Gregory Roth, incoming chair of AIAA’s new Artificial Intelligence (AI) in Aerospace Applications Working Group and aerospace engineer at the Air Force Research Laboratory’s Air Platform Concepts Branch in Ohio, was pleasantly reassured with the approach to AI at AIAA AVIATION Forum 2026. The discussion around AI in Aerospace is now clearly in full swing.
Should We Consider AI a New Tool?
Not exactly; while the Turing Test appeared in the early 1950s, a focused conversation began in 1956 with the Dartmouth Summer Research Project. Today, “artificial intelligence” is used as an umbrella term whose definition continues to evolve alongside different subcategories, including neural networks, large language models, and even conflations with digital engineering. As recently as a decade ago, before the introduction of the Transformer architecture, AI was viewed as an interesting tool with broad applicability, which meant and continues to mean different things to different communities. New terms appear and definitions shift; AI is a dynamic subject of study and series of software packages.
“As old ideas and concepts get leveraged and [interact with] new perspectives and roles, there may also be new problems. The terms are new, but the fundamental challenges and solutions may not be new.” – Gregory Roth
Many engineers, especially seasoned engineers, approach AI with some wariness and skepticism, wondering whether it is actually a novel tool or approach. Meanwhile, other groups that may know less about its history are more open to engagement. Yet with so many resources and choices, one of the greatest difficulties is knowing where to begin. With limited bandwidth, determining the right tool should take the least amount of time. It’s important to make sure people can find a path to the right approach/tool and receive the right training so they are comfortable using it.
Roth was particularly encouraged with how Lockheed Martin’s Renee Pasman framed AI as simply software in her plenary talk, “AI: Hype, Reality and Readiness in Aerospace.” He appreciated her well-grounded framing of AI as software, with its unique value as a tool that can be used throughout the process of finding a solution.
“Software is a recipe and instantiates a model. It is deterministic, not stochastic, even though stochastic processes are approximated. Popular AI use today, such as LLMs (Large Language Models), lean on stochastic approximations, but in safety critical environments, that won’t fly literally.” – Gregory Roth
In calculating risk, there are joint assessments of the likelihood of an occurrence and the impact of that occurrence. The uncertainty introduced by LLMs drive the likelihood of an occurrence up and therefore risk – often which cannot be tolerated. Conversely, decisions range in impact, and treating AI as equally valuable in all circumstances can be its own type of liability. Articulating both the scale and stakes of a decision is critical in determining how to use AI in a given situation. People will not tolerate the automation of decision-making if it results in crisis. Despite AI itself having a recipe, it is difficult to definitively set directions for use and avoidance. A rubric or set of guidelines and recommendations will be more effective.
Is Cognitive Decline a Direct Result of Using AI?
“Technology has always changed the physical characteristics of users and working animals. The horse population declined significantly with the influx of cars in the 1930s. That being said, if you’re using AI to short circuit the learning process, you don’t know what you don’t know. AI does not replace experience.” – Gregory Roth
To remain in peak shape, our brains need exercise just like our muscles. While using AI can help with many routine functions, it also tends to erode the overall achievement curve, bringing everyone closer to average, not closer to high achievement. Low achievers perform better, but high achievers can perform worse if not careful. The overall impact on the user depends on where their capability started and whether they are using AI to enhance learning or solely to solve with minimal effort. On the other hand, AI is not automatically faster if it takes longer to debug AI-written software than doing it right the first time.
AI has great potential if used wisely but can also lead to great dismay if used poorly.
