Authors: Judith Hurwitz and Stuart Frost
Causal Artificial Intelligence
AI is increasingly recognized as the dominant technological trend of the next decade. Although AI has existed for decades, its evolution has reached a stage where it significantly impacts various applications and industries. While recent attention has primarily centered on Generative AI and Large Language Models (LLMs), more organizations are now focusing on Causal AI. For businesses, leveraging AI to enhance decision-making is crucial, and Causal AI is emerging as a vital technique for problem-solving. Unlike correlation, Causal AI aims to uncover the underlying reasons for issues and identify effective solutions.
The Limitations of Correlation-Based AI Approaches
Most AI solutions traditionally rely on correlation, a widely accepted statistical method for examining relationships between data sets. However, this approach has limitations; it assumes that past outcomes can predict future ones. Data scientists have depended on correlation for decades, especially in times when changes in outcomes were incremental. Subtle shifts in patterns could signal significant developments for businesses. By applying data analytics and machine learning, correlation can better predict outcomes based on these patterns. Yet, it falls short of explaining why problems arise or suggesting ways to alter outcomes.
Causal AI: Going Beyond What Happened to Explain Why
This is where Causal AI becomes essential. Consider this simple example: your organization notices a significant slowdown in sales of a particular product. What’s the issue? Historically, sales have increased by 10% annually since the product's introduction. What has changed? Is it a pricing problem? Do customers need new incentives? Is product quality an issue? Are new competitors entering the market? While correlation can analyze statistical changes over time and identify regions with lower sales, it cannot explain why these trends are occurring.
Building Models That Capture the Why Behind the What
In contrast, Causal AI begins by clarifying the problem at hand. In this case, the organization must first identify the variables affecting product marketing, including pricing, economic factors, marketing campaigns, and distribution channels. The model will elucidate the relationships between these elements. For instance, how crucial are distribution channels? Have competing products' prices changed? What role does pricing play in sales? Have product returns increased in the last six months? Once a model is established, the organization can input data related to these factors, enabling the team to utilize Causal AI for deeper insights into outcomes. For example, if a new competitor offers a similar product at a lower price, what would happen if your business reduced its prices? How would that affect sales and profits? What if the company addressed customer complaints? Would resolving these issues lead to increased revenue? While this may seem straightforward, the real challenge could be that the competing product is not just cheaper but also superior in functionality.
Judah Pearl's Insights on the Power of Causal Reasoning
As Judah Pearl, a pioneer in Causal AI, states in his book The Book of Why, “Data does not understand cause and effects: humans do.” The strength of Causal AI lies in its ability to empower organizations to make informed decisions. A key reason Causal AI is transformative is that it provides a consistent framework for all stakeholders—including business leaders, data scientists, and subject matter experts—to share a common understanding of problems. By elucidating the reasons behind issues and offering tools to improve business outcomes, Causal AI holds immense potential.