How Diveplane uses explainable AI to bolster AI adoption

“Black box” artificial intelligence (AI) systems are designed to automate decision-making, mapping a user’s features into a class predicting individual behavioral traits such as credit risk, health status, and so on, without revealing why. This is problematic, not only because of the lack of transparency, but also because of potential biases inherited by algorithms from human prejudices or any hidden elements in the training data that may result in unfair or incorrect decisions. 

As AI continues to proliferate, there is an increasing need for technology companies to demonstrate the ability to trace back through the decision-making process, a functionality called explainable AI. This would essentially help them understand why a certain prediction or decision was made, what the important factors were in making that prediction or decision, and how confident the model is in that prediction or decision.

To help instill user confidence that operational decisions are built on a foundation of fairness and transparency, Diveplane claims its products are designed around three principles: predict, explain and show. 

Explosive growth in the AI software market

Raleigh, North Carolina-based Diveplane today announced that it has raised $25 million in series A funding to bolster its position in the AI software market and invest further in its explainable AI solutions that provide fair and transparent decision-making and data privacy. 

Gartner estimates that the AI software market will reach $62 billion in 2022, and continue to grow at a rate of more than 30% through 2027. Diveplane claims it’s positioned to capitalize on the market opportunities with its support of multiple real world use cases — prediction, anomaly detection, anonymization and the creation of synthetic data -– all from a single model on one platform.  

Diveplane’s jump from gaming to explainable AI  

The company is led by Mike Capps, the former president of Epic Games, and Hazardous Software’s cofounders Chris Hazard and Mike Resnick.

In fact, Diveplane’s AI technology spun out of Hazardous Software, which  Hazard and Resnick founded in 2007. 

The game gained a lot of mainstream attention, even from places the founders didn’t expect — the U.S. Army. Though they had to go back to the drawing room to reconfigure the technology for military purposes, they were able to build AI software for decision support, visualization and simulation for hard strategy problems. 

They eventually spun off Hazardous Software and established Diveplane to start developing an explainable AI system that could support multiple use cases from prediction, anomaly detection, anonymization, and the creation of synthetic data. And the rest, as they say, is history.  

“Today, we offer practical, ethical, efficient machine learning. And you will only need one model to do everything,” Hazard told VentureBeat.  

Explainable, auditable and editable 

At the heart of Diveplane’s offerings is Reactor, a cloud-based machine learning (ML) technology that creates AI decision-making models based on historical data observations. It helps identify anomalies in real-time systems, create shareable synthetic datasets that take the place of highly-sensitive personally-identifiable information, and predict the future. 

“While a lot of folks focus on neural networks or gradient boosting, we use something called instance-based learning, where your data is the model,” said Hazard. “The best part is you don’t need to build 10,000 models to answer 10,000 questions, you need to build one.”

The company claims that the Reactor removes the “black box” entirely by being explainable, auditable and editable. Using Diveplane’s patented “conviction metric,” which represents how surprised the system is by new data, Reactor helps users understand clearly about how it arrived at a certain decision and exactly what data informed that choice. This way, organizations can pinpoint and analyze potentially biased data and remove it from all future analysis.

Reactor is the core technology upon which all of Diveplane’s other products are built — GEMINAI (a synthetic data creator), SONAR (anomaly detection tool), and ALLUVION (commercial real estate prediction tool). 

Partnerships are the way forward, Diveplane says

Diveplane targets highly regulated industries such as finance, healthcare, commercial real estate and defense, among others. However, the company partners with multiple vendors to serve other markets. 

“We’re building trusted partnerships, with a product set that provides a holistic capability for fair and transparent decision-making and data privacy. This support adds rocket fuel to our business, so we can build on our successful approach to helping companies innovate with our Reactor platform,” Capps said in a statement.

Most recently, Diveplane partnered with Scanbuy, a mobile engagement and digital advertising solutions provider, to launch ExtendedAudiences, a privacy-protected, act-alike audience extension of U.S. consumer CPG purchase data. 

“AI- and ML-derived value will be critical to our industry’s future. By requiring that all our models be inherently understandable, Scanbuy allows self-attestation data audits to be extended to modeled audiences,” said Chai Outmezguine, CEO of Scanbuy, in a statement. 

“Model explainability allows us to determine how predictions are built, protect attributes across data partners, and maintain the data subject’s ultimate privacy and control over usage. We also see a bright future in introducing this transparency to reinforcement-learning techniques that help optimize in-flight campaigns.”

Since its inception in 2018,  Diveplane has raised nearly $35 million. It’s secured investments from Shield Capital, Calibrate Ventures, L3Harris Technologies, and Sigma Defense, alongside star-studded investors including U.S. women’s soccer stars Megan Rapinoe, Becky Sauerbrunn, Meghan Klingenberg and Mia Hamm. Philip Bilden, Managing Partner of Shield Capital, will join the Diveplane board of directors.