The next AI leap: From language to real-world engineering

In a significant evolution of artificial intelligence, Dassault Systèmes is spearheading a shift from generative AI focused on language to a more robust application in engineering. This transition is underscored by the need for AI to not only provide answers but also ensure that those answers are applicable in the real world. Gian Paolo Bassi, Senior Vice President of Mainstream Innovation and Customer Role Experience at Dassault Systèmes, emphasizes that while large language models can generate information quickly, they fall short when it comes to the complexities of engineering tasks that require a deep understanding of physical principles.

The limitations of current generative AI become evident when comparing simple text tasks with engineering challenges. For instance, while a language model can succinctly summarize a technical standard, it struggles to assess whether an antenna in a wearable device will maintain a reliable connection to a router situated 100 meters away. Bassi articulates this gap, stating that large language models, by their very design, cannot adequately represent the complexities of the physical world.

The Need for Physics AI

In engineering, the stakes are higher. Beyond retrieving information about product behavior, engineers must understand how a product will function in practice, which necessitates knowledge of physics, materials, and geometric constraints. This is where Bassi believes the next major advancement lies: the development of what he refers to as physics AI. This approach aims to enhance the capabilities of AI by integrating it with a comprehensive understanding of the physical environment.

Simulation technologies have already enabled engineers to analyze and test products before they are physically created. Building on this foundation, Dassault Systèmes is introducing the concept of Industry World Models. These models provide AI with a virtual representation of not only a product but also the environment in which it operates. For example, an autonomous robot in a factory must navigate its surroundings, which include machines, moving goods, and personnel. To function autonomously, the robot requires real-time information about these elements, illustrating the need for context-aware AI.

Bassi elaborates on this context-driven approach, noting that different systems have different 'worlds.' For a robot, the factory is its operational realm, while for a transportation system, the city serves as its environment. This distinction highlights the importance of situational awareness in AI applications within engineering.

The Evolving Role of Engineers

As AI technology advances, a pressing question arises: what will become of the engineer's role? Bassi argues that rather than diminishing the engineer's position, AI will make it more engaging. He identifies curiosity as the defining trait of a successful engineer—an eagerness to explore various solutions to complex problems. Historically, time has often been the limiting factor in this exploration.

Bassi draws parallels between the current AI revolution and previous technological advancements that have transformed engineering practices. Just as 3D CAD revolutionized design comprehension beyond what 2D drawings could offer, AI has the potential to facilitate more interactive and conversational engagements with specialized engineering tools. He asserts that engineers will not be replaced by AI; instead, those who can effectively leverage AI tools will outpace those who rely solely on traditional methods.

To implement this vision, Dassault Systèmes is introducing three specialized AI companions: AURA, LEO, and MARIE. AURA focuses on enterprise knowledge, handling requirements and specifications. LEO is tailored for engineering and manufacturing contexts, assisting with design and simulation tasks. MARIE applies scientific expertise in fields such as materials and chemistry. Bassi provides a practical example involving a hospital robot, where AURA could analyze compliance requirements, LEO could evaluate design adherence, and MARIE could address material suitability for sterilization processes.

Data Accessibility and Security Challenges

For these AI systems to be effective, they require access to company data, which often includes sensitive intellectual property. This necessity creates a tension between maximizing AI utility and safeguarding proprietary information. Bassi emphasizes the importance of technical safeguards and sovereign infrastructure as companies grant AI deeper access to their engineering environments.

He poses a critical question for business leaders: "Where is your data?" This inquiry underscores the idea that if AI cannot access relevant data, its potential value diminishes significantly. Bassi's message to engineers is equally urgent: they should begin experimenting with AI technologies without delay. However, he cautions that this shift involves more than merely learning to interact with a chatbot; it requires creating integrated engineering environments where data, simulation, scientific models, and AI collaboratively function, with engineers asking the right questions to guide the process.

As Dassault Systèmes continues to enhance its 3DEXPERIENCE platform with AURA, LEO, and MARIE, the company is positioning itself at the forefront of a new era in engineering. These AI-powered virtual companions, set to be globally available in 2026, promise to redefine the relationship between engineering and artificial intelligence, moving beyond traditional boundaries into a future where AI and human ingenuity work hand in hand.