What is 'Causal AI' and why will it become increasingly important? 

The UK government recently released a white paper outlining its approach to AI regulation. The paper's purpose is "to ensure there are guardrails in place around the use of AI technologies, whilst allowing innovation to continue and the benefits to be felt far and wide." The approach is very much aligned with the aims of the EU’s Artificial Intelligence (AI) Act, as outlined in a previous Agenda article, as well as recent announcements from the White House.

Most regulations are trying to address similar fundamental issues with AI. We will use the UK as an example of major areas of focus for AI regulations. There are five key principles:

1. Safety, security and robustness

2. Transparency and explainability

3. Fairness

4. Accountability and governance

5. Contestability and redress

For enterprises looking to use AI for their most important decisions, this will become an increasingly difficult set of requirements to navigate, particularly with the current set of black-box AI approaches. Even the most exciting areas, such as generative AI, including large language models (LLMs), are a long way off meeting these requirements. Recently, for example, ChatGPT gave the wrong advice about breast cancer, which could have had a significant impact if this was used for decision-making. So, enterprises that want to experience the full set of benefits that AI can offer may need to look more broadly.

Causal AI is an emerging area of the broader field of AI. It naturally aligns very well with the above principles. The key idea behind Causal AI is to learn cause and effect relationships within data and use this to inform the output of AI models. This is vastly different from the approach that current state-of-the-art ML models, such as LLMs, take: consume a lot of data, learn the patterns and predict the next pattern.

Causal AI is an emerging area of the broader field of AI. It naturally aligns very well with the above principles. The key idea behind Causal AI is to learn cause and effect relationships within data and use this to inform the output of AI models. This is vastly different from the approach that current state-of-the-art ML models, such as LLMs, take: consume a lot of data, learn the patterns and predict the next pattern.

One way to think about LLMs is that rather than providing an answer to a question, what they actually tell you is ​​"when people ask questions like this, this is what the answers that other people usually give tend to look like.” This explains why LLMs have a tendency to ‘hallucinate’ when asked for factual information. LLMs and other generative models can significantly improve productivity when used to e.g. improve an existing piece of text or generate a piece of artwork. But using these tools for mission-critical decisions within the enterprise is a significant challenge and is unlikely to adhere to incoming regulations.

In contrast, Causal AI models are inherently explainable due to the way in which they are constructed. They can be interrogated for explanations as to why a particular output was reached and can easily be assessed for fairness and bias. Additionally, models are often constructed with a human-guided approach providing accountability, governance, contestability and redress.

The Causal AI toolset is one that enterprises can benefit significantly from. Causal AI also recently appeared on the Gartner® Hype Cycle™ for Emerging Technologies, 2022.

The types of questions that Causal AI models are well placed to answer include:

• Should I approve or reject this loan application and if so why?

• What causes my customers to churn and what actions can I take to make them stay?

• What caused this issue in my manufacturing plant and what actions can I take to prevent future issues?

Examples of where Causal AI is being deployed today include:

Causal AI can help identify the causes and effects of various medical conditions and treatment outcomes. For instance, it can simulate the impact of different treatment interventions on patient outcomes, allowing healthcare professionals to assess the effectiveness of alternative treatment options and make informed decisions about patient care.

Causal AI can aid in risk assessment and decision-making within the financial sector. It can analyse the causal relationships between economic factors, market events and investment outcomes. For example, it can provide insights into the causes of market volatility, assess the impact of regulatory changes on financial markets and help identify factors that contribute to fraudulent activities.

Causal AI can be used to optimise manufacturing processes and prevent issues. It can simulate the effects of process adjustments, equipment upgrades or supply chain changes to identify the most effective interventions for improving product quality, reducing defects and optimising production efficiency.

Causal AI can play a crucial role in understanding customer behaviour and improving satisfaction. It can identify the drivers of customer churn, determine the causal factors behind customer preferences and suggest personalised interventions to enhance the customer experience. This can help businesses tailor their products, services and marketing strategies to meet customer needs more effectively.

Enterprises that wish to become fully data-driven, while adhering to the necessary regulations need to use the full array of AI techniques that are available. Generative AI should be part of this, but it does not cover the full set of use cases where enterprises can make a real difference. Causal AI will allow enterprises to start creating solutions that can be used for their most critical tasks, including decisions that have a strong impact on their customers and the broader society.

GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally, and HYPE CYCLE is a registered trademark of Gartner, Inc. and/or its affiliates and are used herein with permission. All rights reserved.