Ending the black box approach to artificial intelligence | FROG.AI

Ending the black box approach to artificial intelligence

Many machine learning models today provide users with results using a black box approach. Modelling conducted using deep neural networks and reinforcement learning has provided impressive results in recent years but their uptake for real decision making has been slow.

The goal of FROG.AI is to provide humans with the confidence to deploy artifical agents for real-world decisions. We do this through extensively testing and understanding the decisions our artifical agents make and providing explantions for those decisions with automated natural language reports.

Data to Actions

We use the latest deep reinforcement learning algorithms to train agents to determine sensible actions from raw data of an environment. Initial testing occurs on historical data and the agents then evolve by adjusting to how they perform in the real-world.

Action Analytics

We continuously perform rigorous automated testing of all our agents. We use sensitivity analysis with real and augmented data to work out the critical inputs that determine the actions our agents make.

Explanations by NLG

The final part of our pipeline is to convert our action analytics into natural language to provide reports for human readers. These explanations provide human operators with the confidence to follow the advise provided by the artificial agents.