06 Jan The human behind the machine

Anis Jayaram, Fire Engineer, Jensen Hughes, looks at how today’s fire engineers are blending the world’s most advanced AI tools with traditional expertise to reshape fire safety, while keeping trust, ethics and responsibility at the heart of innovation.

The first time I came across the term “artificial intelligence (AI)” wasn’t in a classroom or research paper, but within the pages of Isaac Asimov’s science fiction. Those stories sparked a fascination that grew as the world around me changed. What began with printed newspapers and thick encyclopedias slowly shifted to the boundless expanse of the internet, where knowledge was just a click away.

Fast forward to the end of 2022, when OpenAI released ChatGPT to the public, and we witnessed a phenomenal rise in interest in AI. Since then, people have been eager to integrate AI into almost everything, from complex industrial systems to something as simple as a toothbrush. We are now in an era where adding the term AI has become a way to upsell products and services. Interestingly, Google’s AI Principles and Responsible AI documentation includes a thoughtful reminder:

“If a simpler rule-based system can achieve comparable performance, AI should not be used.”

Attention is all you need

Let me pause for a moment to explain a few basics of AI. Behind all the current trends lies a long journey of progress in computer science and advanced statistics.

There are many types of architectures in the world of AI and machine learning, but at the heart of it all is the simplest unit, the neuron. Much like a brain cell, it forms the fundamental building block of a structure known as an artificial neural network (ANN).

Step 1: Takes inputs
Imagine you are trying to decide whether to go out for a walk. You might think about a few things: Is it sunny? Is it too hot? Do I have free time?

Step 2: Converts inputs to numbers
Each of these questions is considered an input. In an ANN, these inputs are represented as numbers that the system can process.

Step 3: Applies weights
Not all inputs are equally important. Maybe the weather matters more than how much free time you have. The neuron assigns each input a weight, which is a number showing how important that input is in making the decision.

Step 4: Adds them up
The neuron adds all the weighted inputs together, similar to calculating a weighted average.

Step 5: Makes a decision (activation)
The neuron then passes this total through a simple rule known as an activation function. This function decides what happens next. If the result crosses a certain threshold, the output might be “Yes, go for a walk.” If it does not, the output might be “No, stay home.”

We have since advanced to far more complex models, yet ANNs remain important. In fact, many of the problems we deal with in the real world can still be reduced to some form of regression.

For readers who are curious to learn more, I recommend the YouTube channel 3Blue1Brown, which beautifully explains these concepts in a way that is engaging and easy to follow.

So as fire engineers, how can we use AI to our advantage? How can it make our work easier? And perhaps most importantly, how can we use it responsibly and ethically?

A field at the crossroads

Fire safety engineering is still an evolving discipline. I would like to say there is a certain degree of engineered anticipation, predicting how a fire develops, how people behave and how systems respond. Yet, despite decades of progress, many of our tools remain reactive. We have long relied on rules, deterministic models and prescriptive codes that are robust but often limited by assumptions. With the integration of AI, we now have the potential to refine these processes and create far more efficient workflows.

Before exploring some of these tools, it is important to note that due diligence should always be exercised when using AI systems, particularly to ensure compliance with your respective corporate IT and data security policies.

One efficient workflow for prescriptive compliance involves the use of Retrieval Augmented Generation (RAG). This framework combines a large language model with an external knowledge retrieval mechanism. In simple terms, you provide the model with reference data such as fire codes and standards and then ask it targeted questions. This allows it to find relevant clauses or identify specific exceptions quickly and accurately. A practical way to experiment with this is through any LLM based Notebook, where you can upload a document and query it directly.

There are also more subtle applications of AI that can improve everyday productivity. Many collaboration tools such as Microsoft Teams now integrate plugins that can transcribe meetings automatically. While the built-in transcriber works, other third-party tools can also generate accurate transcripts and produce minutes of meetings automatically. Similarly, AI-based writing assistants can adjust the tone of a document or correct grammar, helping non-native English speakers communicate more effectively and professionally.

Coding used to be sought after skill that very few people possessed, but recent this can be largely developed using AI , this may not be consumer ready code, but it helps us engineers create nifty tools. Personally, I have developed a bunch of small, specialised tools that can run calculations that would normally take hours, and I consider myself a below average coder. Not understanding a programming language shouldn’t prevent engineers from developing tools; understanding the methodology and its limitation is far more important than coding.

Advanced uses of AI in fire engineering

Some of the more advanced uses of AI in our field involve computer vision, which allows a computer to understand and interpret visual information. This capability can be very useful for reviewing plans, calculating travel distances, and performing prescriptive compliance checks. These tools are mostly proprietary, and help engineers complete reviews faster and with greater accuracy.

There are also AI-assisted tools that support computational fluid dynamics (CFD). Most current progress in this area comes from the finite element analysis (FEA) domain, and only a few practical AI tools exist for fire-specific CFD. However, this is an evolving field with significant research and development underway.

The list of available AI tools is non-exhaustive. If we were to name every product that uses or claims to use AI, it would fill an entire issue on its own. The real value lies not in the number of tools available but in how we integrate them effectively into our workflow. Success with AI depends on thoughtful integration and continuous learning.

Here are a few guiding principles for applying AI in fire engineering practice:

1. Start small: Begin with repetitive or data-heavy tasks, the ones that are usually time-consuming.
2. Validate everything: Always benchmark AI outputs against peer-reviewed work or established reference models.
3. Document assumptions: AI models can produce incorrect or misleading results, so treat outputs as experimental and record all assumptions to avoid creating a black box.
4. Collaborate early: Involve engineers who understand AI early in the process and incorporate their insights into the framework.
5. Educate your team: AI literacy is now as essential as code literacy. Training teams in AI and responsible use is crucial in today’s environment.

How to be responsible?

To use AI effectively in fire safety, we must apply the same structured discipline that we expect from any safety-critical system. As I mentioned earlier, responsible use of AI is not only about innovation but also about control, validation and trust.

1. Transparency: AI models must explain why they predict a certain outcome, not just what the result is. Explainable and auditable AI is essential for accountability.
2. Validation and verification (V and V): Every AI model should face the same level of scrutiny as a CFD or egress simulation. It must include traceable data sources, validation reports and benchmark comparisons.
3. Ethical deployment: Fire data and general arrangement drawings often contain sensitive information. Compliance with privacy and data protection laws is our responsibility.
4. Integration, not isolation: AI tools should connect with existing workflows to be effective. They should enhance engineering practice rather than exist as isolated experiments.
5. Continuous learning: AI systems must evolve as codes, materials and human behaviour change. A model that performs well today may become outdated tomorrow if it is not regularly updated and retrained.

The engineer of the next decade will not be replaced by AI, but by another engineer who knows how to use it. Just as computer-aided design replaced hand drafting and computational fluid dynamics replaced empirical charts, AI is becoming an extension of professional intuition. Today, we already have tools that can check compliance across multiple codes, suggest optimal sprinkler layouts and forecast fire development through probabilistic risk assessment. These are not distant concepts; they exist now, though often in prototype form. The challenge is not technology itself, but trust and adoption.

This feature appeared in issue 68 of Fire Middle East magazine.