Assurative AI: The Future of Safe Engineering Design

Assured AI in Civil Engineering: A Deep Dive with Dr. Michael Rustell

Key Concepts:

• Assured AI: AI systems designed not only to generate solutions but also to verify their compliance with standards, regulations, and employer requirements, ensuring safety and accountability.
• Agentic Workflows: AI systems capable of autonomous planning, execution, and reasoning, automating complex processes.
• LLMs (Large Language Models): AI models capable of understanding and generating human language, crucial for converting unstructured data into structured formats.
• Deterministic vs. Non-Deterministic Systems: Deterministic systems produce the same output for the same input, while non-deterministic systems (like LLMs) can yield varying results.
• RAG (Retrieval Augmented Generation): A technique combining information retrieval with LLM generation to improve accuracy and relevance.
• Cyborg/Centaur Approach: Utilizing AI as a tool to augment human expertise, rather than replacing it entirely.

1. The Imperative of Assured AI in Civil Engineering

The discussion centers on the critical need for “Assured AI” in civil engineering, a field governed by stringent safety and compliance standards. Unlike software development, where rapid iteration and “breaking things quickly” are common, civil engineering demands demonstrable safety and adherence to regulations for insurance, legal, and ethical reasons. Simply asking an AI to “build something” is insufficient; demonstrating the source and justification of its solutions is paramount. A single unchecked error in safety-critical infrastructure could have catastrophic consequences. This necessitates AI systems that not only create solutions but also verify their compliance.

2. Dr. Michael Rustell’s Background and Perspective

Dr. Michael Rustell, CEO of Informatic and lecturer in structural engineering at Brunel University of London, brings a unique perspective stemming from his experience navigating both the software development and civil engineering worlds. His path began with a doctorate in optimization of port terminal layouts, leading to four years of software development in maritime engineering. He then transitioned to a large engineering firm, where he faced the challenge of applying his AI skills to practical design work, highlighting the gap between research and real-world implementation. This experience led him to found Informatic, aiming to bridge the divide between engineering and AI, specifically addressing the frequent failure of AI projects to move beyond the pilot phase. He emphasizes the importance of a human-centered approach to AI, where systems assist engineers rather than replacing them.

3. The Evolution of AI Capabilities & The Role of LLMs

Rustell details the evolution of AI’s applicability to civil engineering. Previously, the reliance on structured data (tables) hindered progress. However, the advent of Large Language Models (LLMs) has been a “secret ingredient,” enabling the conversion of unstructured information (standards, reports, drawings) into structured data. This, combined with the development of “agentic workflows” – AI systems capable of autonomous planning and execution – has unlocked new possibilities for automation and efficiency. He notes the power of modern coding models like Claude Code, but stresses the need to ground these capabilities in robust data and structured knowledge representation.

4. Implementing Assured AI in Practice: Challenges and Approaches

Implementing Assured AI is not straightforward. Key challenges include:

• Understanding Standards: Engineers must possess a deep understanding of relevant standards and regulations.
• Copyright Issues: Obtaining necessary permissions to use copyrighted information within AI systems is crucial.
• Liability Concerns: Determining liability for AI-generated outputs remains a complex legal issue.
• Data Quality & Representation: Ensuring the accuracy and consistency of data used to train and operate AI systems is vital.

Informatic’s approach involves a phased implementation: starting with applying rules from a single standard, then identifying relevant standards, combining them, and ultimately creating reproducible and auditable knowledge maps. The ability of LLMs to understand unstructured human data is highlighted, but the need for deterministic elements and rigorous checking remains paramount.

5. Determinism vs. Non-Determinism & Controlling LLM Behavior

The discussion clarifies the difference between deterministic and non-deterministic systems. While LLMs are inherently non-deterministic (producing varying outputs for the same prompt), their behavior can be controlled through techniques like setting the “temperature” to zero, forcing deterministic outputs when repeatability is required. The ideal approach involves balancing determinism for reliability with flexibility for creative problem-solving.

6. The Importance of Human Oversight & the “Cyborg/Centaur” Model

Rustell and the host emphasize the indispensable role of human expertise. AI should augment, not replace, engineers. The “cyborg” or “centaur” model – where humans and AI collaborate – is favored. Human judgment is essential for validating AI outputs, identifying gaps, and ensuring that solutions align with engineering principles and real-world constraints. The ability to identify what’s missing from an AI-generated solution is a uniquely human skill.

7. Future Trends: Small Agile Firms & the Shift in Business Models

Rustell predicts a shift towards smaller, agile firms specializing in AI-powered engineering solutions. These firms will leverage AI to deliver projects faster, cheaper, and with higher quality. He anticipates a disruption of traditional engineering business models, moving from selling time to selling knowledge and expertise. The ability to build and curate AI agents, combined with deep domain knowledge, will be a key differentiator. He cautions larger firms against complacency, as smaller, more innovative companies could quickly gain a competitive advantage.

8. Practical Advice for AEC Firms

For AEC firms aiming to adopt AI responsibly, Rustell offers the following advice:

• Stay Informed: Monitor developments in software engineering, as that’s where the most significant AI breakthroughs are occurring.
• Embrace Risk: Be willing to experiment and accept that initial AI projects may not be fully successful.
• Focus on Efficiency: Start with automating low-risk, time-consuming tasks.
• Prioritize Human Expertise: Invest in training engineers to effectively use and oversee AI systems.
• Don't Die: (A playful warning to avoid being left behind by the rapid pace of AI innovation).

Notable Quotes:

• “It’s not enough to say, ‘Hey, chatbot, build me whatever.’ You need to demonstrate where information’s come from, how it’s been used.” – Dr. Michael Rustell
• “Assurative really is this concept…it’s a bit like guard rails that tell your AI what it can and can’t say.” – Dr. Michael Rustell
• “You have to have human in this because so much of what we do as engineers relies on engineering judgment.” – Dr. Michael Rustell
• “The size of the prize really is quite large…if you are interested in having some of that then you do need to move.” – Dr. Michael Rustell

Technical Terms:

• LLM (Large Language Model): A type of AI model trained on massive amounts of text data, capable of understanding and generating human language.
• Agentic Workflow: An AI system capable of autonomous planning, execution, and reasoning.
• RAG (Retrieval Augmented Generation): A technique combining information retrieval with LLM generation.
• Deterministic System: A system that produces the same output for the same input.
• Non-Deterministic System: A system that can produce varying outputs for the same input.

Conclusion:

The conversation underscores the transformative potential of AI in civil engineering, but emphasizes the critical importance of “Assured AI” – systems that prioritize safety, compliance, and accountability. The successful integration of AI requires a combination of technical expertise, deep domain knowledge, and a human-centered approach. The future of the industry will likely be shaped by agile firms that can effectively leverage AI to deliver innovative solutions, while larger firms must adapt to avoid being disrupted. The key takeaway is that AI is not a replacement for engineers, but a powerful tool that can augment their capabilities and unlock new levels of efficiency and innovation.