CAC Co-intelligence:
Learning with AI, Not from it
Theory
The Theory Behind CAC Co-intelligence
The concept of co-intelligence is central to Ethan Mollick's book Co-Intelligence: Living and Working with AI, which explores how humans and artificial intelligence can work together. The principle is simple but powerful: by combining human and machine capabilities, we can achieve outcomes that may be difficult for either to accomplish alone.
CAC Co-intelligence takes this concept into the classroom. Rather than focusing primarily on how AI can help us work more efficiently, it asks a different question:
How can we use AI to enhance learning without sacrificing the thinking required to learn?
The goal is to harness AI's capabilities while maintaining human cognitive engagement, independent judgment, and responsibility for learning.
Why This Matters for Learning
Learning requires cognitive engagement. Developing knowledge, understanding, and expertise involves actively processing information, solving problems, making connections, practising skills, and reflecting on our thinking.
Generative AI introduces a new challenge. Students can now produce sophisticated answers and complete complex tasks with relatively little independent cognitive effort.
This creates a fundamental distinction between completing a task and learning from it.
Consider physical fitness. Watching football can help you understand the game, but watching others exercise will not develop your own physical fitness. You need to participate.
The same principle applies to learning. AI can demonstrate, explain, guide, and support, but if students consistently outsource the cognitive work they need to practise, opportunities for developing their own capabilities may be lost.
Cognitive offloading is not inherently harmful. We regularly use tools to reduce unnecessary mental effort. The challenge is knowing which cognitive activities can be delegated to AI and which need to remain with the learner.
The Three Forms of Co-intelligence
CAC Co-intelligence distinguishes between three basic patterns of human–AI interaction:
1. Cognitive → AI
Human thinking initiates the task, followed by AI assistance or execution. This can be useful for efficiency and productivity, but risks leaving AI-generated outcomes without sufficient human evaluation.
2. AI → Cognitive
AI provides the starting point, with humans subsequently interpreting, evaluating, or developing its outputs. This approach can support exploration and idea generation, but may also influence independent thinking through anchoring or premature reliance on AI suggestions.
3. Cognitive → AI → Cognitive (CAC Co-intelligence)
Human thinking initiates the process, AI is used to support or extend that thinking, and human cognition returns to critically evaluate, refine, and take responsibility for the outcome.
CAC is the preferred approach when the educational goal is to develop students' thinking and learning capabilities, rather than simply produce an efficient outcome.
Importantly, cognition remains active throughout the process, including during interaction with AI.
The Theoretical Foundation
CAC Co-intelligence draws on established principles of educational psychology, particularly self-regulated learning, metacognition, cognitive engagement, and cognitive offloading. These concepts help explain why students need to remain actively involved in their thinking before, during, and after using AI.
1. Self-Regulated Learning: Planning, Monitoring, and Reflecting
One of the strongest theoretical connections is Barry Zimmerman's cyclical model of self-regulated learning. Zimmerman describes effective learning as an ongoing cycle involving three interconnected phases:
-
Forethought: Learners analyse the task, activate prior knowledge, set goals, and plan their approach.
-
Performance: Learners apply strategies, monitor their progress, and adjust their actions as they work.
-
Self-reflection: Learners evaluate their performance, consider what they have learned, and identify improvements for future learning.
CAC Co-intelligence applies a similar cyclical approach to learning with AI.
-
Cognitive (Before AI): Students establish their own understanding, intentions, and initial approaches before seeking AI assistance.
-
AI (During Engagement): Students interact purposefully with AI, actively questioning, monitoring, and directing the interaction.
-
Cognitive (After AI): Students evaluate AI contributions, reflect on their learning, and make informed decisions about how to proceed.
The connection is not simply that both approaches have three stages. More importantly, both emphasise the importance of learners taking responsibility for directing, monitoring, and evaluating their learning.
CAC translates this principle into a practical process for AI-supported learning, encouraging students to remain active participants rather than passive recipients of AI-generated responses.
2. Metacognition: Thinking About Our Thinking
Metacognition refers to our ability to understand, monitor, and regulate our own thinking. It plays a central role in effective learning because students need to recognise what they know, identify what they do not understand, and decide which strategies will help them progress.
Generative AI makes metacognition particularly important. AI can produce convincing explanations, solutions, and ideas, even when the user has not fully understood the underlying concepts. This creates a risk of students confusing a high-quality AI-generated response with their own understanding.
CAC encourages metacognitive awareness through questions embedded in each stage:
-
Before AI: What do I already know? What am I trying to learn? Can I make an initial attempt myself?
-
During AI: Does this response make sense? What assumptions is AI making? Am I understanding or simply accepting its suggestions?
-
After AI: What have I actually learned? Can I explain or apply this independently? What should I question, verify, or improve?
These questions help make thinking more deliberate and visible.
The aim is to develop students who can recognise when AI genuinely supports their understanding and when it may be concealing gaps in their knowledge.
3. Cognitive Engagement: Learning Requires Active Thinking
CAC is also consistent with research on the importance of cognitive engagement in learning. For example, Chi and Wylie's ICAP framework (2014) identifies four modes of engagement: Passive, Active, Constructive, and Interactive. The framework proposes that learning generally becomes more effective as students move beyond simply receiving information towards actively processing it, generating new understanding, and engaging in meaningful dialogue.
This distinction is especially relevant to generative AI. A student who asks AI to complete an assignment and copies the response may produce an acceptable outcome while engaging in relatively little meaningful cognitive activity. In contrast, a student who develops an initial response, uses AI to challenge their reasoning, compares alternatives, and justifies their final decisions engages in a richer thinking process.
Importantly, using AI interactively does not automatically mean the student is cognitively engaged. A lengthy conversation with a chatbot can still involve limited learning if the student simply accepts what is generated. CAC encourages students to move beyond receiving AI outputs towards questioning, explaining, evaluating, and constructing their own understanding.
4. Cognitive Offloading: Knowing What to Delegate
Cognitive offloading occurs when we use external tools or resources to reduce the mental effort required to complete a task. For example, Risko and Gilbert (2016) explain that cognitive offloading is a common feature of human thinking. It is not inherently negative. Calculators, reminders, search engines, and other technologies allow us to redirect cognitive resources towards different activities.
Generative AI significantly expands what can be offloaded, including writing, analysing, reasoning, summarising, and generating ideas. The educational concern is not cognitive offloading itself, but offloading the very thinking that students need to practise in order to learn. For example, using AI to explain a difficult concept may support understanding. However, repeatedly asking AI to solve problems without first attempting them may reduce opportunities to develop independent problem-solving capabilities.
CAC encourages students to make deliberate decisions about cognitive offloading:
-
What thinking should I attempt independently?
-
Which aspects of the task would benefit from AI augmentation?
-
What must I understand, evaluate, and take responsibility for myself?
By establishing cognitive engagement before and after AI use, CAC aims to reduce uncritical dependence while retaining the benefits of AI assistance.
Bringing the Theory Together
These four concepts provide a complementary theoretical rationale for CAC Co-intelligence:
-
Self-regulated learning explains the importance of planning, monitoring, and reflecting.
-
Metacognition explains why students need to remain aware of and regulate their own understanding.
-
Cognitive engagement explains why meaningful learning requires students to actively process and construct knowledge.
-
Cognitive offloading explains the need to make deliberate decisions about which thinking activities to delegate to AI.
Together, they support the central principle of CAC:
AI should augment human cognition without unnecessarily replacing the cognitive processes required for learning.
CAC is not a replacement for these established theories. It is a practical framework informed by them, designed to help students develop purposeful and reflective habits when learning with AI.
The three-stage structure provides a simple way to reinforce these principles consistently, while its iterative nature encourages ongoing questioning, evaluation, and improvement.
The Four Core Risks of AI-Supported Learning
The educational theories underpinning CAC Co-intelligence highlight the importance of active cognitive engagement, self-regulation, metacognition, and critical reflection. They also help explain what can go wrong when these processes are neglected during AI-supported learning.
Generative AI can enhance learning by providing explanations, feedback, alternative perspectives, and opportunities to explore complex ideas. However, its ability to perform increasingly sophisticated cognitive tasks introduces risks when students become passive recipients of AI-generated information rather than active participants in their learning.
CAC Co-intelligence is designed to address four interconnected risks:
-
Cognitive Offloading: When AI performs the thinking students need to practise.
-
Automation Bias: When students place excessive trust in AI-generated information and judgments.
-
Skill Erosion: When repeated reliance on AI reduces opportunities to develop and maintain independent capabilities.
-
Illusion of Learning: When students mistake AI-assisted performance for genuine personal understanding.
These risks highlight a fundamental distinction between what students can produce with AI and what they genuinely understand, learn, and can accomplish independently.
The purpose of CAC is not simply to minimise these risks, but to help students use AI in ways that strengthen their thinking and learning.
These four risks focus specifically on learning and cognitive development. Ethical, social, economic, and environmental considerations remain essential to responsible AI use and are addressed through the broader educational application of CAC Co-intelligence.
1. Cognitive Offloading: When AI Does the Thinking
Cognitive offloading occurs when we delegate mental activities to external tools or resources, including AI. As discussed in the theoretical foundation, cognitive offloading is not inherently harmful. It can reduce unnecessary mental effort and allow us to focus on more complex activities.
The educational concern arises when students delegate the very thinking they need to practise in order to learn. For example, a student who asks AI to explain a difficult mathematical problem may develop greater understanding. However, a student who asks AI to solve the problem without first attempting it or engaging meaningfully with the explanation may complete the task without developing the intended problem-solving capabilities.
This creates a distinction between task completion and actual learning. Students may produce sophisticated work without necessarily understanding how it was created or being able to demonstrate the underlying knowledge independently.
The risk: Students may increasingly rely on AI to perform learning-critical cognitive activities, reducing opportunities to develop their own understanding and capabilities.
How CAC responds:
The initial cognitive stage encourages students to activate prior knowledge, formulate ideas, and attempt their own approaches before seeking AI assistance.
During AI engagement, students can explore alternative approaches, receive explanations, and challenge their initial thinking.
The final cognitive stage encourages students to compare and synthesise their original ideas with useful AI contributions, developing a more informed understanding rather than simply adopting an AI-generated solution.
The goal is not to eliminate cognitive offloading, but to ensure that AI supports learning rather than bypassing it.
2. Automation Bias: When AI Becomes the Authority
Automation bias occurs when people place excessive trust in automated systems and accept their outputs or recommendations without sufficient independent evaluation. Generative AI makes this particularly challenging because its responses are often fluent, detailed, and convincing, even when they contain factual errors, unsupported claims, misleading information, or flawed reasoning.
For example, a student researching a complex issue may accept an AI-generated explanation because it appears credible and authoritative, without checking its accuracy, questioning its assumptions, or considering alternative perspectives. This becomes particularly problematic when students lack the subject knowledge needed to recognise errors or weaknesses in AI-generated responses.
AI can also influence thinking through anchoring, where an initial AI-generated suggestion shapes subsequent judgments and makes alternative interpretations less likely to be considered.
The risk: Students may increasingly substitute AI-generated judgments for their own critical evaluation, accepting information and conclusions without sufficient scrutiny.
How CAC responds:
The initial cognitive stage encourages students to establish their own understanding, questions, and preliminary judgments before being influenced by AI-generated suggestions.
During AI engagement, students are encouraged to question responses, challenge assumptions, request justification, and explore alternatives.
The final cognitive stage requires students to critically compare AI contributions with their initial thinking, verify important information, and decide which ideas should be retained, revised, or rejected.
This process recognises that neither the student's initial understanding nor AI-generated responses should be accepted uncritically.
The aim is to ensure that AI informs human judgment rather than replacing it.
3. Skill Erosion: When Independent Capabilities Are Neglected
Cognitive skills develop and are maintained through practice, feedback, application, and reflection. When AI repeatedly performs activities that students would otherwise need to practise, opportunities to develop and maintain those skills may diminish.
For example, a student who consistently relies on AI to construct arguments may become increasingly efficient at producing written assignments without necessarily developing the ability to organise, evaluate, and defend arguments independently. Similarly, students who routinely allow AI to solve unfamiliar problems may miss opportunities to develop persistence, strategic thinking, and confidence in their own problem-solving abilities. Over time, this can create a gap between the sophistication of students' AI-assisted work and their independent cognitive capabilities.
Importantly, skill erosion is not an inevitable consequence of AI use. Appropriately designed AI-supported learning can strengthen skills through targeted feedback, guided practice, explanation, and opportunities for independent application.
The risk: Repeated reliance on AI may contribute to weaker skill development or, in some circumstances, the deterioration of previously acquired capabilities.
How CAC responds:
The initial cognitive stage preserves opportunities for independent thinking, problem-solving, and the application of existing knowledge.
AI is then used to challenge, extend, or refine students' thinking, providing opportunities to explore approaches they may not have considered independently.
In the final cognitive stage, students synthesise these contributions with their original ideas, construct improved explanations or solutions, and reflect on how their understanding has developed.
This process should be supported by opportunities to demonstrate knowledge and skills independently, ensuring that AI-supported performance translates into genuine learning.
The intention is to help students develop their own capabilities alongside their ability to use AI effectively.
4. Illusion of Learning: When Confidence Exceeds Understanding
Generative AI can make complex ideas appear easier to understand by providing clear explanations, structured solutions, and convincing responses. However, recognising or following an explanation does not necessarily mean that a student has developed the knowledge or skills required to explain, recall, or apply that understanding independently. This can create an illusion of learning, where students overestimate their understanding because AI has made the task appear manageable.
For example, a student may confidently follow an AI-generated solution to a complex problem, yet struggle to solve a similar problem without assistance. The student may feel that learning has occurred because the explanation makes sense, even though the underlying knowledge has not been sufficiently developed or consolidated. This represents a metacognitive challenge: students may struggle to distinguish between what they genuinely understand and what AI has helped them accomplish. The concern is particularly significant for novice learners, who may lack the knowledge or experience needed to identify gaps in their understanding.
The risk: Students may develop confidence that exceeds their actual understanding, leaving important knowledge gaps unnoticed and potentially reinforcing dependence on AI.
How CAC responds:
The initial cognitive stage provides an opportunity for students to recognise their existing knowledge, uncertainties, and independent capabilities.
During AI engagement, students can explore explanations, test ideas, and seek clarification.
The final cognitive stage encourages students to critically synthesise what they initially understood with what they have gained from AI, while checking whether their understanding has genuinely improved. This may involve explaining concepts without AI assistance, applying knowledge to unfamiliar situations, justifying decisions, or identifying remaining uncertainties.
Importantly, producing a refined answer with AI is not sufficient evidence of learning. Students also need opportunities to demonstrate what they can understand and accomplish independently.
The aim is to help students distinguish between AI-assisted performance and genuine independent understanding.
How the Four Risks Connect
These four risks are distinct but interconnected, and they can reinforce one another. Excessive cognitive offloading may reduce opportunities for independent thinking. Automation bias may discourage students from questioning AI-generated responses. Reduced opportunities for meaningful practice may contribute to weaker skill development or skill erosion, while the apparent quality of AI-assisted work may create an illusion of learning.
Together, these risks can contribute to a reinforcing cycle in which students become increasingly reliant on AI without necessarily developing the understanding and capabilities needed to work independently.
CAC Co-intelligence aims to interrupt this potential cycle by establishing two deliberate cognitive checkpoints:
-
Before AI: Establish independent thinking, activate prior knowledge, develop initial ideas, and clarify the purpose of AI assistance.
-
After AI: Critically evaluate AI contributions, synthesise them with initial thinking, verify information, reflect on learning, and construct a more informed understanding.
Throughout the AI interaction, students are expected to remain cognitively engaged by questioning, challenging, exploring, and directing the process.
The Importance of Cognitive Synthesis
The final cognitive stage is particularly important because it involves more than checking, editing, or accepting AI-generated outputs. Students return to their original thinking and critically compare it with the ideas, explanations, and perspectives encountered through AI. Through this process, they determine which ideas to retain, revise, combine, or reject, using the interaction to construct a more informed understanding.
For example, a student may begin with an initial explanation, use AI to challenge its assumptions or introduce alternative perspectives, and then develop a revised explanation that is more comprehensive and better supported. This does not mean that human and AI contributions should always be combined equally. Sometimes the student's initial reasoning may be stronger than the AI response. At other times, AI may reveal important misconceptions or introduce valuable perspectives that fundamentally change the student's understanding.
The educational value lies in the student's ability to critically evaluate and synthesise these contributions, rather than simply adopting whichever answer appears most convincing. Importantly, synthesis must involve genuine understanding. Combining ideas into a polished final product is not, by itself, evidence that learning has occurred. Students should be able to explain and justify their synthesis, ideally without relying on AI to do so. This makes the final cognitive stage both a safeguard against uncritical AI dependence and an opportunity for deeper learning.
Beyond Risk Reduction: Strengthening Human Cognition
CAC Co-intelligence does not automatically prevent cognitive offloading, automation bias, skill erosion, or illusions of learning. Its effectiveness depends on how teachers design learning activities, scaffold student thinking, and provide opportunities for independent practice, critical evaluation, and reflection. Nor does every interaction with AI require the same degree of cognitive effort. The appropriate balance between human and AI contributions depends on the task, the learner's existing capabilities, and the intended learning outcomes.
CAC deliberately prioritises the development of human cognition when learning is the primary goal. Its purpose extends beyond protecting students from the potential negative consequences of AI. By encouraging independent thinking, purposeful AI engagement, and critical synthesis, CAC seeks to make human–AI interaction an opportunity for cognitive development.
The objective is not simply to protect thinking from AI, but to use AI interaction as an opportunity to develop better thinking.
​
Recognising and Evaluating the Four Risks in Practice
While CAC Co-intelligence provides a structure intended to encourage meaningful cognitive engagement, educators also need to consider how they can recognise whether the four identified risks are occurring and whether students are developing independent capabilities.
The following questions can guide teaching, observation and evaluation:
-
Cognitive Offloading: Are students delegating cognitive activities they need to practise to achieve the intended learning outcomes?
-
Automation Bias: Can students independently identify and challenge inaccurate, unsupported or inappropriate AI-generated information?
-
Skill Erosion: Are students maintaining and developing their ability to perform relevant tasks independently over time?
-
Illusion of Learning: Does students' confidence in their understanding align with their demonstrated ability to explain, recall and apply knowledge without AI assistance?
These questions provide practical prompts for evaluating AI-supported learning, rather than validated measures of the four risks. Educators can investigate them through appropriate evidence, including independent explanations, unfamiliar problems, delayed assessments and observations of students' reasoning.
Importantly, the presence of the CAC stages should not be treated as evidence that these risks have been eliminated. Their reduction, and any improvement in learning, must be established through evaluation rather than assumed from participation in the process.
The Two Layers of CAC Co-intelligence
CAC Co-intelligence operates through two interconnected layers: the Student Layer and the Teacher Layer.
The Student Layer provides a practical framework for thinking and learning with AI. The Teacher Layer supports educators in designing learning experiences that develop students' cognitive capabilities, AI literacy, and capacity to use AI critically, ethically, and purposefully.
Together, they aim to ensure that AI use does more than improve task completion: it contributes to the development of capable, critical, independent, and responsible learners.
Student Layer
Students need to be aware of and follow the 3 stages: cognitive - AI - Cognitive, or Thinking - AI - Thinking.
Students are encouraged to engage their own thinking before, during, and after interacting with AI.
This process positions AI as a tool for cognitive augmentation rather than a substitute for human thinking. Its purpose is to preserve student agency, promote critical engagement, and strengthen learning through thoughtful human–AI interaction.
Teacher Layer
Introducing students to the three-stage process is necessary, but not sufficient. Effective CAC Co-intelligence requires teachers to intentionally design and facilitate learning experiences that develop students' thinking, deepen their understanding of AI, and support responsible decision-making.
The Teacher Layer focuses on four interconnected dimensions:
1. Guiding Thinking and Learning: Design and scaffold learning experiences that develop students' cognitive and metacognitive capabilities, helping them understand when, why, and how to engage with AI without surrendering responsibility for their thinking.
2. Understanding AI Augmentation Roles: Help students recognise the different roles AI can play in supporting, extending, and challenging human thinking. Develop their ability to select appropriate roles, evaluate their strengths and limitations, and recognise when AI assistance may undermine learning.
3. Exploring Ethical and Societal Implications: Develop students' awareness of the social, economic, and environmental implications of AI, including issues of bias, privacy, equity, sustainability, and accountability. Encourage informed, responsible, and context-sensitive decisions about AI use.
4. Developing AI Literacy: Build students' understanding of different forms of AI, how they function, what they can and cannot reliably do, and how these differences influence their appropriate use in learning.
Applying the CAC Cycle
These principles are put into practice through the three stages of the CAC cycle: Cognitive → AI → Cognitive. Although the framework is illustrated as a linear sequence for simplicity, it is inherently cyclical and iterative. The final cognitive stage is not necessarily the end of the process. Critical evaluation, synthesis, and reflection can generate new questions, reveal gaps in understanding, or identify opportunities for further exploration, initiating another cycle of thinking and AI engagement. Each iteration should build on the previous one, progressively deepening understanding, refining ideas, and strengthening human judgment. The following section explains how each stage can be applied in practice.
​
Step 1: Cognitive
​
In this step, student guidance includes:
-
Think first: Begin with deep reflection. Understand the problem, its context, and what truly needs solving before turning to AI. Develop pilot solutions so that human approaches are considered before being influenced by AI.
-
Plan & Prepare: Define the problem clearly, explore traditional or human-driven approaches first, and decide whether AI is genuinely the right tool for this task.
-
Reflect on proven methods & ethical implications: Draw on policy, existing knowledge and professional standards; anticipate potential ethical, security, or practical consequences of AI involvement. For example, when moving to step 2, is it ethical to engage the GenAI with the associated data? Discuss how learning occurs through this approach.
-
How can AI enhance your work?: Ask why you may wish to use AI. Is it necessary? Will it enhance the outcome, and which AI type or model best aligns with your goal?
Step 2: AI
​
In this step, student guidance includes:
-
Apply as a supportive tool & not as your replacement: Treat AI as a collaborator that enhances capability while preserving human direction and accountability. Use ethically. Discuss how learning occurs in this approach and contrast it with Step 1.
-
Which AI is the best fit? Type/Model: Select the right kind of AI (e.g., generative, predictive, analytical) to match your specific goal or problem.
-
AI outputs can be flawed: Remember that AI can generate errors or biases; critical review is essential before use. Discuss matters such as relevance and accuracy. Discuss differences in prompts and models.
​
Step 3: Cognitive
​
In this step, student guidance includes:
-
Apply Evaluative Judgment: Assess AI results critically for accuracy, logic, and alignment with your intent and values. Discuss how learning and higher-order thinking occur through such synthesis.
-
Refine and validate AI outputs: Edit, verify, and contextualise results to ensure they meet human and professional standards. Discuss how this can be done.
-
Apply human touches, responsibility, and control: Integrate human voice, empathy, ethics, and ownership; maintaining human oversight over final outcomes.
​
Step 4 for Classroom activities
When complete, engage the class in reflection, discussion and connection to assessment or other activities.
Extending CAC Through Learning Design
The CAC cycle provides students with a practical process for maintaining cognitive engagement before, during, and after using AI. However, simply following the three stages is not enough to realise the full educational potential of human–AI collaboration.
Teachers and facilitators need to intentionally design learning experiences that extend students' thinking, develop their understanding of AI, and strengthen their capacity to use it critically, ethically, and purposefully. This involves considering not only when students use AI, but also how they interact with it, what role AI performs, and which cognitive capabilities the activity is intended to develop.
Two important considerations are the different ways AI can augment human cognition and the ethical implications associated with its use.
1. Exploring Different AI Augmentation Approaches
AI does not need to perform the same function in every learning activity. Depending on the intended learning outcomes, it can serve as a tutor, questioner, challenger, creative partner, or component of a broader learning environment. Students can also learn to make deliberate decisions about how responsibilities should be divided between themselves and AI.
Each approach creates different opportunities for cognitive engagement. For example, using AI as a Socratic questioner encourages students to explain and defend their reasoning, while using AI as an adversarial opponent can challenge assumptions and expose weaknesses in an argument. Similarly, using AI as a tutor may support skill acquisition through guided practice and feedback, while a Centaur approach encourages students to decide which tasks should remain human responsibilities and which may benefit from AI assistance.
The role of AI should be selected according to the intended learning outcome, rather than simply what the technology is capable of doing.
AI Augmentation Roles
These approaches are not mutually exclusive. They describe different interaction strategies, divisions of responsibility, and technological arrangements rather than six separate types of AI. For example, an AI tutor may use Socratic questioning, while a co-creation activity may incorporate adversarial feedback. An AI-enabled learning environment may support several of these approaches.
Importantly, CAC Co-intelligence is not another augmentation approach within this table. It provides the overarching cognitive process through which students can engage with any of these approaches. Regardless of the AI role selected, students begin with purposeful thinking, remain cognitively engaged during AI interaction, and return to human evaluation, synthesis, and reflection.
2. Developing Ethical Awareness Through CAC
Effective AI-supported learning also requires students to understand that AI use has consequences beyond the immediate learning task. These include social, economic, and environmental implications. AI has far reaching ethical considerations beyond most other technologies due to its rapid uptake and evolution. Therefore, this awareness is vitally important.
Rather than treating AI ethics as a separate topic addressed only once, teachers should deliberately introduce and revisit different ethical considerations through successive CAC learning experiences. For example, an activity involving AI-generated images could introduce questions about copyright, originality, and creative ownership. An activity involving AI-generated recommendations might focus on bias, fairness, and transparency. An activity involving AI-supported research could explore misinformation, verification, and accountability. Other activities could investigate the environmental resources required to operate AI systems, the potential economic consequences of automation, or issues of equity and access.
The intention is not to address every ethical issue in every activity. Instead, teachers should select considerations relevant to the task, the AI systems being used, and students' developing understanding.
Integrating Ethics into the CAC Cycle
-
Cognitive — Before AI: Students consider whether AI use is appropriate, what information may be shared, who could be affected, and which ethical responsibilities or policies apply.
-
AI — During Engagement: Students monitor how AI handles information, represents perspectives, generates content, or makes recommendations. They remain alert to potential bias, misleading outputs, privacy concerns, and other emerging issues.
-
Cognitive — After AI: Students critically evaluate the ethical implications of AI contributions, consider possible consequences, and take responsibility for how the resulting work or decisions are used.
Through repeated application, students can develop the capacity to recognise and respond to ethical concerns independently, rather than relying solely on teacher direction or prescribed rules.
3. Progressive Learning Through Different CAC Experiences
Developing co-intelligence should not depend on a single lesson or isolated AI activity. Instead, teachers can deliberately vary AI augmentation approaches, cognitive demands, and ethical considerations across successive learning experiences.
It is important to distinguish this progression from iteration within an individual CAC activity. Iteration within a CAC activity occurs when students return to earlier stages of the cycle to refine their understanding, investigate new questions, or improve their reasoning. Progression across CAC learning experiences occurs when teachers design different activities over time to introduce new AI augmentation approaches, deepen ethical understanding, and develop increasingly sophisticated cognitive capabilities. Both processes are important, but they serve different purposes.
Scaffolding Progressively Greater Independence
Initially, students may need structured activities in which teachers explicitly model the CAC process, specify AI's role, and guide critical reflection. As students gain experience, teachers can introduce more complex interactions. These might include using AI to challenge assumptions, explore competing perspectives, or support open-ended inquiry. Over time, students should have greater responsibility for selecting appropriate AI augmentation approaches, recognising ethical considerations, and regulating their own learning.
For example, a sequence of learning experiences might involve:
-
AI Tutor: Students receive guided explanations and feedback while strengthening their understanding of a topic.
-
Socratic AI: Students explain their reasoning in response to probing questions.
-
Adversarial AI: Students defend and refine their arguments against alternative perspectives.
-
Co-creation: Students critically combine their own ideas with AI contributions to develop innovative solutions.
-
Centaur: Students make informed decisions about the division of tasks and responsibilities between themselves and AI.
Ethical considerations can be progressively integrated into these experiences, with students moving from recognising immediate issues such as privacy and academic integrity towards evaluating more complex questions of fairness, accountability, social impact, and sustainability.
This example is not a prescribed sequence. Teachers should select and revisit approaches according to the learning objectives, context, and students' demonstrated readiness. Throughout this progression, the underlying CAC process remains consistent. What develops is the sophistication of students' thinking, their understanding of AI's capabilities and limitations, and their ability to exercise independent judgment.
4. The Teacher's Role in Learning Design
The effectiveness of these experiences depends on purposeful learning design and facilitation. Teachers need to consider:
-
Learning intentions: Which knowledge, skills, and cognitive capabilities should students develop?
-
AI augmentation approaches: Which form of human–AI interaction best supports the intended learning outcomes?
-
Ethical implications: Which ethical, social, economic, or environmental considerations should students investigate?
-
Cognitive scaffolding: What guidance is needed to support meaningful engagement throughout the CAC cycle?
-
Evidence of learning: How will students demonstrate independent understanding, critical evaluation, and the ability to synthesise their thinking with AI contributions?
These decisions are interconnected. The selection of an AI augmentation approach influences the thinking students are encouraged to undertake, while the learning context determines which ethical questions and cognitive skills are most relevant.
Teachers should also consider whether AI assistance is appropriate for a particular activity or whether independent practice would better support the intended learning. Importantly, more sophisticated AI use does not automatically produce deeper learning.
An activity involving a simple AI questioner may develop stronger reasoning than one involving a complex AI-enabled environment if the latter allows students to disengage from the cognitive work. The quality of learning depends on the thinking students undertake, not the sophistication of the AI tools they use.
Developing Capable and Responsible AI Learners
CAC provides a consistent cognitive process, while thoughtful learning design introduces the variety, challenge, and progression needed to develop students' capabilities over time. By exposing students to different AI augmentation approaches and progressively exploring ethical implications, teachers can help them move beyond simply knowing how to use AI towards understanding when, why, and how it should be used.
The longer-term goal is for students to become increasingly independent in selecting appropriate AI approaches, critically evaluating their contributions, recognising ethical implications, and taking responsibility for their learning and decisions.
CAC provides the thinking process. Learning design develops the capability to apply it intelligently, critically, and responsibly across different contexts.
Limitation of CAC Co-intelligence
CAC Co-intelligence provides a practical framework for maintaining human cognitive engagement when learning with AI. It draws on established principles of self-regulated learning, metacognition, cognitive engagement, and cognitive offloading, while addressing risks such as automation bias, skill erosion, and the illusion of learning.
However, CAC is not a universal solution for AI integration in education. Its effectiveness depends on the learning context, students' existing capabilities, the quality of teaching and learning design, and how meaningfully the three stages are applied.
Understanding these limitations is essential to using the framework appropriately.
1. Learning Versus Efficiency: When AI-First May Be Appropriate
CAC deliberately prioritises cognitive engagement and learning over efficiency and productivity. Requiring students to think independently before engaging with AI can provide valuable opportunities to activate prior knowledge, develop ideas, and establish an initial perspective. However, this approach may take longer than allowing AI to generate ideas, explanations, or solutions immediately.
In some situations, particularly brainstorming, initial exploration, or unfamiliar subject areas, AI-first approaches may offer educational advantages. They can expose students to possibilities, explanations, or perspectives that they may not have generated independently. Furthermore, students with limited prior knowledge may struggle to develop meaningful initial responses without additional guidance. Requiring extensive independent thinking before providing assistance could introduce unnecessary frustration or unproductive cognitive demands. The limitation arises when CAC is applied as a rigid sequence rather than a flexible thinking process. Requiring extensive independent work before AI engagement may introduce unnecessary cognitive demands, limit productive exploration, or discourage students who lack sufficient prior knowledge.
CAC therefore requires teachers and students to exercise judgment about the appropriate depth of engagement at each stage.
The first cognitive stage should therefore be adapted to the learner and the task. It may involve identifying existing knowledge, formulating questions, predicting possible outcomes, or recognising what the student does not yet understand, rather than requiring a complete independent solution. In most scenarios, the learning design of the first stage would replicate current practise.
CAC does not suggest that AI-first approaches are inherently inappropriate. Rather, it provides a preferred structure when developing independent thinking and cognitive capabilities is a central learning objective. The key consideration is whether the chosen approach supports the intended learning, not simply whether it follows a prescribed sequence.
2. The Risk of Superficial Compliance
One of CAC's most important limitations is that following the three stages does not automatically ensure meaningful learning. Students could make a minimal initial attempt, use AI to complete most of the task, and then provide a superficial reflection. Although they would appear to have followed the CAC process, the intended cognitive benefits may not have occurred.
Similarly, students may critically review AI outputs without developing a deeper understanding, or combine their initial ideas with AI suggestions without genuinely evaluating the contributions. This is particularly important in the final cognitive stage, where synthesis should represent the development of understanding, not merely the production of a more polished answer.
The value of CAC therefore depends on the quality of cognitive engagement within each stage, rather than simply completing the sequence. Teachers need to provide appropriate scaffolding, ask questions that encourage deeper reasoning, and create opportunities for students to explain and justify their decisions. Where appropriate, students should also demonstrate their understanding independently, rather than relying solely on AI-supported work as evidence of learning.
CAC provides a structure for thinking, but the structure alone cannot guarantee the quality of that thinking.
3. Human-First Thinking Can Also Introduce Bias
Although CAC begins with independent cognition to reduce premature reliance on AI, human-first thinking is not without limitations. Students enter learning activities with existing assumptions, misconceptions, and incomplete knowledge. Developing an initial position can sometimes strengthen attachment to those ideas, particularly when students become reluctant to revise their reasoning.
This creates a potential tension within CAC. While the initial cognitive stage can help reduce anchoring to AI-generated suggestions, it may also introduce anchoring to the student's own initial ideas. Consequently, the final cognitive stage must involve more than defending or refining an original position.
Students need to remain willing to reconsider their assumptions, recognise mistakes, and substantially change their understanding when supported by credible evidence or stronger reasoning. The aim is not to privilege human thinking simply because it came first, but to ensure that human judgment remains active, open-minded, and accountable throughout the process.
4. The Cycle Is Iterative, but Iteration Does Not Guarantee Improvement
Although CAC is represented as Cognitive → AI → Cognitive, the process is inherently cyclical. The final cognitive stage may reveal gaps in understanding, generate new questions, or identify opportunities to explore alternative approaches, initiating another cycle. However, repetition alone does not guarantee improved learning.
Students may repeatedly interact with AI without making meaningful progress, particularly if they fail to question assumptions, verify information, or reflect on how their thinking is changing. There is also a risk that successive AI-generated suggestions may increasingly shape students' reasoning, making it difficult to distinguish independent understanding from ideas adopted through repeated interaction.
Effective iteration therefore requires a clear purpose. Each cycle should respond to an identified question, uncertainty, or opportunity for improvement. Students should be encouraged to consider what has changed, what has been learned, and whether further AI engagement is necessary. The educational value of iteration lies in progressively refining understanding and judgment, not simply increasing the number of AI interactions.
5. AI Literacy and Prompting Skills Require Explicit Instruction
CAC does not prescribe detailed prompting strategies or provide comprehensive instruction on the capabilities and limitations of different AI systems. The framework establishes when and why students should engage their thinking, but it does not independently teach them how to interact effectively with AI.
Without appropriate instruction, students may develop ineffective prompting habits, misunderstand AI capabilities, or accept responses without recognising their limitations. For example, students may not understand how to ask follow-up questions, request alternative perspectives, challenge AI-generated reasoning, or select appropriate AI tools for different tasks. These capabilities cannot be assumed to develop automatically through repeated AI use.
Teachers therefore need to incorporate explicit instruction, modelling, guided practice, and feedback to support effective AI interaction. As AI systems continue to evolve, students will also need opportunities to update their understanding of different models, tools, and capabilities. CAC should be supported by AI literacy instruction rather than treated as a substitute for it.
6. Dependence on Learning Design and Teacher Facilitation
CAC is intended to develop student agency, but its effectiveness depends considerably on how learning experiences are designed and facilitated. Simply introducing students to the three-stage process is unlikely to produce meaningful changes in thinking or learning behaviour.
Teachers need to consider the intended learning outcomes, students' prior knowledge, the cognitive demands of the task, and the appropriate role of AI. They also need to recognise when independent practice is necessary, when AI assistance may provide useful scaffolding, and when students are ready to take greater responsibility for regulating their own learning.
The level of support required will vary considerably. Novice learners may need explicit guidance and structured checkpoints, while more experienced learners may apply the process with greater flexibility and independence. Assessing the quality of cognitive engagement also presents challenges. A completed CAC activity or reflection may demonstrate that students followed a process, but it does not necessarily establish that meaningful learning occurred.
Teachers may need additional evidence through independent explanations, transfer tasks, discussions, demonstrations, or other appropriate assessments. CAC is therefore a pedagogical tool whose effectiveness relies on purposeful implementation, not a self-sufficient teaching method.
7. Scope: A Micro-Level Framework
CAC is primarily a micro-level framework focused on how individuals engage cognitively during AI-supported learning activities. It does not independently provide a comprehensive approach to curriculum redesign, institutional policy, assessment reform, or decisions about which knowledge and skills should be prioritised in an AI-integrated education system.
Similarly, although ethical considerations should inform decisions throughout the CAC cycle, the framework does not provide a complete model for addressing the broader social, economic, and environmental implications of AI. Issues such as privacy, academic integrity, equity, bias, intellectual property, accountability, and environmental sustainability require additional guidance, policies, and educational approaches.
CAC should therefore be understood as one component of responsible and effective AI integration, complemented by broader pedagogical frameworks, AI literacy education, ethical guidance, and institutional support.
8. The Need for Further Empirical Validation
CAC is informed by established educational theories and has been developed through teaching practice, reflection, and action research. However, a strong theoretical rationale does not, by itself, demonstrate that a framework improves learning outcomes.
Further empirical investigation is needed and is underway to determine how effectively CAC supports independent thinking, metacognitive awareness, evaluative judgment, and the development of cognitive capabilities across different educational contexts.
Important questions include whether CAC produces measurable learning benefits compared with alternative approaches, which students benefit most, how the framework should be adapted across subjects and levels of expertise, and whether its effects are sustained over time.
Research is also needed to examine whether the framework meaningfully reduces the four identified learning risks, rather than simply increasing students' awareness of them. Such investigation would help identify where CAC is most effective, where modifications may be needed, and which implementation strategies provide the greatest educational value. The framework should therefore remain open to ongoing evaluation, refinement, and evidence-informed development. This is where I ask the community for help. If you decide to use CAC, and find it beneficial, let's engage for research purposes.
Understanding CAC's Value and Boundaries
These limitations do not negate the value of CAC Co-intelligence. Rather, they help define its purpose and appropriate application. CAC is not intended to maximise productivity, prescribe every aspect of AI interaction, or guarantee learning simply by following three stages.
It is designed to provide a practical and adaptable structure that encourages students to think independently, engage purposefully with AI, and return to human cognition through critical evaluation, synthesis, and reflection. Its successful application requires recognising when AI assistance is beneficial, when independent cognitive effort is essential, and when different approaches may better serve the learning objectives.
Ultimately, CAC is not about rigidly following a sequence. It is about developing the capacity to make informed decisions about how human and artificial intelligence can work together to enhance learning while strengthening, rather than diminishing, human cognitive capabilities.