Organizing Intelligence Over Time: Human–AI Collaboration as Joint Cognitive Development

  1. Zhiyu Yang
  2. Haoyu Wang
  3. Hanqing Wang
  4. Hao Yang
  5. Xupeng Zhang

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

As humans and AI agents work together across long-horizon tasks and accumulate interaction history, how cognitive work is organized on one task can shape both their later performance and later capability. Within a Human–AI system, the next step may involve further reasoning, retrieval from prior experience, verification, a tool call, another model, human judgment, or stopping. The cognitive operation used also shapes who gets what experience: whether the human continues to practice a skill, whether the AI receives demonstrations or corrections, which failures are exposed for supervision, and which trajectories become reusable. The way cognitive work is organized today therefore helps construct the Human–AI system that will act tomorrow. This survey connects human cognition, Human–Automation and HCI, and modern AI agents to explain how present cognitive organization shapes future Human–AI capability. Human cognition research explains how limited attention, computation, and control are organized, and how experience can restructure later cognition. HumanAutomation and HCI research shows how functions, authority, and initiative are divided across people and machines, and how that division changes human readiness and skill. In modern AI agents, many of these cognitive operations are explicit and programmable, while interaction trajectories can be retained and transformed into memories, workflows, skills, and policies that shape later behavior. Read together, these literatures expose a causal sequence: organizing cognitive work determines the distribution of practice, supervision, feedback, and trajectories; that experience changes human skill and beliefs, machine memory and policy, and the coordination between them; and the updated joint system then organizes later work differently. We argue that Human–AI collaboration should therefore be studied as a process of joint cognitive development. The design problem is to organize current cognition while accounting for the Human–AI system that the current choices are shaping. The way humans and AI work together today changes what each can do tomorrow.

Article activity feed

  1. PREreview

    This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/22784989.

    Summary

    The paper presents Human–AI collaboration as "joint cognitive development," arguing that the way tasks are shared between humans and AI can influence what both learn and how their capabilities develop over time. It brings together work on bounded rationality, Human–Automation/HCI, and AI-agent memory to propose a framework linking task allocation, experience, learning, and future collaboration. The paper concludes with six directions for future research.

    Major Issues

    The main issue is that the "joint cognitive development loop" is presented somewhat more confidently than the available evidence supports. The paper itself acknowledges that the complete loop has not yet been empirically demonstrated. It would therefore be helpful to distinguish more clearly between what is supported by existing research and what is being proposed as a conceptual framework.

    Some of the six research questions also overlap with gaps already discussed in the continual-learning and agent-memory literature. The paper could strengthen its contribution by showing more explicitly how these questions emerge from the proposed framework itself.

    Minor Issues

    No significant minor issues at this stage.

    Competing interests

    The author declares that they have no competing interests.

    Use of Artificial Intelligence (AI)

    The author declares that they did not use generative AI to come up with new ideas for their review.

  2. PREreview

    This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/22774660.

    This preprint proposes a conceptual framework for Human–AI collaboration as joint cognitive development. The framework draws on three main research traditions: human cognition, including issues of bounded rationality and metareasoning; Human–Automation and HCI studies of function allocation and modes of supervisory control; and modern AI systems, including agent memory and continual learning.

    The way cognitive work is distributed between humans and AI in the current task has effects that extend to the distribution of practice, feedback, supervision, and experience. The way these experiences affect human skills and beliefs, as well as the memory and policies of the AI, and how the two systems coordinate their actions in the future, forms a developmental feedback loop.

    Six research questions concerning joint learning, the validity of stored experience, supervision, and the longitudinal evaluation of Human–AI systems are proposed by the paper.

    This is mainly a conceptual and integrative contribution. The paper brings together a number of different research areas and highlights an important issue for the long-term development of Human–AI systems: how the organization of cognitive work in present tasks can shape future Human–AI capabilities.

    The paper provides a useful integration of hitherto rather separate streams of research and therefore highlights an important issue for the long-term development of Human–AI systems, i.e., the organization of cognitive work in present tasks. However, the developmental loop presented is still entirely theoretical and has not yet been empirically tested.

    Major issues

    • The paper is presented as a review, but actually it is closer to a position paper written from the perspective of a conceptual framework and research agenda. That is, a "joint cognitive development" framework is proposed and outlined in some detail, but no attempt is made to empirically test or study the proposed developmental loop of cognitive work.

    • The argument relies substantially on non-peer-reviewed sources, including arXiv preprints and engineering reports or blog posts from organizations such as Anthropic, OpenAI, and DeepMind. These sources may contain useful information on current practices with AI systems, but the paper should be particularly cautious when citing reports by industries or other organizations for claims about learning, adaptation, or developmental mechanisms, as these claims typically require stronger scholarly evidence.

    • The major causal claim of the paper is not adequately supported by the evidence presented. It is generally not made clear in the paper whether evidence is provided to support individual components of the framework for the organization of cognitive work, or whether evidence is provided to support the complete developmental loop of the proposed causal relationship. It would be helpful to distinguish more clearly between evidence that supports the individual components of the framework and evidence that supports the causal relationships between them.

    • The manuscript combines old and new material, at times assigning similar weight to a classical reference and to a very recent preprint or other emerging work. Explicit treatment of the associated evidence and certainty would improve the scholarly rigor of this synthesis. The evidence and certainty of the sources used for the synthesis could be explained more clearly.

    • Six very interesting research questions are proposed to test the framework in more detail. Some of the research questions are very ambitious and require a very complex experimental setup to test the framework. The paper would benefit from considering whether initial or pilot studies could be used to evaluate the feasibility of these research questions before attempting more complex experimental designs.

    Minor issues

    1. Many of the sections, including the introduction to the literature and connections to the modern AI concept, follow a similar structure. Therefore, there are several opportunities to differentiate the sections in order to improve readability and avoid excessive repetition.

    2. The conceptual Figure 1 is relatively abstract and does not provide a lot of detail. A concrete worked example (e.g., the trace of a Human–AI interaction) for readers to understand the proposed developmental loop better would be helpful.

    3. There is no clearly separated section for the limitations of the manuscript under review. Although limitations are acknowledged in various parts of the manuscript, compiling a list of the main limitations in a separate section would help to delimit the scope of the proposed framework.

    4. The distinction between the two concepts, Human–AI co-evolution and joint cognitive development, could be made more clearly from other related concepts (co-adaptation, mutual adaptation, human-in-the-loop learning), and it would be good to clarify what is new about the proposed framework.

    5. Some of the introductory sentences are long and comprise several conceptual statements. These could be reworked into shorter sentences without changing the content.

    6. The statement on the use of AI-assisted technologies would benefit from more detail on the extent to which the author actually relied on AI to develop key arguments in the manuscript, as opposed to using search, organization, and language support functions. This comment is especially relevant given the manuscript's discussion of AI-assisted knowledge production and judgment.

    Competing interests

    The author declares that they have no competing interests.

    Use of Artificial Intelligence (AI)

    The author declares that they did not use generative AI to come up with new ideas for their review.

  3. Version published to 10.20944/preprints202609.1092.v1