AI models, particularly black-box models derived from machine learning, often lack transparency, making it difficult to understand and trust their predictions. Upcoming regulations like the AI Act require companies to explain these models, creating a need for tools that can provide insights into their decision-making processes.

AI-vidence offers a software solution designed to enhance the explainability of AI models, enabling users to understand and validate their behavior. The platform employs a regional explanation principle, identifying the appropriate level of detail to analyze model predictions effectively. By incorporating logical AI components, AI-vidence aims to improve model robustness and provide operational guarantees. The platform offers tools for data scientists to work within a Jupyter notebook, as well as tools for business and compliance teams that facilitate real-time collaboration and collective intelligence.

Regional explanation approach to identify the right scale of observation for analysis

Complexity reduction through simpler model substitution and variable reduction

Causal link identification using the Causalink Python library

Predictive engine (Causalgo) adapted for formalization and prediction

Multi-scale analysis of language models (LLMs) using topic modeling

Tools for computer vision that classify components of a scene

Prototype interfaces for dynamically testing hybrid models

Support for both on-premises and cloud hosting

Open-source version (AntakIA) available

AI-vidence targets data scientists, business teams, and compliance teams seeking to understand, validate, and explain AI model predictions, as well as organizations preparing for AI Act compliance.

AI-vidence primarily generates revenue through software license sales, with annual licenses available for on-premises and cloud-hosted versions of its explainability software.