Explainability is a cutting-edge technology to show how AI reaches decisions — in terms that a human can understand.
Explainable AI emerged as a solution to the problem of black box AI models, or models that are difficult to parse or understand why they reached a decision. In many highly-regulated industries, stakeholders are legally required to “show their work” or show proof of how an AI system reached a decision. Explainable AI emerged as one way to show model interpretability and audits logs.
Although explainability is touted as a solution to enable transparency and create more trustworthy AI, not all models can be explained, nor should they be. There is a tradeoff between performance accuracy of a model and explainability, and it is up to the end user to decide whether the risk of a lack of transparency is worth assuming. Certain deep neural networks (DNNs) remain too complex for today’s explainability methods, but are still leveraged because they perform very highly. Modzy’s approach to explainable AI and commitment to transparency allows users insight into how models were developed, trained, and overall performance.
Modzy's Explainability Solution
At Modzy, we recognize the link between explainability and trustworthy AI, and where possible, are building explainability features into our models and platform to align with our commitment to transparency. Our patent-pending approach is five times faster than the leading approach, offering both time and cost savings, as well as an increased level of trust. We recognize that many of our customers are applying AI in high-stakes situations, and that explainability is just one piece of what they need to secure, manage, and govern their enterprise AI applications. By building explainability in from the beginning, we’re helping our customers stay ahead of and accountable to the governance requirements of their industries.
Fastest Solution in the World
|Explainability Technique||Single CPU Intel Core i5.7360U||Single GPU Tesla V 100-SXM2|
|LIME||105 seconds||5.8 seconds|
|SHAP (Gradient Explainer)||35 seconds||3.8 seconds|
|Modzy's AXAI||2.5 seconds||1.4 seconds|