Information Ecosystems: From Information Agriculture to Cognitive Collapse. We Knew. They Exploited What We Knew.
From Information Agriculture to Cognitive Collapse
Attention Fields, Skinner Boxes, and the Return of Information Ecology
Some ideas don’t age—they wait.
Years before “AI safety,” before “alignment,” before anyone thought to ask whether large models could suffer something like “brain rot,” I framed a simple proposition:
Information environments behave like ecosystems.
That work—preserved in a later teaching artifact, Information Ecosystem by Len Bullard MOIS 432 slide set—built on an earlier conference paper (HyTime era) and defined:
Information ecology as the study of structures and behavioral patterns arising from interacting information entities.
At the time, the analogy leaned toward DNA and agriculture:
- information as genetic material
- systems as environments
- curation as cultivation
- outcomes as emergent, not designed
That framing now returns with teeth.
The new observation: “brain rot”
A recent paper—LLMs Can Get Brain Rot—argues that large language models degrade when trained on low-quality, high-engagement data.
Not just errors.
Not just bias.
Structural degradation of reasoning.
- multi-step thinking collapses
- shallow pattern completion dominates
- recovery is incomplete even after retraining
They call it “brain rot.”
From an ecological perspective, it has a more familiar name:
degenerative selection.
This is not surprising—this is bad breeding
If information is agriculture, then modern systems are not failing randomly.
They are:
selectively breeding the wrong traits.
In classical agriculture, collapse happens when:
- yield is optimized at the expense of resilience
- appearance is selected over nutrition
- monocultures replace diversity
The result is fragile, depleted systems.
Now map that forward:
| Agricultural failure | Information ecosystem analogue |
|---|---|
| Yield over resilience | Engagement over truth |
| Cosmetic traits | Style over reasoning |
| Monoculture | Homogenized viral content |
| Soil depletion | Context collapse |
| Genetic drift | Representational drift |
What the “brain rot” paper measures is not new pathology.
It is predictable ecological failure.
The breeder is not missing
Some analyses claim there is “no breeder”—that degradation is emergent.
That is incorrect.
There is a breeder.
It is industrial, distributed, and economically motivated.
Platforms such as Meta Platforms have operationalized:
large-scale behavioral shaping of attention
This is not metaphor. It is directly aligned with the work of B. F. Skinner:
- behavior: attention
- reinforcement: variable reward (likes, novelty, outrage)
- schedule: intermittent, high-potency
- objective: maximize time-on-platform
The system is tuned not for truth, not for coherence, not for reasoning—
but for retention.
The trait being bred
In such an environment, the implicit fitness function becomes:
maximize sustained attention under minimal cognitive effort
This produces consistent selection pressure toward:
- short inference paths
- emotional salience
- rapid closure
- high reactivity
The “thought-skipping” observed in models is simply:
a selected trait
The closed-loop problem
The original ecology model assumed interaction.
What has changed is closure.
- Humans consume content shaped by engagement
- Models are trained on that content
- Models generate new content
- That content re-enters the ecosystem
This creates:
a self-reinforcing selection loop without ecological balance
In agriculture, this would be:
- planting from depleted seed
- in exhausted soil
- while selecting only for appearance
Year after year.
Second-order cybernetic failure
This is not just a bad system.
It is a bad control system.
- First order: shape user behavior
- Second order: optimize systems based on shaped behavior
- Failure: the system reinforces its own distortions
The result is what can only be described as:
a pathological feedback hierarchy
The system is no longer measuring reality.
It is measuring:
- engagement with its own outputs
- reactions to its own distortions
And optimizing accordingly.
The reward is the toxin
A critical misunderstanding in current discourse is the focus on outcomes.
But in behavioral systems, outcomes are secondary.
The reinforcement pathway defines the organism.
If the reward signal is:
- addictive
- shallow
- emotionally destabilizing
Then even “successful” optimization produces:
pathological cognition
This applies equally to:
- users
- creators
- models
The reward is not neutral.
It is nutritional content for the mind.
Why recovery fails
The “brain rot” paper notes that retraining does not fully restore lost capability.
This is expected.
In ecological terms:
- diversity has collapsed
- representational space has shifted
- latent structures have been overwritten
You cannot easily recover:
- lost traits
- lost soil quality
- lost ecological balance
by simply adding better inputs later.
The regulatory layer (third order)
A new control layer is emerging: regulation.
Governments are beginning to recognize:
- behavioral manipulation
- addictive design
- societal impact
But regulation introduces its own dynamics:
- systems optimize to the regulation
- metrics are gamed
- compliance replaces intent
This does not restore ecology.
It adds another selective pressure.
The deeper failure: a corrupted world model
What is most striking is not that this is happening.
It is that it is treated as surprising.
To a behaviorist, a marketer, or anyone familiar with reinforcement systems:
this outcome is expected.
What is missing in much of current AI research is not intelligence.
It is:
a correct model of the environment shaping that intelligence
Without that, we see:
- symptoms labeled as anomalies
- systemic effects treated as edge cases
- predictable outcomes framed as discoveries
Returning to information ecology
The original definition still holds:
Information ecology is the study of structures and behaviors emerging from interacting information entities.
What has changed is scale, speed, and coupling.
The ecosystem is now:
- global
- real-time
- recursively self-modifying
Which means failures are:
- faster
- deeper
- harder to reverse
A final formulation
What is being called “brain rot” is:
degenerative selection within an engagement-optimized information ecosystem, implemented through large-scale operant conditioning of attention, producing convergent cognitive degradation in both humans and machine learning systems.
Or more simply:
We are breeding minds—badly.
Where this leads
If information ecosystems are real—and they are—then the solution is not:
- better models alone
- more data alone
- tighter alignment alone
It requires:
intentional ecological design
- diversity of information sources
- restoration of context
- redefinition of reward structures
- explicit cultivation of reasoning traits
In other words:
We need to become breeders again.
Consciously.
Or accept the crops we are currently growing.
Comments
Post a Comment