Presented by Carahsoft
Artificial intelligence can accelerate mission capabilities, but speed without assurance will not earn the confidence needed for widespread deployment. Thomas Baker, Deputy Chief Artificial Intelligence Officer at the Defense Intelligence Agency, says trust is the central problem DIA is working to solve as it expands its use of AI.
That trust has both technical and cultural dimensions. On the technical side, agencies need AI security, red teaming, testing, evaluation, verification and validation. Those disciplines help an organization understand how a system performs, where it may fail and what controls are necessary for its intended use. On the cultural side, senior leaders need enough confidence in that assurance framework to allow their teams to use the technology.
Baker offers the development of a junior analyst as an analogy. A new analyst’s first paper does not move directly to the highest level of government. It passes through doctrine, standard operating procedures and layers of review. A leader may not trust every sentence produced by a new employee, but the leader trusts the system that develops, checks and approves the work.
AI needs a comparable structure. The objective is not to persuade decision-makers that a model is infallible. It is to establish a system around the model that makes its use appropriately controlled, observable and reviewable. That allows leaders to judge where an AI capability is safe to use and where additional human scrutiny is required.
The workforce implications extend beyond learning how to write a prompt. AI can reduce the effort required for research, drafting and repetitive processes, but Baker does not believe the fundamental job of an intelligence analyst changes completely. The strongest analysts still develop deep expertise, apply critical thinking and identify the “so what” that matters to a decision-maker.
As the cost and time required for execution fall, judgment becomes more important. The differentiator is the person who asks the right question, recognizes whether an answer is relevant and understands what the model may have missed. AI can remove friction between systems and processes, giving employees more time for creative and consequential work, but it cannot replace the expertise required to evaluate that work.
Baker uses the term AI fluency to describe the broader capability employees need. Fluency includes an understanding of how AI works, its strengths and limitations, and how it can be applied within a mission and risk context. It will evolve as the technology changes. What counts as fluency today may be only a baseline as agents and other advanced capabilities become more common.
Adoption strategy matters as much as training. Baker argues that forcing employees to use AI is unlikely to create lasting change. Instead, DIA can “pave the road” by providing approved, useful capabilities, reducing barriers and sharing examples of how colleagues have used them successfully. Word of mouth from trusted peers can be more persuasive than a directive.
Different users will move at different speeds. Early adopters may be willing to turn over too much and benefit from additional skepticism. Others may avoid the tools until nearly everyone else has demonstrated their value. A mature program allows employees to define where they are comfortable using AI while maintaining common security and assurance requirements.
Trust is not a one-time certification. It is the product of governance, evidence, human judgment and repeated mission success. By building an assurance system around AI and an increasingly fluent workforce around that system, DIA can pursue speed without asking leaders or analysts to accept risks they cannot understand.