The Moment You Can Clone Your Best Engineer | Daniel Siegel | TEDxHochschuleBremerhaven

The Future of Engineering: AI-Powered Collaboration

Key Concepts:

• Large Language Models (LLMs): AI models like ChatGPT, Claude, and Perplexity, foundational for generating text, estimating, and controlling tasks.
• Agents: AI systems combining LLMs, context, and tools to perform specific tasks, particularly in engineering.
• Transformer Models: The architecture underlying LLMs, originating with the "Attention Is All You Need" paper in 2017.
• Modality (Geometric Understanding): The ability of AI models to understand and process geometric data, crucial for engineering applications.
• Multi-Agent Systems: Utilizing multiple specialized agents working collaboratively to solve complex engineering problems.
• Computer-Aided Engineering (CAE): The use of computer software to assist engineers in design, analysis, and manufacturing – the foundation of modern engineering tools.

---

I. The Shift Towards AI-Assisted Engineering

The video highlights a paradigm shift in engineering, moving from traditional methods to a collaborative model where engineers work alongside AI “digital co-workers.” This isn’t a question of if this will happen, but when. The speaker, Daniel, draws parallels to the transition from drawing boards to Computer-Aided Engineering (CAE) over 30 years ago, noting that current engineering tools, while updated, still operate on a fundamentally outdated foundation. While previous technological advancements like the internet and cloud computing improved engineering processes, they didn’t fundamentally alter how engineers worked. The rise of LLMs, however, represents a true inflection point.

II. The Rise of Agents: LLMs + Context + Tools

The key breakthrough, according to Daniel, began in 2017 with the publication of the “Attention Is All You Need” paper, birthing the transformer models that power LLMs. While AI existed in engineering before, LLMs offered a new level of “general intelligence” – the ability to estimate, generate, and even control and execute tasks. However, LLMs alone are insufficient for engineering. The emergence of “agents” in 2023 bridged this gap.

An agent is defined by three core components:

1. Large Language Model (LLM): Provides the reasoning and language processing capabilities.
2. Context: The specific knowledge base relevant to the task at hand.
3. Tools: Access to engineering software and functionalities (e.g., CAD software, simulation tools, manufacturability analysis).

The ability to connect LLMs to engineering tools is the critical advancement, allowing them to manipulate geometry, analyze simulation results, and assess manufacturability.

III. Improving Agent Quality: Model, Context, and Tools

Daniel, referencing insights from his colleague Rahm, outlines a formula for improving agent quality: Quality = Model + Context + Tools. Each component requires focused development:

• Model: While foundation models are rapidly improving (new models published almost monthly), they currently lack the ability to understand geometry – a crucial modality for engineering. However, research in spatial geometry is accelerating, promising future models with this capability.
• Context: A significant challenge lies in accessing and utilizing the vast amount of engineering knowledge currently residing within companies and in the minds of experienced engineers. Investment in prompting strategies, memory functionalities, and knowledge capture is essential.
• Tools: Existing engineering tools are often outdated and not “agent-ready.” New technologies are needed to bridge the gap and provide LLMs with access to these functionalities, ideally through standardized protocols for scalability.

IV. From Agents to Multi-Agent Systems: A New Paradigm

The evolution continues with the development of multi-agent systems. The principle is to create specialized agents, each with a specific model, context, and toolset, tailored to a particular task – mirroring the organizational structure of many companies. These agents then communicate and collaborate to solve complex problems. Daniel describes experiencing a “goosebumps moment” witnessing these systems in action, with agents autonomously performing engineering tasks and coordinating with each other. Companies are already building and deploying these “digital coworkers.”

V. Timeline for Adoption: Lessons from Computer Science

To estimate the timeline for widespread adoption, Daniel draws a parallel to the evolution of programming. He notes that it took approximately two years for AI models to reach a level of proficiency in coding comparable to basic auto-completion, and now AI-native companies are using these agents to develop over 50% of their production code daily.

He presents a graph illustrating the relationship between skill level (normalized based on benchmarks) and time. Starting with ChatGPT at 30% skill, the introduction of agents provided a slight increase, and granting agents access to engineering tools resulted in a significant boost.

He further highlights the adoption rate of the internet (14 years to reach 800 million weekly active users) compared to ChatGPT (2 years to reach the same number), suggesting that AI adoption could accelerate rapidly.

VI. Real-World Example: Wheel Design Collaboration

The video begins with a practical demonstration of AI-assisted engineering. Daniel and a colleague collaborate on designing a 20-inch off-road wheel. The process showcases:

• Requirement Specification: The colleague outlines desired characteristics: lightweight, rigid, impact-resistant, balanced comfort/aerodynamics/strength.
• Design Generation: The AI (represented by Daniel’s digital assistant) presents nine design options.
• Iterative Refinement: The colleague selects a design and requests modifications (adjusting an area and twisting spikes).
• Automated Updates: The AI instantly generates an updated 3D CAD design.
• Simulation & Analysis: The AI performs simulations based on specified loads (cornering, pothole impact, braking) and confirms the design meets safety standards with a safety factor of 3.2x.
• Prototype Ordering: The AI automatically requests quotes from suppliers and places an order based on delivery time and budget.

This example demonstrates the potential for AI to streamline the entire design process, from initial concept to prototype production.

VII. Conclusion: Embracing the Future

The video concludes with a call to embrace the future of engineering, emphasizing that the question isn’t if AI will become integral to the process, but when. The rapid pace of development suggests this future is closer than many anticipate. The final scene shows a rendering of the newly designed wheel, symbolizing the tangible results of this collaborative approach. Daniel’s closing statement, “Let’s embrace the future together,” underscores the transformative potential of AI in engineering.

Notable Quote:

“So, the question is not if this becomes a reality, but when it becomes a reality.” – Daniel, highlighting the inevitability of AI integration in engineering.