Abstract

This paper proposes a concise explanation based on the Free Energy Principle (FEP) to account for the behavioral evolution of early Homo erectus from a theoretical biological perspective, particularly focusing on their production of stone axes. “Cognitive surprise” may have prompted early Homo erectus to occasionally exhibit non-traditional or anomalous behaviors. The co-evolutionary dynamics of these behaviors reveal patterns of emergence, disappearance, and re-emergence among Stone Age humans, akin to the game of “Snakes and Ladders.”

When these artifacts appear in the records of the Early and Middle Pleistocene, anthropologists and archaeologists usually interpret them as evidence of early humans climbing the imagined genealogical “ladder.” This interpretation is used to explain how human cognitive abilities gradually developed, leading to increasingly innovative skills and ultimately resulting in the cognitive superiority of Homo sapiens.

However, Héctor Marín Manrique, Karl Friston, and Michael Walker propose a different hypothesis: the behaviors of anomalous individuals are not always accepted by the group. The group may be unable to understand or imagine the potential advantages of these novel behaviors, or even incapable of expressing these differences. This failure in understanding, combined with sporadic demographic events, may lead to these anomalous behaviors being ignored and thus not transmitted to future generations. This situation is akin to sliding down a “snake” in the game of “Snakes and Ladders,” potentially creating discontinuities in the course of human behavioral evolution and resulting in evolutionary mysteries that are difficult to unravel.

▷ Manrique, Héctor Marín, Karl John Friston, and Michael John Walker. “‘Snakes and ladders’ in paleoanthropology: From cognitive surprise to skillfulness a million years ago.” Physics of life reviews (2024).
 

1. “Handaxes”: An Archaeological Case Study

Stone handaxes are among the evidence of human activity during the Early and Middle Pleistocene epochs [1-2]. They are classified as bifacial large cutting tools (BFLCTs), a category that also includes cleavers.

▷ An early handaxe and a cleaver from Western Europe (scale in centimeters).

The longitudinal and transverse symmetry of these handaxes is a common and remarkable feature, which existed as early as approximately 1.7 to 1.6 million years ago [3]. It is widely believed that this symmetry reflects the maker’s intentionality—that is, the shape of the tool was preconceived before its creation [4].

This preconception reflects a neurobiological cognitive ability of the makers: they were able to sculpt a desired three-dimensional shape, such as a handaxe, from a raw stone core [5-9]. Paleolithic archaeologists generally agree that morphological and technical regularities can be perceived in the shapes of these tools. This marks an essential difference between handaxes and other simple stone tools lacking such features, regardless of whether these simple tools originate from handaxe sites.

Although BFLCTs represent complex stone cutting tools made by removing large flakes from stone cores, they are not the earliest cutting tools. The oldest flaked stone tools appeared in Africa’s Late Pliocene around 3.4 million years ago, possibly created by Australopithecines. The oldest fossil bones attributed to humans can be traced back to 2.8 million years ago. Around the beginning of the Early Pleistocene, about 2.58 million years ago, our ancient ancestors began to show signs of cooperation. One of these signs is the tools they made by striking stones together, creating shell-like shapes well-suited for handheld gripping.

Compared to BFLCTs, these earlier tools were simpler to manufacture. More refined handaxes first appeared in East Africa around 1.76 million years ago, by which time Homo erectus had replaced several earlier hominins [10-14]. In South Africa, handaxes appeared around 1.6 million years ago [15]. Subsequently, handaxes sporadically appeared in the Paleolithic records of Africa and Eurasia, but their distribution in time and space was uneven. From the later Early Pleistocene onwards, between 1.5 million and 1 million years ago, BFLCTs appeared at a few sites in West Asia and South Asia. In other words, BFLCTs were continuously manufactured by various forms of humans over a period of more than 1.5 million years.

The simple stone tools mentioned earlier, such as choppers that appeared as early as 3 million years ago, continued to be found archaeologically until recent millennia. In other words, even after people began making more refined bifacial handaxes around 1.76 million years ago, the earlier, simpler techniques remained widespread, in sharp contrast to the intermittent manufacture of handaxes. The latter’s record was intermittent and sparse over the more than one million years following their first appearance, indicating it was not a common manufacturing technique. Their scarcity also suggests that hundreds of thousands of years may have passed between the invention of these bifacial stone tools and their important role in Paleolithic life.

▷ Chopper

This raises a series of questions. Since the manufacture of handaxes reflects clear intentionality, how did this understanding arise among early humans? How many times might it have arisen? Did it appear and disappear multiple times in different times and places? [16] Can we assume it arose only once and was skillfully passed down through generations and transmitted between communities, having a profound impact across time and space?

Some Paleolithic archaeologists and paleoanthropologists consider this possibility credible. It is based on a “progressivist” assumption that BFLCTs provided functional advantages to early humans. These advantages had obvious adaptive value, aiding survival and promoting gradual population growth and reproductive success. These skills facilitated population growth and geographic expansion by enabling early humans to extensively exploit resources across different ecological zones and biomes.

But this view has been challenged from many quarters. First, the temporal and spatial span of handaxes is enormous. Even if several archaeological assemblages containing handaxes are found within a geographical area of ≥500 kilometers and can be dated to roughly similar periods (possibly ≥200,000 years), it is difficult to assert that this necessarily represents a continuous, intergenerational “cultural tradition.” If an average generation lasts 25 years, then 200,000 years represents 8,000 generations. This poses a significant challenge to the possibility (let alone the plausibility) of explaining the archaeological record using “social transmission,” “cumulative culture,” and “cultural history” approaches.

Secondly, the temporal and spatial distribution of handaxes is irregular, exhibiting multiple instances of “appearance, disappearance, and reappearance” over time, and their spatial distribution is also relatively scattered. If we consider the manufacture of handaxes as a result of cultural transmission that brought survival advantages, it is hard to explain this sparsity and intermittence. Some suggest that the preservation of handaxes’ “social transmission” or “cultural transmission” in the archaeological record is incomplete, but this is merely a remedial explanation. Finally, if handaxes coexisted with simpler choppers, and the latter did not show significant disadvantages, why were choppers continuously produced?

This paper provides a concise explanation of this phenomenon based on the Free Energy Principle (FEP) [17]. It can explain not only stone tools like handaxes but also other technological developments of ancient humans. To this end, we first introduce a helpful analogy—the game of “Snakes and Ladders.”
 

2. The Analogy of the “Snakes and Ladders” Game

The ancient game of “Snakes and Ladders” is a board game suitable for two or more players. The game board is shown below. The ladders and snakes on the board connect different squares. At the start of the game, players place their tokens on the starting point (square numbered 1) and roll the dice to determine the number of steps to move forward; if a player rolls a six, they get an extra turn.

If a token lands on the bottom of a ladder, it can ascend to the corresponding top of the ladder; if it lands on the head of a snake, it must slide down along the snake’s body to its tail. The first player to reach the finish wins. Obviously, during the game, players’ tokens undergo multiple ascents and descents, sometimes even returning to earlier positions.

▷ Snakes and Ladders: https://en.wikipedia.org/wiki/Snakes_and_ladders

From the perspective of cumulative culture, the development of human technology and culture is like climbing a ladder, continuously ascending. But in reality, the “ascent of humanity” over two million years is much like this game—it has not been smooth sailing, nor has it progressed in a straight line. In an exaggerated sense, it can almost be seen as a fable describing how, in the process of technological and cultural development, we move at a snail’s pace—slowly, hesitantly, stumbling, and sometimes even regressing. The irregular temporal and spatial appearances of technologies like the handaxe can be understood in this way.

The following argument will explain why this developmental pattern is intermittent. This argument is rooted in the theory of self-organizing systems in physics. Simply put, self-organizing systems far from equilibrium inevitably exhibit a kind of wandering behavior, which in many ways is similar to the characteristics of the “Snakes and Ladders” game [17].
 

3. The Free Energy Principle, Active Inference, and Human Evolution

The Free Energy Principle (FEP) offers a fundamental approach based on statistical physics to understand how self-organizing systems (such as organisms) adapt through evolution and how sentient beings produce behavioral responses. In this context, we focus on animals possessing a Hierarchically Mechanistic Mind (HMM) [18-19].

HMM defines the embodied, situated brain as a complex adaptive system that generates perception-action cycles through dynamic interactions among hierarchically organized neurocognitive mechanisms, actively minimizing the entropy (i.e., dispersion or decay) of sensations and bodily states [19]. HMM can be viewed as a neurobiological inference machine [20], operating in accordance with the Free Energy Principle.

The Free Energy Principle states that all organisms aim to minimize free energy to maintain their survival. However, self-organizing systems that can adapt to their environment will, through active inference, respond to changing environments beyond mere perceptual predictions, consuming more energy. The quantity they attempt to minimize is defined as free energy, which measures the difference between external perceptual predictions and internal beliefs (preferences, priors). Simply put, perceptual predictions are “objective inferences” based on externally received information—for example, when walking and encountering rain, we predict we will get wet based on visual observation and the tactile sensation of raindrops—while internal belief preferences are the subjective desire not to get wet. The former is unrelated to whether we dislike getting wet.

Biological systems need to minimize the difference between sensory predictions and internal beliefs as much as possible. To achieve this, besides adjusting beliefs based on external information (e.g., accepting the fact of getting wet in the above scenario), they can also change the state of the external world through actions, such as running to take shelter from the rain. Under the framework of active inference based on the Free Energy Principle, an organism’s cognition and behavior follow the same rules, serving to minimize perceptual surprise—which is a form of prediction error. In information theory and Bayesian conditional probability analysis of predictive coding, this prediction error is often called surprisal or self-information, calculated as the negative logarithm of the probability of possible events.

▷ In the Free Energy Principle, the system’s states can be divided into four categories: the external state representing the environment, the sensation state of the agent’s observations, the internal state, and the action state.

For organisms in an ecosystem, negative free energy is regarded as a manifestation of their fitness, requiring them to optimize thermodynamic efficiency in their interactions with ecological niches. Free energy, as an informational measure, quantifies the average surprise of various outcomes for an organism. Therefore, when an organism can accurately model and predict its interactions with its ecological niche, we can say it has adapted to that environment because it avoids surprising exchanges with the environment (e.g., deviating from homeostatic set points) or avoids being in extremely low-probability states (e.g., injury or death).

Certain organisms, such as humans, have developed deep generative models and possess the ability to predict the outcomes of their actions. Such organisms can envisage counterfactual futures under different actions; in simple terms, they can plan [21-25]. Because these organisms incorporate their impending actions into their perceptual predictions, they can forecast potential outcomes and make decisions to perform active inference (planning).

This is a key aspect of active inference: selecting actions and plans based on minimizing expected free energy. Simply put, the choice of action aims to minimize expected surprise, to avoid low-probability adverse events (such as injury or death), and to reduce environmental uncertainty through the outcomes produced. Reducing uncertainty is crucial because it means that perceptual behavior under the deep generative model has an epistemic aspect, keeping perceptual behavior sensitive to salience and novelty.

Only systems with such deep generative models will exhibit this exploratory behavior because they are the only ones capable of responding to cognitive revelations and answering the question, “What will happen if I do this?” [26-28]. Therefore, the expected free energy driving plan or policy selection can be decomposed into pragmatic and epistemic components, supporting exploitative and exploratory behaviors, respectively.

To put it another way, if an organism anticipates the consequences of future behaviors, its mind is filled with possibilities and surprises before the action unfolds. This expectation of future results prompts the individual to try to reduce perceptual surprise, thereby enhancing control over the environment and gaining a deeper understanding of the possible consequences of various behaviors. This is the driving force behind exploratory behavior.

In the broader context of biological evolution, minimizing free energy supports not only the survival of organisms but also their ability to successfully reproduce offspring [29-30]. Natural selection is gradual and conservative; in the face of environmental interdependencies and regularities, the integrity of organisms’ generative self-organizing systems is supported by adaptive interactions, including active inference. This active inference is closely related to prior expectations induced by generative models and endowed by evolution [31-33]. Active inference instantiates a generative model whose components are neural networks in the brain capable of predicting the most probable upcoming sensations. In the Bayesian statistics of conditional probability, evolution and natural selection can be regarded as natural Bayesian model selection (also known as structure learning [34]).

Therefore, evolution proceeds gradually and intermittently at biological, technological, and even psychosocial levels [35-37]. The Free Energy Principle is related to biology, especially neurobiology, whether at the ontogenetic level of cellular dynamics, neural circuits, and behavior, or at the phylogenetic level of populations evolving through natural selection of biological adaptation and fitness.
 

4. “Cognitive Surprise” and the “Snakes and Ladders” Diagram

Now, let’s apply the earlier perspectives to interpret the archaeological findings related to handaxes. In discussing this topic, we need not overemphasize the role of “group size” in driving the social transmission of handaxe-making technology. Instead, we should focus more on the individual level. That is, what sparked the cognitive awareness in the makers regarding the possibility of crafting stone handaxes? And how did observers (other companions or the makers themselves) react to this innovative approach?

In an environment where traditional tools were mainstream, an individual who spends more time and effort to produce refined and more complex handaxes requires an exploratory inclination. This includes being aware of what results might arise from adopting different manufacturing methods (as seen in the first part of this paper, the form of the handaxe reflects the maker’s preconceived notions of shape). It also involves recognizing that the products of new technology might bring long-term practical value, namely the potential to reduce subsequent energy consumption. All these require the individual to possess deeper generative models.

Thus, we might consider that throughout the temporal and spatial span of handaxe existence, the overall cognitive abilities of various ancient human groups did not reach a level at which they could recognize the innovativeness and long-term value of handaxes [38-39]. However, occasionally, individual members might possess more expressive (i.e., deeper) generative models, discerning both the exploratory (cognitive) and exploitative (practical) value, thereby developing these more complex crafts under chance circumstances. Moreover, other individuals in the group (sometimes including these anomalous individuals themselves) often failed to realize the uniqueness of such innovative behaviors by their companions and did not experience “cognitive surprise” [40].

For bystanders, such behavior did not align with the typical activities in their behavioral repertoire, which were based on their practical prior knowledge. Their strong adherence to prior normative beliefs often overwhelmed the exploration of new cognition. This led to these new creations not being accepted by the group; the creators’ “skills” were seen as a waste of the group’s invaluable and indispensable energy and time needed for survival.

It is worth noting that, in maintaining basic homeostasis, our brain consumes 11.2 watts of energy per kilogram at rest, whereas the entire body consumes only 1.25 watts per kilogram. The brain uses 50% of our glucose intake and 20% of our oxygen intake, with 75-80% of the energy supporting neuronal activity in the brain [41].

Therefore, the activities of Homo erectus one million years ago had to meet greater daily energy demands, which was essential for their survival and, ultimately, for our existence today. In such circumstances, tried-and-tested routines became dominant in life. Those who could efficiently perform these tasks earned the group’s trust, while eccentric, unorthodox, or heterogeneous behaviors were ignored, leaving no traces in collective memory or lore. Moreover, when our ancient ancestors made stone tools by striking large stone blocks, they occasionally knocked off some peculiar shapes in the process. These shapes were actually produced unintentionally, and people at the time might not have realized that these accidentally formed stones could also be used as tools [42-43], even though archaeologists classify them as tools from a comparative morphological perspective.

▷ Vadim Sherbakov

In general, we believe that during the Early and Middle Pleistocene, the fate of those anomalous, heterogeneous behaviors among humans was one of ups and downs. The personal achievements seemingly reached on the technological “ladder” were likely ignored by their companions, who could not imagine or express the survival advantages these behaviors might bring. Furthermore, humans at that time lacked sufficiently fluent communication abilities, and, coupled with various demographic accidents, records of heterogeneous behaviors were often lost. Even if these technologies were remembered in some small hunter-gatherer groups, they might disappear upon extinction due to population fluctuations.

Human communication during that period was very limited; even if language existed, it was quite primitive. Additionally, humans at that time had shorter lifespans, reached biological maturity earlier than now, and had limited brain capacity (these smaller brains were only two-thirds the size of our brains today). In any case, this limited the neurobiological plasticity in small adult brains, thereby affecting deeper cognitive abilities*.

The fourth section of the original paper focuses on the detailed brain differences between Homo erectus and modern humans. Since it is not directly relevant to the overall logic of this article and serves only as supporting evidence, we will not elaborate here.

Moreover, groups faced various risks of extinction, such as unequal sex ratios, death during childbirth or congenital disabilities, infections, tooth loss, or scarcity of food or water caused by plagues, blights, droughts, floods, wildfires, frosts, blizzards, or other violent climatic events. These factors combined to make it difficult for the results of anomalous behaviors to be stably transmitted within populations.

The final question is, if these handaxes distributed across various locations and times were independently made by different individuals, why do they have certain morphological similarities? For example, transverse and longitudinal symmetry.

In this regard, we might understand that the nature of self-organization has inherent periodicity, randomness, and a tendency to minimize free energy. This causes the system to repeatedly explore certain specific states within a vast state space, leading to deep generative models exhibiting certain similarities in preferences (such as an awareness of symmetry; evidently, symmetry is cognitively more concise and conforms to the principle of minimization). At the same time, limitations of raw materials and physiological structures (such as ease of grip) also constrained the forms of manufacture. This allows tools like handaxes unearthed from different places and made at different times to have certain morphological similarities under independent manufacture. Apart from coincidence, this similarity also has certain evolutionary inevitability and is not necessarily the result of cultural transmission.

In summary, when examining the evolution of handaxe technology, we find irregularity and periodicity in technological development. This phenomenon is not a unique historical script of humans but a self-organizing behavior commonly existing in biological systems. The Free Energy Principle provides us with a powerful tool to understand this phenomenon.

Each technological “leap” or “slide” may be an adaptive response to changes in the external environment, including physical environmental changes and shifts in social, economic, and cultural contexts. In the contemporary society of globalization and rapid information flow, human technology and culture may exhibit even more complex dynamic changes.

For this reason, we should promote open innovation systems, allowing the existence and development of “non-traditional” thinking. Through such openness and collaboration, we can minimize the “surprise” elements in technological advancement and foster a more stable and sustainable environment for technological and cultural development.