What is causal AI?

Causal AI can answer questions about how changing a specific variable might impact the outcome of a calculation.

Fault tree analysis is a practical causal AI example. It’s a top-down method that uses Boolean logic to trace the events that can lead to system failures. By mapping relationships between component failures and system malfunctions, fault tree analysis helps identify the root causes of problems.

The work of causal AI involves several key steps:

1. Observational data collection
   Causal AI systems gather large amounts of observational data that track events and metrics over time.
2. Causal relationships discovering
   Algorithms analyze data patterns to detect potential causal links between variables, using techniques like causal discovery to create causal models.
3. Causal models building
   The identified relationships are used to build causal models, such as Bayesian networks or structural causal models, representing dependencies between variables.
4. Domain expertise incorporation
   Experts provide input to refine causal models, specifying known relationships and integrating human insights with data-driven models.
5. Causal effects estimation
   Using techniques like counterfactual analysis, causal models estimate the impact of hypothetical interventions by simulating changes in variables.
6. Test interventions
   Once the models are established, causal AI uses them to estimate the effects of hypothetical interventions through counterfactual analysis. Organizations can then test these interventions on a small scale or in simulations to predict their effectiveness before wider implementation.
7. Iteration
   As new data becomes available, causal models are continuously refined to enhance their accuracy and provide ongoing insights into causal relationships.

Causal AI focuses on understanding the underlying why behind events. It aims to identify the cause-and-effect relationships between variables in a system. Generative AI, on the other hand, is concerned with creating new content. It learns patterns from existing data and uses that knowledge to generate new, similar data.

Example of causal AI vs. generative AI:

Causal AI could analyze historical traffic data to determine which factors, such as road construction, traffic signals, or public transportation availability, are most correlated with traffic congestion.

Generative AI could generate synthetic traffic data based on real-world patterns. This would allow city planners to simulate various scenarios, such as changes in traffic patterns, road closures, or the introduction of new transportation systems.

The key difference between causal AI and correlation AI lies in their ability to understand the "why?" behind relationships:

• Correlation AI identifies patterns and relationships between variables in data. It can tell you if two things are related, but not why.
• Causal AI goes a step further by determining whether one variable causes changes in another. It aims to understand the underlying mechanisms and causal relationships, not just the correlations.