Because what we consider AI currently is so new, there are a lot of uncertainties and nuances around how to embed responsibility into AI systems.

As discussed previously, responsibility includes: accuracy, explainability and interpretability, privacy, reliability, robustness, safety, security, the mitigation of harmful bias, and more as the field evolves. A challenge to science generally is that there are no perfect answers or approaches. Due to the speed and scale of progress in AI, practitioners in the field likely will be learning by trial and error for the foreseeable future.

Some foreign governments, international entities, and U.S. agencies have already begun to create high-level AI principles, and even some policies around AI’s responsible and trustworthy use. These are important first steps, but next these principles must be translated into actionable steps that agencies can use throughout the AI development process.

When it comes to the practical implementation of AI in government — again with the fundamental requirement of responsible and trustworthy AI — researchers and practitioners are continually iterating and learning. If readers of this guide want to dive more deeply into responsible AI, there are numerous sources including within the Federal government such as the Department of Defense Ethical Principles for AI and the Executive Order on Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government.