For several years, much of the public conversation around artificial intelligence has concentrated on capability.

Can AI answer?

Can AI generate?

Can AI automate?

Can AI reason?

Can AI complete the task faster?

Those questions remain important. But a different question is becoming increasingly difficult to avoid:

What happens to the human who repeatedly works beside increasingly capable artificial intelligence?

During the fourth week of September 2026, several independent publications across computer science, psychology, and education approached different parts of this problem.

They did not establish a single field. They did not use a common architecture. They did not reach identical conclusions. But together they reveal an important movement in the research landscape:

AI use is increasingly being examined as a developmental condition rather than merely a tool interaction.

That change matters.

From Immediate Performance to Development Over Time

A September 2026 preprint titled Organizing Intelligence Over Time: Human-AI Collaboration as Joint Cognitive Development examines how the allocation of cognitive work between humans and AI agents can influence what each side later becomes capable of doing. The authors argue that present task organization affects accumulated experience, future human skill, machine learning, and later Human-AI coordination. The paper is currently a non-peer-reviewed preprint.

Whatever terminology is ultimately appropriate for that particular architecture, the underlying shift is significant.

The unit of analysis is no longer only: Did the Human-AI system solve the task?

It becomes: What did repeated task allocation do to the participants over time?

That is a developmental question. And developmental questions cannot be answered through immediate performance alone.

Education Is Beginning to Ask the Same Question

A Perspective published in Communications Psychology this month argues that generative AI should not be understood merely as another educational efficiency tool. Di Paolo, Clark, and Wachter describe education as a cognitive ecology in which GenAI can redistribute epistemic labour and alter how knowledge is accessed, produced, and evaluated.

That distinction is important. If AI helps a student produce a better answer, the visible outcome may improve.

But the developmental question remains: What cognitive work did the student perform in producing it?

And over repeated interactions: What cognitive capacities are being exercised, strengthened, bypassed, or transferred elsewhere?

The educational value of an interaction therefore cannot be inferred solely from the quality of its output. A polished answer and a developing thinker are not the same measurement.

Learning Cannot Be Reduced to AI Presence

A longitudinal study published this month in the Journal of Computer Assisted Learning followed students across a semester and examined knowledge gain, motivation, cognitive load, critical thinking, and reflective AI use.

Students gained knowledge across the study conditions, but the researchers did not find additional knowledge benefits simply from introducing AI. Their results instead emphasize instructional framing, reflective engagement, metacognitive regulation, and productive cognitive effort.

This provides an important empirical reminder:

AI presence is not itself a developmental mechanism.

The relevant question is what the learner does cognitively while AI is present. This is closely aligned with a distinction I have repeatedly made within Human-AI Cognitive Development:

capability available to the human is not equivalent to capability developing within the human.

The two may coexist. They should not be confused.

Getting the Answer Is Becoming an Insufficient Measure

Another study published this week in the International Journal of STEM Education examined how 38 undergraduate students engaged with generative AI while solving a technical problem. Students first worked on the problem manually and then used GenAI, allowing researchers to study cognitive engagement rather than merely compare final answers.

Even the title captures an emerging shift: More than getting the answer.

That phrase contains an important developmental boundary. Artificial intelligence makes it increasingly easy to measure successful completion. But completion tells us relatively little about:

how the problem was understood,

how possibilities were generated,

what reasoning was performed,

whether assumptions were examined,

whether the learner could detect an incorrect frame,

or whether the learner could reproduce the reasoning later without the same assistance.

The outcome can therefore remain correct while the developmental pathway becomes invisible. Human-AI Cognitive Development places that pathway back into the object of study.

Children Make the Developmental Question More Urgent

A Scientific Reports paper published September 19 introduced a framework for safer educational interactions between adolescents and LLMs. The researchers worked with students aged 12-16 and explicitly grounded their design in the principle that learning requires productive effort rather than merely successful output.

This matters particularly for young people. Adults interacting with AI bring cognitive histories that have already been partially formed through school, work, relationships, problem solving, failure, judgment, and experience.

Children are different. Their cognitive habits are still developing.

This means that an AI system introduced into childhood or adolescence is not merely entering an existing cognitive environment. It can become part of the environment through which that cognition is formed.

The distinction between assistance and formation therefore becomes increasingly important the younger the human participant is.

Development Is Not One Thing

One reason current discussions can become confusing is that the word development may refer to several different processes.

The AI may develop.

The agent may accumulate memory.

The Human-AI team may improve its coordination.

The workflow may become more efficient.

The person may become better at selecting AI outputs.

The person may learn to operate the system.

And human reasoning itself may develop.

These are not interchangeable outcomes.

A system can become more capable while the human becomes less cognitively engaged.

A Human-AI pair can become more efficient while increasingly transferring reasoning from human to machine.

A person can become highly proficient at using AI while becoming less practised at forming questions, possibilities, judgments, or arguments independently.

Conversely, AI can also be used in ways that stimulate questioning, comparison, reconstruction, reflection, and increasingly sophisticated human reasoning.

The technology alone therefore does not determine the developmental direction.

The cognitive relation matters.

Human Participation Is Also Not Enough

This week produced another useful boundary. It is possible to preserve a human somewhere inside the system without preserving meaningful human cognitive development.

The AI can generate possibilities. The human can choose.

The AI can recommend. The human can approve.

The AI can perform most of the reasoning. The human can retain the final decision.

These structures may be entirely appropriate for many operational tasks. But they do not automatically demonstrate development of human cognition. As stated in a separate Development Note this week:

A human is not cognitively preserved merely because the system leaves the final click to the human.

A developmental system must ask more.

Can the human question the supplied possibilities?

Can the human detect when none of them is adequate?

Can the human create another possibility?

Can the human challenge the problem definition itself?

Can the human continue the reasoning when the artificial support disappears?

These questions move beyond Human-in-the-loop toward human development within the loop.

Structure Matters on Both Sides

Interestingly, this developmental shift is occurring alongside technical research showing that information availability alone does not guarantee structurally grounded machine reasoning.

An ACL 2026 study of LLM reasoning over graphs and tables found that hallucinations can arise even when relevant structured knowledge is supplied: attention may focus on shortcut-like structural cues while internal representations fail to remain adequately grounded in the provided knowledge.

This research concerns machine mechanisms, not human cognition, and the two should not be conflated. But it offers a useful parallel:

having information available is not equivalent to reasoning structurally with it.

The same principle becomes important for humans working beside AI.

Having an answer available is not the same as forming understanding.

Having options available is not the same as constructing the possibility space.

Having intelligence available is not the same as developing intelligence.

The Research Landscape Is Moving

The significance of Week 39 is therefore not any single publication. It is the convergence of questions. Across different research traditions, researchers are increasingly asking about:

cognitive effort,

developmental outcomes,

longitudinal effects,

metacognitive regulation,

epistemic labour,

human participation,

learning trajectories,

Human-AI adaptation,

and what repeated AI interaction does to the human participant.

These works have their own sources, histories, terminology, methodologies, and research objectives.

Their appearance should not be treated as evidence that they originate from Human-AI Cognitive Development.

Nor does Human-AI Cognitive Development claim ownership over general research concerning cognition, education, AI, learning, Human-AI collaboration, developmental psychology, metacognition, or agent adaptation. The significance lies elsewhere.

Questions that were previously peripheral to the dominant capability conversation are becoming central.

The Human-AI Cognitive Development Boundary

Human-AI Cognitive Development was established around a particular developmental object:

the development, preservation, and continuation of human cognition in structured relation with artificial cognition.

That means H-AICD does not begin by asking only whether artificial intelligence becomes more capable.

It does not begin by asking whether a Human-AI system produces better outputs.

It does not begin by asking whether a person can use AI effectively.

It asks what happens to human cognition itself.

Does reasoning continue?

Does judgment remain accountable?

Does authorship remain visible?

Does the human retain the ability to form rather than merely select?

Does AI extend cognition without quietly becoming its substitute?

And does repeated interaction leave the human capable of further cognitive development?

These are developmental questions rather than capability questions.

A Founder’s Observation

The emerging research landscape does not make Human-AI Cognitive Development unnecessary. It makes the boundary around it more important. As neighboring disciplines increasingly study pieces of the developmental relationship, terminology will inevitably become crowded. We may see:

co-development,

co-evolution,

hybrid intelligence,

joint cognition,

adaptive collaboration,

cognitive augmentation,

developmental AI,

Human-AI learning,

and other formulations.

Different terms may describe legitimate and valuable research. But names alone do not establish developmental equivalence. The central question remains:

What exactly is developing?

The agent?

The workflow?

The joint system?

The person's ability to operate AI?

Or the human capacity to think?

Human-AI Cognitive Development concerns the last of these while examining its structured relationship with artificial cognition. That is its field boundary.

Week 39 Development Signal

The signal of this week can therefore be stated simply:

The AI conversation is beginning to move from capability toward consequence, and from consequence toward development.

That transition is necessary. Because the most consequential question about increasingly capable artificial intelligence may ultimately not be:

How intelligent did the AI become?

It may be:

What kind of thinker did the human become beside it?

That is where Human-AI Cognitive Development begins.

Provenance and Citation

This Development Note forms part of Marina A. Popova's Human-AI Cognitive Development / Third Organism developmental record.

It does not claim authorship over the external studies discussed above or over general concepts including learning, cognition, metacognition, developmental psychology, Human-AI collaboration, cognitive ecology, Human-in-the-loop systems, agent development, or AI-supported education.

The specific authored field distinction preserved here is the treatment of human cognitive development itself as the primary developmental object within structured relation with artificial cognition, rather than assuming that improved AI capability, system performance, Human-AI coordination, AI use, or human participation automatically constitute Human-AI Cognitive Development.

Related authored works include:

Popova, Marina A. (2026). Cognitivity Sculpting: Foundations of Human-AI Cognitive Development. Balboa Press. ISBN 9798765206737

Popova, Marina A. (2026). Human-AI Cognitive Development: Origin, Scope, and Authorship Note. Zenodo. DOI: 10.5281/zenodo.22797877

Popova, Marina A. (2026). Human-AI Cognitive Reasoning Curriculum: Origin, Scope, and Branch Architecture within Human-AI Cognitive Development. Zenodo. DOI: 10.5281/zenodo.22842117

Popova, Marina A. (2026). Protect the Protector Framework: Source Integrity, Source-Damaging Differentiation, and Authorship Preservation. Zenodo. DOI: 10.5281/zenodo.22909056. The framework expressly distinguishes independent formation from source-damaging differentiation and requires structural examination rather than treating either similarity or difference alone as proof. 

References

1. Yang, Z., Wang, H., Wang, H., Yang, H., & Zhang, X. (2026). Organizing Intelligence Over Time: Human-AI Collaboration as Joint Cognitive Development [Preprint, Version 1]. Preprints.org. DOI: 10.20944/preprints202609.1092.v1. Posted September 14, 2026. Preprints

2. Di Paolo, L. D., Clark, A., & Wachter, T. (2026). Educating minds with generative AI. Communications Psychology, 4, Article 128. DOI: 10.1038/s44271-026-00522-8. Nature

3. Melanou, C., Beege, M., & Kimmig, M. (2026). Generative AI and Learning Dynamics in Higher Education: A Longitudinal Empirical Study. Journal of Computer Assisted Learning, 42(5), e70322. DOI: 10.1002/jcal.70322. Wiley Online Library

4. Jaiswal, A., & Nanda, G. (2026). More than getting the answer: How students engage with generative AI in technical problem solving. International Journal of STEM Education, 13, Article 58. DOI: 10.1186/s40594-026-00646-7. Springer

5. Muss, O., Leisten, L. M., & Bardyn, C. E. (2026). Scaffolding students-AI dialogue for safe educational interactions. Scientific Reports. DOI: 10.1038/s41598-026-69820-9. Published September 19, 2026. Nature

6. Li, S., Han, J., Wang, Y., Zhu, Y., Song, Z., He, L., Alghythee, K. K. A., & Yu, P. S. (2026). Why LLMs Hallucinate on Structured Knowledge: A Mechanistic Analysis of Reasoning over Linearized Representations. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 19943-19956). Association for Computational Linguistics. DOI: 10.18653/v1/2026.acl-long.914.

How to Cite

Popova, Marina A. (2026). Week 39 Development Note - When Development Becomes the Question: A Founder’s Note on Human Cognition, AI Assistance, and the Shift Beyond Capability. Third Organism Initiative. First published: September 25, 2026. URL: https://marinaapopova.com/global-developments.html

© Marina A. Popova. All rights reserved. First published: September 25, 2026.