Bob Ritchie is a fast learner. Fresh out of Virginia Tech, the technologist joined SAIC as an intern and just three years later became a lead software architect supporting the U.S. Navy and Marine Corps in the digital transformation of existing Department of Defense tactical decision-making and command and control (C2) systems. He remained committed to the company throughout his career, working part-time until he eventually left in the late 2010s for a new opportunity at Capital One, where he led a major migration to the Amazon Web Services public cloud.
Given his incredibly precocious career and rapidly accumulated technical acumen, it’s no surprise that in 2022 he was named Chief Technology Officer of the SAIC Innovation Factory, where he is proud to play a role in the first-of-its-kind, confidential, publicly accessible, cloud-based software factory and the Department of Defense’s Cloud One enterprise.
GovCon Wire recently had the opportunity to speak with Ritchie about that most pervasive technology: artificial intelligence, and it was no surprise that Ritchie shared plenty of insights about what the federal government is currently looking for in an AI system and where the field is headed in the future.
GovCon Wire: For which applications of AI/ML are you seeing the most demand from your federal government customers, and can you explain what is driving that demand?
Bob Ritchie: The demand I see in the Federal space typically falls into two segments. The first is what I would characterize as productivity demand, which is almost dominated by the large-scale language model, search augmented grid, and large-scale action model marketplaces. Specifically within Federal Government, this market is very focused on how to safely and reliably leverage closed and open source models to offload the tedious and error-prone aspects of business and mission workflows. This segment is similar to prior technologies such as robotic process automation, but with a significantly more powerful set of tools in the toolbox.
The second segment, and frankly the more challenging, is the continued development and advancement of models that provide insights and forecasts in near real time, moving into the realm of predictive and prescriptive analytics. With great power competition in the areas of national security and defense, improving the overall experience of citizen services, and conquering new frontiers in autonomous systems in space, air, land and sea, the demand for truly operational and relevant AI solutions has never been greater.
GovCon Wire: Where do you see AI going in the future?
Bob Ritchie: One of the fundamental directions that AI solutions are focusing on is governance, transparency, and explainability. While not too far removed from the old marketing/influence perspective (“4 in 5 dentists agree”), people need a higher level of understanding and trust in the output before they can accept recommendations or leverage machine-to-machine decision making.
Over the past two decades, I have been involved in several AI/ML user acceptance testing scenarios where solutions were rejected for reasons that can be paraphrased as “this isn't the way I meant to do it, so it can't be right.” There are numerous options to improve the accuracy, relevance, context, and explainability of AI-based solutions, such as utilizing expert mixture models, orchestrating recursive multi-shot action models, and providing advanced visualizations of nested probabilistic decision trees. Given the current barriers of cost and availability of computing power, and the exhaustion of publicly available training data sets, near-term advances in AI will all revolve around how to efficiently balance fine-tuned and fit-for-purpose models in combination with state-of-the-art LLMs.
GovCon Wire: To use data effectively, it must be collected, analyzed, and understood – which is key to running AI systems effectively. What key challenges and opportunities do you see emerging as organizations leverage data to make decisions?
Bob Ritchie: It's almost a cliché to say at this point, but one of the major challenges and opportunities that always exists is the lack of proper data hygiene as a starting point within organizations. Many organizations go all in and want to leverage large language models, but many of them don't have their data house organized properly enough to truly realize the value proposition. However, an interesting phenomenon is that organizations that understand this gap and the limitations of large language and/or action models can leverage those same models to significantly reduce the effort of understanding their organization's data and remove barriers by establishing semi-automated processes for defining, refining, and maintaining semantic ontologies.
Another challenge is the cost and impact of the raw compute power required to keep up with existing demand, especially when it comes to inference in operationally critical scenarios. Many companies are using this innovation opportunity to develop alternative chip architectures that significantly outperform GPUs for inference, and we expect to see many more fit-for-purpose chip architectures emerge.
GovCon Wire: What does Zero Trust success look like in your opinion, and how is SAIC working to help its federal government customers achieve it?
Bob Ritchie: One of the beauties of the Zero Trust ethos is that there is no ambiguity in its objectives. What degree of trust? Zero. But like anything else, there are objectives, there are thresholds for success, and it all starts with a relentless commitment to minimizing implicit trust and maximizing data protection. SAIC is pioneering the design and implementation of the journey from net-centric solutions to the data-centric ecosystem of the future across the Federal Government and our allies. By embracing Zero Trust as a journey, balancing innovation and speed with evidence, risk management, and objective policy change recommendations, and collaborating openly across industry and the Federal Government, SAIC is committed to ensuring the right data is securely accessed by the right resources at the right time, in the right context, to drive truly differentiated mission outcomes.
