The Explainability Factor: Lockheed Martin and DarwinAI’s Game-Changing AI Platform

As Lockheed Martin continues to invest in and expand the depth of our artificial intelligence (AI) projects, our specialists are continually on the lookout for new technologies to meet the challenge of an increasingly complex battlespace. On May 20, 2020, Lockheed Martin and DarwinAI, an explainable AI company based in Waterloo, Canada, announced a strategic collaboration aimed at deepening our company’s current understanding of AI solutions and responding to the black box problem using the startup’s unique explainability platform.

Explainable AI or “explainability” illuminates how neural networks – complex constructions that mimic the human brain – effectively reach their decisions for a more robust end product for the warfighter. According to DARPA, the future warfighter faces challenges that demand more intelligent, autonomous and symbiotic systems. However, the effectiveness of these systems will be limited if the warfighter cannot understand – or appropriately trust – the emerging generation of artificially intelligent machine partners.

In an article describing the dilemma of the black box, DarwinAI CEO Sheldon Fernandez explained that the system’s behavior is a reflection of the data that the network is trained against and the human labelers who annotate that data. Explainability, which is a key component of the GenSynth platform, provides the ability to peek inside this black box and understand the decision-making process of a neural network – a major benefit for the aerospace industry which is especially conscientious about issues of ethics, regulation and reliability. 

DarwinAI’s GenSynth (short for Generative Synthesis) platform mitigates the black box challenge in several ways: it improves neural network efficiencies and reduces the time it takes to produce robust and accurate models through the insights it generates. The platform also provides a detailed breakdown of how a model performs for specific tasks at the neuron level. According to DarwinAI’s public product release in November 2018, this deep understanding of the network’s components and their involvement in specific tasks enables a developer to fine-tune the model designs for efficiency and accuracy, effectively operationalizing AI.

In addition, the GenSynth platform enables developers to:

• Automatically generate new models which meet performance targets (such as accuracy) within operating restrictions (for example, parameters, size, FLOPs) – reducing development timelines from months down to days. 
• Provide transparency at every stage of the development process, from revealing bottlenecks in models, to visualizing performance comparisons of different experiments, to identifying errors, biases and issues in both models and data. 
• Reveal critical factors which cause the model to make decisions – showing why decisions are made, unearthing hidden bias and enabling more effective and efficient audits. (Such insights are key in implementing robust and trustworthy AI systems.)

Embracing AI promises considerable benefits for the warfighter through its contributions to productivity growth and innovation. According to Lockheed Martin’s AI Center senior manager Dave Casey, AI can sort through data faster than any human mind, reducing cognitive load of the soldier in the field so that they can focus on other components of the mission. Considering AI’s impact on how militaries operate and protect their forces, it is fundamental to comprehend how its internal processes reach their decisions.

With so much at stake, Fernandez added, it is critical that AI works in all scenarios in the field – even the unexpected scenarios – in a robust fashion: “In layman’s terms, this means that the AI will behave in a way that you intend, that it will be trustworthy under a variety of permutations.”

As Lockheed Martin continues to deliver state-of-the-art solutions that help our customer confront complex warfare scenarios, our robust AI products will seamlessly secure and connect all assets in the joint battlespace to enable prompt action. With the ability to facilitate autonomous operations, streamline military decision-making and increase the speed and scale of military action, it is essential that future AI applications are explainable and trustworthy.