General-purpose AI chatbots may be useful for everyday tasks, but their limitations become more serious when engineers rely on them for production decisions. Maor Farid, co-founder and CEO of Leo AI, argues that engineering-grade AI must meet much higher standards for accuracy, transparency, and technical understanding, tells Machine Design.
One fundamental problem is that general AI models do not truly understand CAD geometry. They typically process text and images rather than the boundary representation, topology, tolerances, mating constraints, and feature history that define an engineering model. Without access to this information, conclusions about dimensions, clearances, or interference may depend on incomplete descriptions.
The quality of technical sources creates another challenge. Engineering decisions rely on validated textbooks, standards, materials databases, internal design guides, and documents from organizations such as ASME and ISO. General models trained largely on internet content can mix authoritative information with unreliable material.
Farid identifies five requirements for engineering-grade AI. Systems should understand native CAD geometry, use curated engineering information, cite sources for technical claims, show formulas and assumptions behind calculations, and recognize when available information is insufficient to provide a reliable answer. Data security is equally important, particularly when companies work with proprietary designs.
Engineers can test AI tools by asking questions with intentionally incorrect assumptions, verifying citations, providing actual geometry, and asking questions that cannot be answered with the supplied information. A trustworthy system should identify errors and request additional data rather than produce a confident guess.
Even AI systems that pass technical tests cannot replace engineering judgment. Calculations and information retrieval are only part of engineering work. Engineers must still determine whether assumptions make sense, whether specifications are appropriate, and whether designs reflect manufacturing realities.
The article ultimately presents AI as a supporting engineering tool. Its value lies in reducing time spent searching for information and repeating existing work while keeping critical decisions in engineers’ hands.