3. Build trust and “explainability” in your AI

Humans still decide how to use and act on the analytical information that AI provides, so its adoption hinges on understanding how it works. As government missions increasingly look to use AI, users must be comfortable with the outputs and decisions made by learning models. Having humans in the loop, with users continuously validating and training models, is critical to establishing “responsible AI” and trusted human-machine partnerships.

Establish transparency in your AI processes and solutions that allows machine-learning products to be traced and audited. “Explainability” is being able to go back and understand, starting from data ingest, what datasets were used, how the data was prepared, and how learning models were developed and by whom to make proper decisions and perform accurately. Once a model has been operationalized, look at its deep metrics for biases and drift. Seek out transparent tooling solutions to promote trustworthy and ethical AI.

4. Make AI accessible and unintimidating

It's very hard to find data scientists, and even harder to find those with clearances for defense and national security missions. I believe in creating “citizen data scientists” by giving them the ability to interact with data and machine-learning models intuitively. For the non-technical members on mission teams, it’s about usability and comfort and empowering them to be able to work with analytics tools and extract the insights they need self-sufficiently.

There are powerful AI solutions on the market that do great things, but focus on those that provide step-by-step repeatability for users and open-architecture integration for current and future mission systems. Maybe surprising is that across government — the Department of Defense, intelligence community and civilian agencies — different missions are looking to AI to solve very similar problems, such as detecting objects within images, videos and text. Listening to our government partners and their users, SAIC has developed prebuilt, easy-to-use and tailorable AI models for things like computer vision and data fusion — what we call Mission Accelerators in our Tenjin AI-development and orchestration tool.

AI is not about replacing human capital but about augmenting it

Despite challenges encountered whenever new technologies are brought on, the federal government is implementing AI initiatives. It's not enough to solve the data problem and build powerful AI tools; we must ensure that users trust and embrace them, while also addressing the main obstacles of dealing with data, policy and adoption.

Responsible AI with explainability and intuitive user interactions is crucial to uptake across government missions. It's also important to keep in mind that AI is meant to enhance our abilities and speed up our progress towards our goals. Augmenting workflows and allowing us to do higher-value work, AI empowers us to accomplish missions faster and better.

As part of SAIC’s LinkedIn Live series, I recently had the opportunity to discuss these topics and more with Manmeet Singh from Koverse, Inc., an SAIC Company, to help our audience members make informed decisions when implementing AI. Watch the LinkedIn Live recording for the full discussion.