Transparent AI for actuaries
So far, the power of AI has not really revolutionized insurance in critical areas such as pricing. Today, most actuaries understand what AI can bring to actuarial pricing, but their hands are tied, because AI models with superior predictive power lack the direct transparency actuaries need.
We have developed our in-house algorithm, built on the latest advancements in AI, that enables a fundamentally different modeling approach, allowing your actuaries to build models that are on-par with black-box models (e.g., neural networks, random forests) when it comes to predictive power while maintaining the transparency of traditional methods (e.g. GLM).
Why Tsetlin Machines?
The type of the Tsetlin Machines (TMs) we use as the basis for our algorithm is very recently developed and is considered a neuro-symbolic AI model (combining symbolic and connectionist AI).
Let's compare Bernoly Tsetlin Machines to Generalized Linear Models (GLMs) that have been the actuarial gold standard for decades:
Automatic Interaction Discovery:
Unlike GLMs, where you need to manually engineer interaction terms, TMs learn interactions automatically without the risk of overfitting from over-parameterization.
Non-linear Modeling, Without Losing Transparency:
TMs can model complex non-linear relationships without becoming black boxes like neural networks, random forests or gradient-boosted trees.
Resilience to Noise and Sparsity:
Insurance data is often noisy or imbalanced (e.g., many zero-claim records). TMs have shown strong resilience in such conditions, especially when features are binary or categorical.
Tsetlin Machines use cases in insurance
We have developed several insurance applications with our in-house Tsetlin Machine model:
- Pricing
- Fraud detection
- Underwriting automation
- Risk capital modeling (under development)
- Actuarial modeling
Why insurers need this now?
