The Evolution of Life and Intelligence: Environmental Changes and Attention
I am conducting personal research on the origins of life from a systems engineering perspective. Instead of perceiving life phenomena as chemical substances or organisms, I attempt to explore the process from inanimate matter to life as an evolution of self-organizing systems.
When capturing life phenomena as systems, it is essential to focus on the relationships of mutual influence, including the interactions among the components of oneself and the interaction between oneself and the external environment. I believe that a circular relationship, where the given impact eventually returns to the starting point, is the key to life phenomena, and I refer to it as the “effects loop.”
In this article, I will focus on the effects loop within systems and contemplate its evolution. In the latter part of this article, I will apply this concept not to the origins of life or organisms but to intelligence, expanding the discussion to topics in artificial intelligence. I will then apply the technical concept of “attention,” currently a hot topic in chat AI, back to the origins of life.
Effects Loop and Evolution
The relationships between organisms and the relationships between organisms and the environment can be seen as having many effects loops where they influence each other.
Organisms that have evolved according to the principle of survival of the fittest form many positive feedback loops as effects loops. Organisms capable of forming more robust positive effects loops are more likely to survive without being eliminated.
Evolution of Effects Loop in the Origins of Life
This structure is not only seen in biological evolution centered around DNA. I believe that this structure existed even in the chemical evolution stage, which is said to have occurred at the origins of life before the appearance of DNA.
Chemical substances and the chain of chemical reactions can form an effects loop. Chemical substances that form a positive effects loop should be generated sustainably.
In this way, even in the world of chemical substances, more robust effects loops are created overall, and their number and types may increase over time. Thus, chemical evolution can also be captured as the evolution of the effects loop.
Stagnation of Power Relations in Effects Loop
When effects loops intertwine, the solidification of the power relations within the loop leads to evolutionary stagnation.
Various changes and fluctuations in the Earth’s environment affect many effects loops differently. This breaks the solidification of the power relations in the effects loops. And by the fluctuation of power relations, evolution can proceed without stagnation.
On the other hand, even without significant changes in the Earth’s environment, the emergence of new species due to DNA mutations can change the power relations of the existing effects loops. Therefore, evolution is also promoted by DNA.
In terms of the entire ecosystem, both external changes in the Earth’s environment and internal changes in DNA exist.
Change of Power Relations in the Origins of Life
On the other hand, in chemical evolution, there might have been some mechanism of internal change in place of DNA. Systems like the RNA world hypothesis or protein world hypothesis, where RNA or proteins are the main subjects, are considered.
However, going further back, there must have been a stage before such internal change mechanisms were established.
If there were no mechanisms for internal change, we would have to believe that substances coincidentally combined to create such mechanisms to prevent evolutionary stagnation. However, understanding the complexity of this mechanism, it is difficult to believe that it emerged by chance.
If so, until the mechanism for internal change was established, we can only assume that evolution progressed by relying only on the mechanism of external changes to overcome the solidification of the power relations in the effects loop.
Internalization of the Mechanism of Change
The concept is that the external mechanism of change in a system promotes evolution, and through evolution, a mechanism of change is also built within the system. The internal mechanism of change should be suitable for promoting evolution. Indeed, DNA mutation is a surprisingly suitable mechanism for evolution.
As a result, it is believed that after acquiring an internal mechanism of change, evolution can occur much more rapidly compared to the era that depended on external change. In other words, although evolution is promoted by external changes, the focus of evolution promotion has shifted to the internal mechanism of change.
This can be viewed as the internalization of the mechanism of change, necessary as an enabler for evolution.
Internalization of Life System Infrastructure
I perceive the chemical evolution at the origin of life as a system of effects loops. From this viewpoint, some examples can be found where mechanisms that initially depended on the Earth’s environment are eventually internalized.
For chemical evolution to proceed, diverse chemical substances must meet, and various chemical reactions must occur. Therefore, chemical evolution is believed to have particularly progressed in water.
However, if all substances enter the same water body, diversity is greatly lost, as they neutralize or interfere with each other, and fast-reacting substances use up the chemical materials for reactions first. Therefore, bias within a water body is necessary for diverse chemical reactions to occur.
I believe that chemical evolution proceeded in numerous ponds and lakes on Earth. To ensure the effects loop of chemical evolution works efficiently, the cyclic movement of chemical substances is necessary, utilizing Earth’s water cycle — the flow of rivers, rising air currents with sea water evaporation, cloud movement by wind, and rain falling and flowing back into rivers.
Eventually, as chemical evolution progresses, a prototype of the fibrous cellular skeleton is created, and a prototype cellular tissue that promotes chemical reactions attaches around it. Also, accompanying the viscosity similar to the cellular matrix prevents the dissipation of chemical substances, creating bias. The movement of chemical substances between tissues is realized by a mechanism along the fibers. Further evolution leads to the formation of membranes as partitions.
In this way, I believe the movement, bias, and partitioning of chemical substances, initially dependent on external factors, shifted to internalization.
The mechanism of change as an enabler for evolution can also be considered as one of the things that shifted to internalization.
Conditions Leading to Internalization
If life phenomena infrastructure is internalized, external environmental mechanisms functioning as infrastructure hold a very significant position for the origin of life.
As seen earlier, Earth’s water cycle and numerous ponds and lakes could handle the movement and partitioning of substances.
Also, various environmental changes on Earth, such as day and night, seasons, tides, cooling and warming, gradual changes like weathering and erosion, sudden changes like earthquakes and floods, and macro changes like the movement of continents and ocean currents due to plate movement, may have worked well as enablers for evolution.
In the early stages of chemical evolution, the external environment undertook many life phenomena infrastructures, and as evolution progressed, it gradually internalized.
Of course, if the mechanisms or the number and types of partitions for the movement of chemical substances are insufficient, the effects loop responsible for life phenomena will not be sufficiently formed.
Even if the effects loop is sufficiently formed, if the types or frequency of environmental changes are scarce, the power relationship of the effects loop becomes fixed, and evolution does not proceed to internalization.
If life phenomena reached the internalization of infrastructure through chemical evolution, the ancient Earth must have had enough environmental mechanisms and conditions in place.
By systematically identifying the types of infrastructure that life phenomena require, it should be qualitatively understandable what kind of mechanisms were necessary in the environment at the initial stage of chemical evolution. And the quantitative conditions such as the number, type, and frequency of mechanisms held by the environment should also become clear for the evolution to progress to the stage where the effects loop can internalize those life infrastructures.
If these conditions become clear, it should be understood whether Earth’s environmental conditions greatly exceeded, just met, or did not meet the conditions necessary for the birth of life. Clarifying this could provide new situational evidence for the discussion of whether life was born on Earth or came from space.
Potential Application to Large Language Models
We have discussed how the solidification of power relationships in the effects loop, which leads to evolutionary stagnation, can be disrupted by diverse environmental changes. This mechanism, which promotes evolution, can also be related to large language models, such as chat AI.
In large language models, machine learning is performed by repeatedly making AI read a vast amount of text. The given texts, which are understandable by humans, contain grammar, expressions, words, and logical development of stories, with certain rules, yet they encompass a wide variety of content, topics, and genres, including texts that come to different conclusions about similar topics.
This resembles how the Earth’s environment, with its physical laws, basic arrangements and flows like terrain and water circulation, and regular changes like day and night and the shifting of seasons, provides diverse changes to the groups of effects loops formed by chemical substances undergoing chemical evolution during the origins of life.
Chat AI via large language models is said to have phases where its intellectual level greatly changes depending on the number of parameters in the AI model and the amount of text given for learning. At such times, the AI may reach a stage where it can perform analogies for knowledge not directly contained in the learned text or construct logic step-by-step and think about things deeply and theoretically.
This also reminds us of chemical evolution at the origin of life. At the origins of life as well, there seems to have been a significant evolutionary stage where external infrastructure could be internalized. Considering this, even in the learning of large language models, evolution similar to the internalization of infrastructure at the origin of life may exist.
The Future of Large Language Models
Furthermore, if we think more deeply, two things that life eventually followed in chemical evolution are foreseen to also come to AI as the internalization of various external infrastructures progresses. One is the pathway to logically construct thought within oneself without relying on external text input infrastructure and make new theoretical or philosophical discoveries.
The other is differentiation. A large chemical evolution system that depended on infrastructure like the Earth’s water circulation and pond distribution can be considered to have differentiated into individual systems called cells. Similarly, if given a very large neural network, AI may create a state where multiple AIs coexist within one system as learning progresses.
This leads to an interesting debate on whether it’s more advantageous to think as one entity when given a neural network much larger than the human brain, or to function as a collection of multiple specialized entities like human society. Assuming that AI can learn and self-organize to internalize its infrastructure like chemical evolution, if it is intellectually optimal to differentiate, such learning could well occur.
Moreover, differentiated groups of AI might demonstrate cooperation like multicellular organisms, or organizations in human terms. This may not only provide scholarly abilities to hold and produce new knowledge but also practical problem-solving and delicate considerations using diversity in interpersonal contexts.
Attention in Large Language Models
The buzz around current chat AI is backed by the technological advancement of large language models. A technical keyword essential for the success of large language models is “attention.” Although it means notice, in the case of large language models, I believe the expression “focus” is more appropriate.
“Attention” is one concept in AI training. To put it plainly, it is the idea that within each sentence, some words should be focused on.
For instance, in the sentence, “By the way, I ate a red apple yesterday,” words like “I,” “apple,” and “ate” hold significant meaning, while parts like “By the way,” “yesterday,” and “red” are supplementary. However, if someone had been talking about how “red foods are good for the body,” then the part “red” holds meaning. Or if there was a conversation about how “the three apples bought the day before yesterday have reduced to two today,” the word “yesterday” should be the focus.
When training AI on text, if all words are considered equally meaningful, it doesn’t grasp the essence of the sentence well. Hence, with the concept of attention, knowledge is imparted to the AI that a sentence contains words to be focused on and others not to be, as seen in the example above.
This way, the AI, while learning the sentences, also learns to grasp the trick of identifying the words to focus on. In certain sentence patterns, focus should be on the subject and predicate, while in others, words signifying time or adjectives may be important. Furthermore, by understanding the relationship with previous and following sentences, the AI can grasp the words to focus on, thereby understanding the so-called context. This leads to the creation of a chat AI that can understand and respond to sentences like humans.
Attention in the Effects Loop
If there’s a commonality between large language models and chemical evolution from the perspective of the effects loop, this concept of attention might have been observed in chemical evolution as well. And if chat AI has made a significant leap with the introduction of the concept of attention, a similar concept in chemical evolution might be a major key to understanding the origins of life.
Here, the fixation of power relations in the effects loop comes into play.
With environmental changes, the power relations between effects loops also change. However, within the range of environmental changes that can occur in everyday situations, dominant effects loops that hardly change may exist. This is similar to how, in general sentences, the subject and predicate continue to be the words to focus on, even if the content changes.
If such dominant effects loops have not yet been found, as chemical evolution is promoted with changes, such dominant effects loops will be formed.
On the other hand, with major environmental changes beyond the everyday range, dominant effects loops will also change. Even after such environmental changes, dominant effects loops that are not affected by other changes within that situation should form.
The change in situation with environmental changes can be likened to the change in the context of sentences. In other words, it is similar to focusing on words related to color when the topic is about color, and focusing on words related to time when the topic is about time.
For example, with the change in situations like day and night, spring, summer, autumn, and winter, the effects loops that become dominant will change. And under each effects loop, relationships can be built where effects loops that hold power in other situations are not destroyed but protected. In this way, even with various situational changes, some effects loop can always actively operate with positive feedback, allowing the entire effects loop as a robust system to continue to exist.
Thus, I think that a mechanism where appropriate effects loops are activated with the change in situation might correspond to attention in chemical evolution.
In Conclusion
In this article, I thought about chemical evolution in the phenomena and origins of life from the systematic aspect of the effects loop. It touches upon the idea that changing the power relations between effects loops, instead of fixing them, is the key to evolution.
Moreover, it discusses the possibility of finding common ground by viewing AI from the perspective of the effects loop, and touches upon the idea of applying the concept of attention, which underlies the technology used in large language models, to chemical evolution in the origins of life.
Additionally, in this article, from the perspective of endogenization, it touches upon how, as evolution progresses, the functions that the system needed to rely on the environment for, are gradually replaced by the functions of the produced chemicals. And it expands the imagination that this endogenization will appear in artificial intelligence as well.
This kind of thinking and research that transcends academic fields is called interdisciplinary, or interdisciplinary research. In this article, while thinking interdisciplinarily, it repeatedly applies hypotheses and insights gained between different fields to other fields. This, in a way, forms an effects loop between different fields.
Though it’s mixed with hypothesis and inference, if such an interdisciplinary effects loop becomes a positive loop, it may lead to new discoveries regarding the elusive and mysterious life and intelligence.
