Causal AI:
AI Humans Can Trust

The organization doesn’t use artificial intelligence to make decisions. Most organizations are stuck at L0. AI is confined to data science experiments and has no impact on real decision-making. The limitations of standard machine learning technology are to blame.

Core problems include a lack of adaptability to real-world dynamics, and a lack of explainability. Data scientists may attempt to implement “post hoc explainability” methods, but these methods do not produce actionable insights. 

There are many other limitations of standard machine learning (see L1). However, these two mean that business decision-makers do not trust AI systems sufficiently to let them out of the lab. Read up on the shortfalls of current state-of-the-art AI here. 

Organizations stuck at L0 are wasting resources on AI investment which has zero impact.

AI is deployed in a narrow range of use cases across the organization and may factor in some real decision-making. Fewer than one in ten organizations have reached L1. They can begin to participate in some of the benefits of AI. However, at L1 the organization is still frustrated with its AI systems, which have minimal impact on significant decisions. 

Machine learning technology is again responsible. The core problems at L1 include:

Lack of adaptability and explainability. The key problems at L0 remain, but at L1 organizations cope with these problems by deploying AI in low-stakes contexts. 

Predictions only. Conventional machine learning models are only suitable for making predictions. They can’t meaningfully help with decision-making, which requires a model of how interventions impact an environment. 

Need for big data. Machine learning techniques and open-source software developed by big tech are designed for data-intensive problems. They have little transferable applicability to most industries which need to work with small data. Enterprise AI platforms simply automate open-source software that is not designed for the enterprise. 

Limited human-machine interaction. Users can direct machine learning algorithms to focus on certain variables, but that’s where interaction begins and ends. They can’t convey context to the AI or constrain its assumptions.

Fairness. Machine learning algorithms amplify human biases and perpetuate historical injustices. This creates ethical, legal, and reputational liabilities.  

These shortcomings of machine learning severely limit the range of applications of AI. Decision-makers do not trust them to assist with important decisions. At L1, AI has no impact on most stakeholders and their decisions.

The organization harnesses Causal AI, the only technology that learns and reasons about the world as humans do. Causal AI has three key pillars: an ability to learn cause-and-effect relationships; the capability to simulate interventions; and a computer imagination that can reason beyond the constraints of historical data. Causal AI is the only AI that decision-makers can trust. 

While L1 technology is essentially a form of “pattern-matching” or “curve fitting”, L2 technology is genuinely intelligent. 

Causal AI provides:

Intrinsic explainability. Causal AI can explain why it took a decision in human-friendly terms. The assumptions behind the model can be scrutinized, constrained, and corrected by humans before the model is fully trained. This reinforces trust that the model will perform as expected in production. Read more here. 

Adaptability to regime shifts. Causal AI continuously discovers invariant relationships that tend to hold over time, and so it can be relied on to make sound decisions even as the world changes. Learn more here. 

Optimal decision-making. The AI can go beyond predictions and directly take decisions and make recommendations. The most mature Causal AI systems are capable of autonomously optimizing for abstract business goals and KPIs (autoKPI™). View our product pages here. 

Human-machine interface. Causal AI keeps humans in the loop — users can communicate business context to the AI, shaping the way the algorithm “thinks” via a shared language of causality. Read up on how Causal AI empowers experts here. 

Small data handling. Causal AI autonomously finds the critical information in limited data and harnesses human knowledge, allowing it to make recommendations with small datasets that are ubiquitous in business and government. Find out more from our blog. 

Algorithmic fairness. Causal AI can envision futures that are decoupled from historical data, enabling users to eliminate biases in input data. Causal AI empowers humans to impose fairness constraints before models are deployed. Read more in this article. 

These breakthrough capabilities enable Causal AI to make trustworthy decisions in applied settings. This is what’s needed to take AI from the lab to the field where it can make an impact and unlock enormous value.

The enterprise is controlled by an Artificial General Intelligence (AGI) — a general-purpose technology at least as intelligent as humans. Higher causal reasoning is a defining feature of human cognition — equipping machines with this ability is a necessary condition for reaching AGI. L3 is beyond the technological horizon for now — we estimate it is some decades away.

If reached, we envision AGI having a dramatic impact on businesses. Unconstrained by biological substrate, an AGI’s digital mind may innovate, strategize, and execute on superhuman timescales. There may be a “winner-takes-all” effect for companies that first reach L3. The current economic development model may change beyond recognition.