration-exploitation dilemma, as did the cortex as an auto-associative network making pattern recognition possible.
The perception of timing followed, enabling animals to learn not only what to do but also when to do it. This led to the development of the hippocampus and other structures that allowed the creation of the perception of three-dimensional space, enabling animals to recognize their location and remember the location of things relative to themselves and to other things.
In short, steering made it possible for vertebrates to learn through trial and error, and trial and error in vertebrates made the perception of three-dimensional space possible. This led early mammals to the third breakthrough: learning not just by doing but also by imagining.
Breakthrough # 3: Simulating
As mentioned in breakthrough #2, the neocortex, a brain structure that emerged in early mammals, marked the third breakthrough in the evolutionary story of intelligence-the gift of simulation. This ability was nothing short of revolutionary.
With the sensory capabilities of the neocortex, mammals could not only create their own world model but also a self-model, a simulation that provided a status report on their external movements and internal states. This allowed them to construct intent, explaining their behavior to themselves, and the ability to pause and imagine aspects of the world that were not currently being experienced.
In essence, they could engage in vicarious model-based counterfactual trial and error learning, simulating possible future actions and deciding which path to take solely based on imagined outcomes.
Furthermore, these simulations enabled episodic memory, allowing them to recall past events and actions. They could then use these recollections to adjust their future behavior. Shortly after this, the motor cortex evolved, enabling them to plan and simulate specific body movements.
While most vertebrates at the time could perform most of these tasks, only higher primates could plan. It set us apart from the rest of the primate family, which lacked the ability to plan.
In essence, the true story of our intelligence lies in how our neocortex weaponized our imagination by creating an inner simulation of the world that allowed us to plan, significantly enhancing our chances of survival.
As we will explore, it was from this hard-won superpower that the next breakthrough, mentalizing, would eventually emerge.
Breakthrough # 4: Mentalizing
Early primates exhibited three broad abilities: theory of mind (inferring intent and knowledge of others), imitation of learning (acquiring novel skills through observation), and anticipating future needs (taking action now to satisfy a future desire even if it may not be immediately needed).
These abilities may not have been distinct entities but rather emergent properties of a single breakthrough called mentalizing the ability to construct a generative model of ones own mind.
This notion is partially supported by several factors: shared neural structures that evolved first in primates, the simultaneous development of these abilities in children, and the tendency for damage to one ability affecting the others. However, the most significant evidence lies in the fact that the structures from which these skills emerged are the same areas responsible for our ability to reason about our own minds.
These new areas appear to be unique to primates and are essential for simulating the minds of others, projecting ourselves into our imagined futures, identifying ourselves in the mirror, and recognizing our own movements. We observe this pattern in children, where the ability to reason about their own minds often follows the development of all three abilities.
However, the most compelling evidence for this idea comes from neuroscientists like Vernon B. Mountcastle. Mountcastle proposed that every area of the neocortex is composed of identical microcircuits. Ramon Santiago de Cabajals Nobel Prize-winning work confirmed this, imposing strict constraints on how we can explain the emergence of these new abilities in primates. It means that new intellectual skills must have originated from a novel application of the neocortex rather than from a computational trick.
This interpretation suggests that the theory of mind, imitation, learning, and anticipating future needs are merely repurposed uses of the neocortex. All three abilities: theory of mind, imitation learning, and anticipating future needs fit seamlessly into the unique niche of early primates.
Now, lets delve into breakthrough #5, the final divergence between humans and our closest living relative, the chimpanzee, which occurred seven million years ago.
Breakthrough number five: Speaking
Early humans found themselves entangled in an unexpected confluence of factors. The dying forest of the African savannah compelled them to adopt a niche as tool-makers and meat-eaters, necessitating the precise fabrication and transmission of tools across generations. These factors proved so effective that they catalyzed the development of pro-language skills.
The language abilities they fostered were not a result of a novel neurological structure, but rather adjustments to ancient structures that established a conducive learning environment for language acquisition.
The learning environment commenced with proto-language conversations within families, requiring joint attention among participants and facilitating the use of names associated with their inner simulations. Through this curriculum, all regions of the neocortex were soon repurposed for language processing.
Initially, the training program targeted children within families, but it soon expanded to unrelated individuals outside families. This expansion ignited a feedback loop of gossip, altruism, and punishment, which continuously favored more sophisticated language skills.
As social groups expanded and ideas traversed the brain, the human hive mind emerged, serving as a transient medium for idea propagation and accumulation across generations. As this language bubble expanded, larger brains were required to store and share the accumulated knowledge.
Harnessing the use of fire led to cooking, which not only provided new opportunities for language expression but also established the foundation for a substantial caloric surplus that was utilized to triple the size of brains.
Thus, from this perfect storm emerged the fifth and final pivotal breakthrough in the evolutionary trajectory of the human brain: language. Alongside language, humans developed numerous distinctive traits, including altruism and cruelty.
What truly sets humans apart is that the mind has transcended individual use, becoming interconnected with others as culture through a rich history of accumulated ideas. Five stars
Post Article Comment and Rate This Article
