Software Engineering Institute presents three pillars of AI engineering

PITTSBURGH, June 30, 2021 / PRNewswire / – SEI today announced the publication of white papers outlining the challenges and opportunities of three early pillars of artificial intelligence (AI) engineering: human-centered, scalable, robust and secure.

To mature AI practices and help national defense and security agencies adopt AI, SEI has started to formalize the field of AI engineering, just as it has done for the software engineering in the 1980s. AI engineering is an emerging area of ​​research and practice that combines the principles of systems engineering, software engineering, computer science, and business-centric design. human to create AI systems that meet human needs for mission results.

In October 2020, the Office of the Director of National Intelligence sponsored UTE to lead an initiative to advance the discipline of AI engineering for defense and national security. SEI had previously published 11 Fundamentals of AI Engineering and held a workshop in 2019 with thought leaders to identify areas of interest for AI initiatives. Findings from the workshop and subsequent collaborations with government, military, industry and academia led to the new pillars of AI engineering.

The government sphere has many obstacles that can prevent successful implementation of AI, such as increased scrutiny, limited data resources, a rigorous acquisition process, and high-stakes application areas. SEI government partners cited scalability challenges in the private sector, amplified by public sector barriers, as of particular concern. “It has been reported that most AI projects fail to capture the expected business value,” said Rachel Dzombak, Head of Digital Transformation at the Emerging Technology Center of SEI and leader of SEI’s work in AI engineering. “A lot of it comes from the inability to turn prototypes into systems that get the right results over time and at scale. ”

After consultation with its partners, SEI has developed its AI engineering scalability pillar, which includes three areas of intervention:

  • scalable management of data and models
  • enterprise scalability of AI development and deployment
  • scalable algorithms and infrastructure

Even highly scalable systems will not achieve mission results if they are not robust and secure. AI systems must be robust against variations in the real world, those that systems can reason about and those that they cannot. The UTE White Paper on Robust and Secure AI calls for three areas of intervention:

  • improve the robustness of AI components and systems, including going beyond measuring accuracy to measuring mission results
  • development of processes and tools to test, evaluate and analyze AI systems
  • design for security challenges in modern AI systems

While security is a must for AI implementations in DoD, so are the humans at the center. “If your smart device at home recommends the wrong song, it doesn’t necessarily have long-term effects,” Dzombak said. “But for applications and issues in the national security space, the exit from AI has consequences for human lives.”

The pillar of human-centered AI engineering aims to ensure that AI systems are built according to the ethical principles of DoD and other government agencies. “We challenge ourselves to ask ourselves what transparent and accountable systems really are,” said Dzombak, “and how to measure and ensure the integrity of the system over time”.

The Human-Centered AI Engineering White Paper highlights these areas:

  • the need for designers and systems to understand the context of use and detect changes over time
  • development of tools, processes and practices to define and facilitate the human-machine team
  • methods, mechanisms and mindsets for engaging in critical surveillance

SEI sees AI engineering as a discipline founded on the three pillars: scalable, robust and secure, and human-centered. Such a discipline would produce AI systems that not only have these qualities, but that perform their function. “By putting these pillars in place early in the design and development of the AI ​​system,” Dzombak said, “you are more likely to build systems that achieve mission results.”

Dzombak sees the three white papers as the start of a conversation. “These documents set out the open questions we see on the ground and identify gaps where work is needed. If we are to move the field forward, we must start taking steps to define and answer these difficult questions.”

With the recent creation of an AI division within SEI, the team is exploring these questions with new and ongoing AI projects, examining project portfolios for insight into the engineering of AI and preparing a roadmap for the discipline based on AI use cases. He also invites the AI ​​community to join the effort. “UTE doesn’t have all the answers,” Dzombak said. “A big part of our role is to bring together perspectives on best practice.”

The AI ​​engineering team maintains open office hours and invites those interested in collaboration, research, and the discipline in general to attend. The team also publishes an AI Engineering newsletter and can be contacted directly at [email protected].

Download the white papers Scalable AI, Robust and secure AI, and Human-centered AI from the UTE Digital Library, where you can also explore other articles, videos and presentations on UTE’s AI engineering work.

About Carnegie Mellon University Institute of Software Engineering
The Software Engineering Institute (SEI) is a federally funded research and development center sponsored by the United States Department of Defense and operated by Carnegie Mellon University. UTE works with organizations to make measurable improvements to their software engineering capabilities by providing technical leadership to advance the practice of software engineering. SEI’s CERT division is the world’s leading trusted authority dedicated to improving the security and resilience of IT systems and networks and a national asset in cybersecurity. For more information, visit the SEI website at

THE SOURCE Carnegie Mellon University Institute of Software Engineering (SEI)

Gordon K. Morehouse