There’s a notion called chemical evolution, which suggests that inorganic chemical substances evolved to approach life, describing the process leading up to the appearance of the first living cells on Earth.
While biological evolution can be explained by focusing on DNA with its capabilities of self-replication and mutation, chemical evolution doesn’t rely on these functions of DNA. This raises the question: what did it rely on?
My hypothesis is that the Earth’s water cycle system serves as the foundation for chemical evolution. If you think in this way, it can be seen that the structure of rivers, being water flows, had a significant impact on the birth of life through chemical evolution.
In this article, I will first provide a brief explanation of my hypothesis regarding chemical evolution. Then, I’ll touch upon the network structure of rivers, explaining how its structure resembles that of neural networks, a key technology in artificial intelligence.
While there are differences between rivers and neural networks, if one considers rivers as creators of life and neural networks as creators of intelligence, it seems that there may be some meaningful correlation in their structural similarities.
Both life and intelligence are enveloped in many mysteries, and investigations are ongoing scientifically, philosophically, and technically. There are many similarities pointed out between these two phenomena. By exploring the similarities between rivers and neural networks, we might find clues to deepen our understanding of the commonalities between life and intelligence.
Let’s delve deeper.
Hypothesis on the Mechanism of Chemical Evolution
My answer to the question regarding chemical evolution is that the Earth’s water cycle and terrain were used to create a self-reinforcing feedback loop, enabling various chemical combinations.
Inorganic and organic materials can move from mountains to seas along water flows. Substances smaller than cells can easily drift in the air, moving from the sea to mountains via water vapor and clouds.
Riding this water cycle, chemical substances also circulate, allowing various chemicals to meet and react using energy from the sun’s heat and light, geothermal energy, and lightning. This can produce diverse chemical compounds.
Moreover, this water cycle loop can form a self-reinforcing feedback loop.
When a new chemical compound is synthesized, it might induce some changes in the loop, depending on its nature and the properties of other chemicals already present. If such changes promote the synthesis of that new compound, it will proliferate and establish itself in the loop. This can be interpreted as a step forward in chemical evolution.
Such steps accumulate, increasing the diversity of chemicals within the loop. As evolution progresses, more complex chemical structures will be formed.
In this way, the variety of stably reproduced chemical substances increases, and more complex chemical structures are produced. My hypothesis is that, without relying on the self-replication and mutation functions of DNA, the chemical evolution of substances progressed by utilizing the Earth’s water cycle.
River Network Structure
Within my hypothesis, water circulation is key. However, for new chemical substances to encounter each other, the structure of rivers becomes significant. If rivers simply flow in a straight line from mountains to the sea, opportunities for these chemical encounters would likely be limited.
Rivers on Earth are intricate and diverse.
Water evaporated from the sea forms clouds, which rain down on mountains. This water then flows as streams. Over time, these small streams converge to form larger rivers. Rivers don’t just merge; they also branch out at certain points.
Convergences and divergences occur throughout a river’s course. Additionally, bodies of water like ponds and lakes are fed by multiple rivers and, in turn, flow out into various rivers, resulting in merging and splitting.
I believe chemical reactions can also occur within flowing rivers, but stable reactions would ideally take place where water and substances are stored. Hence, the presence of ponds and lakes along rivers is also crucial for chemical evolution.
This river structure resembles the structure of neural networks.
Typically, in a neural network, processes flow in one direction, from the input layer to the output layer. Neural networks also have nodes where information accumulates, and the network is formed through connections between these nodes. In the case of rivers, ponds and lakes correspond to these nodes.
The structures of river networks and neural networks are similar, in that they have a defined flow direction and repeatedly merge and split around nodes.
Differences between Neural Networks and River Networks
While river networks and neural networks have similar structures, they also have differences. Let’s briefly discuss these differences.
Difference in Flow Complexity
Firstly, what flows between nodes varies in complexity.
In neural networks, information flows, but this information is typically a single number, rather than complex structured data. On the other hand, rivers carry substances, some of which have complex structures. This means simulations of chemical evolution in river networks might require exchanging more complex information between nodes than in neural networks.
However, not all nodes in a neural network are simply connected flatly. Networks can be constructed with blocks of nodes connecting with each other. From a macro perspective, focusing on exchanges between these blocks, more than one value might be exchanged. Thus, at this macro level, one could potentially handle complex information analogous to structured chemical substances in rivers.
Difference in Feedback Mechanisms
Another significant difference lies in feedback mechanisms.
In neural networks, if the output isn’t as expected, feedback is given to each node to improve subsequent results. This process is called “learning.” For neural networks, when giving feedback for learning, information flows backward from the output layer.
In contrast, while rivers don’t “learn,” they form feedback loops to promote chemical evolution, with feedback flowing from the sea back towards the mountains. In neural network terms, this is akin to taking the output from the output layer and feeding it back into the input layer.
This difference is because while neural networks enhance their abilities towards a “correct” answer through “learning,” rivers facilitate “creation” in their chemical evolution without a specific “correct” pathway.
Unfortunately, due to my limited knowledge, I’m not fully aware of the latest advancements and research in neural networks. However, for neural networks designed for “creative” applications like generative AI, techniques may already exist that provide feedback in a manner similar to river structures.
Summary of Differences
The differences mentioned above, in terms of flow complexity and feedback mechanisms, seem apparent in basic learning-type neural networks as I understand them.
However, if more advanced “creative” neural networks have been realized, they might be very similar to rivers in terms of flow complexity and feedback mechanisms.
Nature of Neural Network Size
From the perspective of seeing similarities in the structure of rivers and neural networks, it’s possible that a characteristic observable in one could be highly informative for understanding the other.
Given this view, insights gained from studying the impact of network size in neural networks may offer valuable insights into chemical evolution in rivers.
In large neural networks, commonly referred to as “large language models”, like the one used in chat AIs, there seems to be a tendency for intellectual capability to increase as the network size grows up to a certain point. This doesn’t simply mean that the network can store more information or answer questions more accurately.
Up to a certain size, a neural network might merely return answers from memorized sentences, but beyond that size, it starts to produce associative responses. And as it scales further, it gains the ability to answer in a more inferential manner.
This wasn’t intentionally designed by AI developers. As the number of nodes in the neural net and their connections increased based on the fundamental AI technologies, unanticipated capabilities emerged.
About River Network Size
Applying this idea of network size to the relationship between chemical evolution and rivers brings interesting possibilities to light.
If the structure of Earth’s rivers — that is, the network of streams flowing from mountains to seas — were simpler, the structures or functions of Earth’s organisms might also have been simpler. Or if a certain level of complexity, like primitive cells, was needed for life, life might not have arisen in simpler terrains.
This relates to discussions about the possibility of life on exoplanets. Even if all conditions resembled Earth’s, if by chance the terrain was simple with insufficient numbers of river networks, ponds, or lakes, that planet might not harbor life like Earth’s.
If river network size is vital, we could determine which ancient Earth regions had more active chemical evolution. Also, by identifying the required number of ponds, lakes, and river junctions in the network, we could potentially specify regions where life could have originated.
Furthermore, as research progresses in this area, we might determine whether the emergence of life on Earth was a miraculous low probability event or a highly probable one.
Thus, by exploring the relationship between neural networks and rivers, we might gain these insights.
In Conclusion
The idea that Earth’s water cycle served as the foundation for life is a hypothesis from my personal research. Thus, this article, which seeks similarities between river network structures and neural network structures, is essentially layering one hypothesis on another, which might seem a bit ambitious.
However, as described in this article, the reality of neural networks capable of intelligent functions might offer a new way to view my hypothesis about life emerging from the water cycle. By accumulating insights from this perspective, if we can discern similarities and characteristics between intelligence and life, it could, in turn, reinforce my hypothesis about life’s origins from the water cycle.
Especially as mentioned in the article, applying the intricacies and feedback loops observed in river networks to neural network research might be beneficial. If this leads to something akin to creativity in AI, it would strongly support the hypothesis that river networks possess creativity.
