A Brief History of Intelligence: Five Breakthroughs in Intelligence

Every breakthrough originates from a new reorganization of brain architecture,

and endows animals with a new set of intelligent attributes.

Below are the core facts about how different levels of intelligence evolve:

First breakthrough: Turning (seeking profit and avoiding harm)

Biological level: The first bilateral animals about 550 million years ago (such as ancestors similar to nematodes).

Anatomical basis: The emergence of the first brain (a few hundred neurons) and bilateral body structure.

Intelligent features:

Valence: Dividing the world into “good” and “bad,”

Generating primitive decisions to approach or avoid.

Basic navigation: Capable of turning based on chemical signal concentrations,

avoiding dangers (high temperature,

bright light) and swimming toward food.

Primitive emotions: Early functions of dopamine and serotonin evolved,

representing signals of “good things coming” and “good things happening.”

Second breakthrough: Reinforcement learning

Biological level: The first vertebrates about 500 million years ago (such as primitive fish similar to lampreys).

Anatomical basis: Formation of the six basic structures of the modern brain (basal ganglia,

thalamus,

hypothalamus,

midbrain,

hindbrain,

primitive cortex).

Intelligent features:

Trial-and-error learning: Evolution of basal ganglia,

using dopamine as a “temporal difference learning” signal,

enabling animals to learn complex action sequences through trial,

not just reflexes.

Curiosity: Surprising events become reinforcement signals,

driving animals to explore new environments.

Pattern recognition: The primitive cortex appears,

allowing the brain to recognize specific odors or visual patterns,

no longer relying solely on single stimuli.

Third breakthrough: Simulation and planning

Biological level: Early mammals (such as small mammals living in burrows during the age of dinosaurs).

Anatomical basis: The emergence of the neocortex.

Intelligent features:

Internal simulation: The neocortex grants animals the ability to perform “internal simulations” before action,

pre-visualizing different possibilities in the mind.

Slow thinking: Transition from purely reactive behavior to planning,

weighing different imagined outcomes and making choices,

corresponding to the “System 2” thinking in psychology.

Fourth breakthrough: Theory of mind (mentalization)

Biological level: Primates.

Intelligent features:

Social simulation: Capable of simulating others’ intentions,

thoughts, and emotions (i.e., theory of mind),

enabling complex social interactions and “political manipulation.”

Imitative learning: Able to understand others’ actions’ intentions,

and quickly acquire skills through imitation,

without having to trial and error each time.

Fifth breakthrough: Language and abstract thinking

Biological level: Humans.

Intelligent features:

Symbol system: Language is not just a communication tool,

but a social symbol system,

allowing humans to directly transmit complex imaginations or models.

Cultural inheritance: Language enables knowledge to transcend individual experience,

accelerating the evolution of thought beyond biological evolution’s slow pace,

giving rise to human civilization.

$Google(GOOGL)$

Criticism and potential limitations

Although “A Brief History of Intelligence” has been praised by scholars like Daniel Kahneman for its grand perspective and interdisciplinary integration,

from the perspectives of evolutionary biology,

neuroscience, and artificial intelligence (AI),

there are also some criticisms and potential limitations:

1. Simplification of the “Five Breakthroughs” framework

Critics argue that,

Bennett’s simplification of the complex 600-million-year evolutionary history into “five leap-like breakthroughs” makes it easier to spread,

but carries a strong teleological tone.

Evolution is not a ladder: The evolutionary biology community generally believes that,

evolution is not a ladder climbing toward “human intelligence,”

but a tree spreading in all directions.

Over-simplification: The book tightly links complex brain region functions with specific evolutionary stages (e.g., equating neocortex entirely with simulation and planning),

which may be overly simplistic from a neuroscience perspective.

In reality,

there is a highly complex synergy between old brain regions (like basal ganglia) and new brain regions (neocortex),

rather than simple “module addition.”

2. Controversy over the “AI mimicking evolution” path

Bennett hints in the book that,

to achieve true Artificial General Intelligence (AGI),

AI development should reenact the five stages of biological evolution (embodiment,

reinforcement learning,

simulation, etc.).

Computational efficiency issues: Critics point out that,

AI does not necessarily need to mimic biological pathways to surpass humans.

Just as airplanes can fly higher than birds without flapping wings,

large-scale data-driven Transformer architectures (like GPT-4) have already demonstrated strong reasoning and language abilities without undergoing “five evolutionary breakthroughs.”

Overlooking non-biological advantages: Some argue that,

overemphasizing “biological origins” might overlook the unique advantages of silicon-based life (such as super-high computing power,

infinite memory, and instant knowledge sharing),

limiting the imagination of AI’s potential.

3. Limitations of “language” as the final line of defense

The book views language and abstract thinking as the pinnacle and last breakthrough of human intelligence.

Impact of LLMs: With the rise of large language models (LLMs),

people find that language ability can seem to “emerge” from massive text statistics,

without the need for complex social simulation and theory of mind like humans.

This challenges the book’s argument that “language must be built on socialization and mental simulation” at a technical level.

4. Sacrifice of interdisciplinary depth

Because the book spans paleontology,

neuroanatomy,

psychology, and computer science,

some domain experts believe it lacks rigor in certain details:

Anatomical details: Some neurobiologists point out that,

the descriptions of certain brain regions’ functions lean more toward “functional localization,”

and neglect the increasingly important “distributed network” perspective in modern neuroscience.

On AI history: To align with the evolutionary narrative,

the book discusses less about parts of AI development that do not fit this logic (e.g., symbolic logic approaches).

5. Concerns about human-centeredness

Although Bennett emphasizes the evolution of intelligence,

the narrative still follows a “from simple organisms to complex humans” logic,

which some ecophilosophers criticize as “anthropocentrism.”

This view might lead readers to mistakenly see other animals (like octopuses or crows) as merely “intermediate products” toward human intelligence,

overlooking their own highly specialized evolutionary paths.

Summary: “A Brief History of Intelligence” is regarded as an excellent “big history” work,

very successful in popular science and interdisciplinary inspiration.

But for researchers seeking utmost precision,

it functions more as an insightful “hypothesis model,”

rather than an absolute scientific conclusion.

Its greatest value lies in reminding AI developers: intelligence is not just computation,

but a long-term accumulation of life solving survival problems in complex environments.