Critical Realism

Theory: Critical Realism | Template: The Deep Dive | Words: 1,433

# Critical Realism: Beyond the AI Correlation Trap

We often hear terms like "objective research" versus "subjective experience." It’s a classic fork in the road for anyone trying to understand the world. Most people feel they have to choose one path or the other. This binary thinking, this idea that reality is either black or white, misses a crucial, deeper truth. It's why a profound philosophical lens, Critical Realism, is so often cited, yet rarely understood in its full power.

The Popular Version

In many professional circles, Critical Realism is often seen as a sophisticated way to say, "both qualitative and quantitative methods are good." It's interpreted as a call for mixed methods research, or simply an acknowledgement that context matters. People might nod and agree that "reality is complex" and "there are objective facts, but our understanding is subjective." While these statements aren't wrong, they skim the surface. They capture a sliver of Critical Realism's spirit, but they miss its radical refusal to accept the either/or choice that underpins much of our research and decision-making.

What the Original Actually Says

To truly grasp Critical Realism, we must go back to its origin with Roy Bhaskar. In his seminal work, A Realist Theory of Science (Bhaskar, 1975), he didn't just suggest a middle ground. He argued for a fundamentally different way of understanding reality itself. Bhaskar proposed that reality is stratified, meaning it exists in layers, much like an onion.

He identified three distinct, yet interconnected, domains: 1. The Empirical: This is what we experience. It’s our observations, our data, what we can directly perceive or measure. Think of it as the surface-level information. 2. The Actual: This refers to what happens, the events themselves, whether or not we observe them. A tree falling in a forest makes a sound, even if no one is there to hear it. 3. The Real: This is the deepest layer. It consists of the underlying structures, mechanisms, and powers that produce the events we see in the actual domain, and which we might then experience in the empirical. These mechanisms exist and operate independently of our awareness. They are the "why" behind the "what."

For Bhaskar, the goal of science isn't just to describe what we observe (the empirical) or even just what happens (the actual). The true aim is to uncover the generative mechanisms in the real domain that cause those events. This means moving beyond mere correlations to understand the causal powers at play. Our access to this deep reality is always partial and influenced by our theories and perspectives, but the mechanisms themselves are real, regardless.

What Changed Since

Bhaskar laid the groundwork, but Critical Realism has evolved significantly, offering powerful tools for understanding complex systems. Subsequent researchers have built on his stratified ontology, making it more actionable for various fields.

One crucial development comes from Margaret Archer, who expanded Critical Realism into a "morphogenetic approach" (Archer, 1995). Archer focused on the dynamic interplay between social structures and human agency. She argued that structures (like educational systems or organizational cultures) aren't just static backdrops. They both enable and constrain our actions. Our actions, in turn, can either reproduce those structures (morphostasis) or transform them (morphogenesis). This offers a powerful lens for understanding how complex systems, like an adaptive learning platform in a school, influence individuals, and how individuals then shape the system over time.

Andrew Sayer further clarified the methodological implications of Critical Realism (Sayer, 2000). He emphasized the need for context-sensitive explanations, urging researchers to identify the "causal powers" inherent in social structures and agents. It’s not enough to say "X is correlated with Y." We need to ask: what properties of X give it the power to influence Y, and under what specific conditions? This pushes us beyond statistical averages to detailed, nuanced understanding.

More recently, Danermark and colleagues provided a practical guide for conducting Critical Realist research, highlighting the importance of "abduction" and "retroduction" (Danermark et al., 2019). These are forms of reasoning where we move from observed effects back to their plausible underlying causes, then test those causal hypotheses. This isn't just about finding patterns; it's about explaining how those patterns came to be.

Interestingly, Critical Realism isn't always at odds with quantitative methods. Tsang and Kwan (1999) argued that realism and positivism can even be combined, particularly in replication studies, to deepen theory development. This shows that Critical Realism isn't about rejecting empirical data, but about using it to probe deeper for the mechanisms that generate it. For example, a study at the University of Cambridge investigated a new e-learning system using a critical realist approach (University of Cambridge, 2010). Instead of just measuring its effectiveness, researchers focused on identifying the underlying mechanisms influencing adoption. They found success depended on factors like alignment with existing teaching practices and perceived value, leading to targeted improvements.

The Modern Application

This deeper understanding of reality has profound implications, especially in the age of Artificial Intelligence. AI excels at identifying correlations and patterns within vast datasets. It can tell us what is happening, and often predict what will happen next, with incredible accuracy. This is AI's superpower: describing the empirical and actual domains.

However, AI often stops there. It can show us that students who use a certain adaptive learning module perform better on tests. It can even predict which students are at risk of failing. But it often doesn't tell us why. What are the underlying cognitive, pedagogical, or motivational mechanisms that led to that improved performance or that risk of failure? This gap between correlation and causation is where Critical Realism becomes indispensable.

Consider the alarming statistics around AI adoption and success. A 2021 study found that 74% of AI projects fail to deliver on their intended business outcomes (Gartner, 2021). Another survey revealed that only 53% of organizations successfully put AI projects into production (VentureBeat, 2019). And a 2020 survey showed that a staggering 87% of data science projects never make it into production (VentureBeat, 2020). While the global AI market is projected to reach $150 billion in 2023 (Statista, 2023), a significant barrier to adoption is a lack of understanding of AI itself, cited by 40% of companies (PwC, 2023).

These failures often stem from a focus on surface-level correlations without understanding the deeper mechanisms. An AI model might predict equipment failure, leading to improved uptime. But a critical realist investigation in a large manufacturing company revealed that the AI system didn't just predict; it also prompted changes in operator behavior and maintenance schedules (A Large Manufacturing Company, 2018). These human-driven changes, triggered by the AI's insights, were the true causal mechanisms behind the improved uptime, not solely the AI's predictive accuracy. Without this deeper dive, we might wrongly attribute success, replicate ineffective strategies, or fail to truly optimize.

The Reference Guide

Critical Realism offers a powerful lens to move beyond superficial observations and unlock deeper insights, particularly when working with AI and complex systems like adaptive learning. Here are its core principles in action:

• Layered Reality: Understand that what you see (empirical data) and what happens (actual events) are products of deeper, unobservable structures and mechanisms (the real).
• Beyond Correlation: Don't stop at identifying patterns. Ask what causal powers or generative mechanisms produced those patterns. AI shows what; Critical Realism seeks why.
• Context Matters, Deeply: Acknowledge that mechanisms operate differently under varying conditions. Research isn't just about finding universal laws, but understanding how causal powers manifest in specific contexts.
• Structure and Agency: Recognize that individuals are shaped by social structures, but also have the power to act and, over time, reshape those structures. This dynamic is key in learning environments.
• Refusal of Binary Thinking: Reject the false choice between objective facts and subjective interpretations. Both are part of a richer, stratified reality.
• Informing AI, Not Replacing It: Use AI's power for prediction and pattern recognition, then apply Critical Realism to interpret those patterns and uncover the underlying causal mechanisms, leading to more robust and explainable solutions.

The Challenge

AI is a phenomenal tool for uncovering correlations and making predictions. But if we deploy AI-driven learning systems without genuinely understanding the causal mechanisms at play – the "why" behind the "what" – we risk building powerful black boxes. We might optimize for metrics without knowing if we're truly fostering deep learning, or simply gaming a system. Next time you see an impressive AI outcome, ask yourself: Can AI truly transform education without understanding the 'why' behind the 'what'?