Evolve Before Individual Learning Becomes Collective History

Evolve is a long-term research project focused on group-level emergence and digital evolution. Its central concern is not how to predesign a complex society, but a more fundamental question: when large numbers of simple individuals are placed under shared environmental constraints and can alter their behavior through action and feedback, can differences at the microscopic level gradually accumulate and eventually produce observable structures at the group level?

The project first established a digital environment that could run continuously and be observed over time. Populations could expand through finite space, encounter resource constraints, form boundaries with other groups, and undergo conflict, decline, and redistribution. This stage demonstrated something basic but necessary: complex macroscopic patterns do not have to be written in advance as a narrative. They can arise over time from the interaction of many local rules.

Yet such a system still had a clear limitation. Individuals were primarily components within predetermined dynamics, while changes in the world were driven mainly by predefined rules and stochastic processes. The individuals themselves were not the primary causal source of world change.

Evolve therefore shifted toward a different problem: bringing individual action directly into the causal chain of the world.

In the new experimental environment, individuals are no longer merely positions or colors moved by rules. They can take actions based on limited local states; those actions affect the environment, and their outcomes can in turn alter subsequent behavioral tendencies. Reproduction, resource acquisition, movement, conflict, and limited information exchange are therefore no longer only consequences automatically generated by global dynamics. They begin to depend on choices made by the individuals themselves.

The significance of this transition is not that the system appears more complex, but that the causal structure of the experiment has changed. Where the earlier system allowed us to study how rules generate population dynamics, the new system begins to let us study whether differences in individual behavior can accumulate into differences at the group level.

A Valid Negative Result

After this stage was completed, Evolve conducted comparative experiments on its central scientific hypothesis.

The basic hypothesis was that if individual learning exerted a sufficiently strong long-term effect within the system, then worlds in which individuals could adjust their behavior through experience should gradually become measurably different from worlds without that adjustment process after sustained evolution. Such differences should not remain confined to a single action by a single individual, but should extend into more persistent behavioral patterns, lineage structures, and group-level evolutionary outcomes.

The experimental results did not support this expectation.

The data showed that the learning mechanism did alter behavior during individual lifetimes. Individuals with learning enabled exhibited more behavioral adjustment, indicating that feedback was not merely formal: it had a real effect on subsequent actions.

However, these differences did not continue to amplify reliably at higher levels. At the predefined group and lineage scales, the differences between learning and control conditions remained limited and did not reach the level required to support the original hypothesis. In other words, the current system can demonstrate that experience can change individual behavior, but it cannot yet demonstrate that those changes are sufficient to reshape long-term group structure.

This is a negative result, but not an invalid one.

The experiment completed the comparisons specified in advance, and the data quality was sufficient to support a judgment. There is therefore no reason to revise the criteria, reselect samples, or redefine success simply because the outcome did not match expectations. Instead, the result narrows the problem space. The central question is no longer whether learning exists, but why learning that clearly occurs has not produced a sufficiently strong macroscopic amplification effect.

A Gap Between Individual Change and Group Change

The central problem now exposed by Evolve can be summarized as a causal chain that has not yet been closed:

Individual experience → behavioral change → information retention → intergenerational transmission → persistent divergence → group structure

Existing experiments provide some evidence for the beginning of this chain. Individuals can change their behavior in response to feedback, but the evidence becomes much weaker beyond that point. Whether behavioral differences can be preserved over long periods, whether they can cross reproduction into the next generation, whether they can continue to accumulate across multiple generations, and whether that accumulation is ultimately sufficient to alter group-level historical trajectories all remain unresolved.

This also means that simply adding more behaviors, more complex decision structures, or larger models is not the most valuable next step. If it is still unclear whether information already being produced can persist through time, increasing individual complexity may generate more short-term variation without addressing the actual bottleneck.

What matters more at this stage is the persistence of information.

Behavioral differences formed during an individual's lifetime must leave some kind of trace if they are to participate in evolution at longer timescales. If those differences largely disappear when the individual dies, or are rapidly overwritten by stochastic variation during reproduction, then even substantial individual learning may remain confined to the local scale.

Conversely, if small behavioral differences can persist across multiple generations and continue to influence how descendants respond to their environment, then time itself may become an amplification mechanism. Small initial differences would not need to transform an entire population immediately. They could instead accumulate gradually through deeper generational histories and eventually produce divergent evolutionary trajectories.

The Next Step Is to Study History, Not Simply Run Longer

Evolve's next research focus will therefore move from whether individuals can learn to whether information produced by learning can enter history.

History here is not equivalent to the number of hours a program has been running, nor to the number of state updates it has undergone. For an evolutionary system, a more meaningful timescale is generational continuity: whether an individual's experience can influence descendants through reproduction, whether that influence can continue into subsequent generations, and whether differences between distinct starting points remain after sufficiently deep generational chains.

This requires a research perspective different from simply observing population size or spatial distribution. We need to distinguish actual genealogical relationships among individuals, trace how behavioral strategies change over a parent's lifetime, and observe how much of those changes enter the offspring and how much remains across later generations.

This form of observation does not alter the individuals themselves, nor does it introduce new capabilities merely to produce a more visible result. It is first and foremost a measurement problem: making intergenerational information, currently hidden behind group-level patterns, observable.

If longer histories amplify existing individual differences, that would indicate that previous experimental timescales were insufficient for the effect to become visible. If the differences still disappear rapidly, the problem can be narrowed further toward transmission, retention, selection, or environmental constraints.

Either outcome would be more scientifically valuable than simply increasing system complexity.

Evolve Is Moving from a Running World Toward a World with History

Evolve initially asked whether a digital world could generate observable population dynamics. The research then moved to the individual level, making action and feedback part of the processes that change the world. The current negative result shows that action and learning alone are still insufficient to guarantee significant divergence in macroscopic structure.

This places the project in a more clearly defined research position.

Two things that are often conflated now need to be separated. A system can continuously change without those changes becoming history. An individual can continuously learn without the information acquired through learning becoming a persistent participant in group evolution.

The important transition occurs when information gains continuity through time.

If individual experience can survive across generations and continue to be selected, modified, and recombined over long periods, then group-level structure may eventually become not only the result of immediate dynamics, but also the result of accumulated history.

Evolve's next research will focus on this boundary.

The aim is not to prove that complex structures must emerge, but to determine under what conditions experience at the individual level can become history at the group level.