To AI, or not to AI, that is the question? Why information is not the same as expertise.
- 15 hours ago
- 8 min read
AI has made technical information available to everyone, but ultimately, it is sparking a conversation that every professional industry is about to face.
Artificial Intelligence is changing how we work. AI has quickly become part of many people's daily workflow: to draft an email, summarise a document, or look something up faster than Google would. It's genuinely impressive. Engineers can use it to organise information, explain concepts, identify relevant standards, improve productivity, and even help experienced professionals explore alternative approaches. Lawyers can use it to research case precedents or summarise legislation. Doctors can look up drug interactions or scan recent research. Financial advisers can screen investment options or model scenarios quickly.
So, in all of these cases, the AI does the groundwork, and a qualified professional steers the outcome. But there is an important distinction that is often overlooked: access to information is not the same thing as expertise.
As AI becomes more accessible, we're increasingly seeing situations where people receive technically correct information from AI but draw completely incorrect conclusions because they don't have the experience needed to interpret it.
We think it is worth discussing using AI in technical aspects, such as engineering, because it affects the quality of the work being done on your projects and ultimately, the safety and cost of what gets built.
The part that worries us
We are not saying AI is bad. At DTCE, we use AI tools too. The problem is that AI tools are treated as a shortcut through the difficult technical part of engineering work.
Building standards, design codes, calculations, geotechnical reports, loading assumptions, durability requirements, and construction constraints all interact. You cannot pull out one piece of information, look at it on its own, without understanding the wider design context, and expect to get the right answer.
Think about what it looks like in practice:
Imagine someone receives a geotechnical investigation report. Forty pages of soil profiles, bore log data, liquefaction assessments, and foundation recommendations. Without an engineering background, the client drops it into an AI tool and asks: "What does this mean for my project? What should I do?" The AI will give an answer that sounds confident, and it might even use the right terminology. But if client doesn't have the expertise to check what it is saying, there is no way of knowing whether it's right. AI does not understand your specific site conditions. It doesn't know your council's requirements. It does not know which New Zealand standards apply unless it has been told clearly and correctly. It does not have the practical judgement that engineers build from years of working with Wellington’s ground conditions on real projects. It cannot weigh the factors the way a qualified engineer can. |
Oh yeah, and don’t forget about AI hallucinations. Most people who use AI have seen this happen.
Have you ever noticed that, when you're in the middle of a conversation, you suddenly catch AI on a wrong answer or at least not quite right, in an area you actually know well and understand?
So, you push back, and the AI quickly changes direction, saying: "You're right, good catch, or you are actually right about…, apologies for the confusion…"
That is fine when you are drafting an email, but it is not fine when the answer affects a foundation, a retaining wall, or a building consent.
The same scenario shows up every day across other professions.
A patient googles their symptoms, runs the results through AI, and arrives at their GP having already convinced they know the diagnosis and treatment.
Someone receives legal advice and asks an AI whether they should follow it, without understanding what the advice is actually doing.
A financial report gets fed into a chat window, and the AI suggests an entirely different strategy, because it doesn't know the person's tax position, their risk tolerance, or history with their adviser.
The same applies in real estate. Property managers are seeing routine tenant issues turn into bigger, more expensive disputes after AI gets involved. A recent OneRoof article covered this problem in more detail. Read it here: ‘Zero to 100’: ChatGPT disputes clogging up the rental system, agents warn of price hikes"
In every one of these cases, AI produces an answer that looks reasonable on its own. The problem is that the person reading it may not be able to tell whether it is right or dangerously wrong.
AI is very good at sounding like it understands you. It brilliantly mimics users’ tone, follows their mood, and then gives the results users are looking for, which feels convincing.
But you would not trust a medical diagnosis from someone who just asked ChatGPT, right? The same logic applies in engineering, law, finance, and every other field where expertise is earned, not generated.
The real experience we recently encountered: when AI contradicted our engineers.
Recently, for the first time in our experience, we faced a case on a small residential project where AI-generated information caused a client to question part of a completed engineering design. Basically, one of our CPEng engineers was faced with a technical discussion driven by an AI agent. So, it's worth telling the story in short, because it illustrates the problem better than any hypothetical.
The client had received our calculations, drawings, and foundation design. After doing their own AI search, they came back with concerns about whether one of the specified foundation details was suitable and also suggested the option that was the opposite of what our engineers had specified. The AI response raised issues around lateral stability, wind loading, sway, and structural performance.
At first glance, those concerns sounded reasonable. The client was not trying to challenge the design for the sake of it. The client was confused and simply wanted reassurance because the AI answer had created doubt.
The issue was that the AI was responding in general terms. It did not understand the project-specific context, the design assumptions, the site conditions, or the code pathway used for that particular structure.
A statement about a foundation detail being “possibly inadequate” can be right in one situation and completely wrong in another.
The correct answer depends on factors such as:
Building weight
Wind zone
Earthquake loads
Soil conditions
Foundation layout
Code compliance requirements
The actual design system being used and lots of other important factors.
We explained to the client the context the AI had completely missed. The specified foundation design by our engineers was appropriate for that structure, that site, and that code pathway. It was not selected by default; it was selected because the project-specific information supported it.
The AI was almost certainly pattern-matching against a different question. If someone asks a broad question about whether a certain foundation type is suitable, the answer may be very different from the answer for a specific structure on a specific site.
The client then looked further into the issue and came back with more detailed technical questions. Some of those questions referred to real engineering concepts, but they still needed to be checked against the actual project. When our Chartered Professional Engineers reviewed the concerns, we found that AI had identified topics that can matter in foundation design, but it did not have the project-specific information needed to know whether those topics actually applied.
This is exactly the risk. It gives answers that sound right to someone who doesn't have the expertise to tell the difference. This case showed something we expect to see more often: a motivated layperson with AI and technical documents can generate questions that look detailed and credible, without knowing whether those questions point to a real problem.
In this case, there was no design problem. But it took several rounds of back-and-forth, detailed explanations, and significant time to explain why the AI answer did not apply to that specific project: all because a search engine produced a context-free answer that planted doubt in a client's mind.
The right way to challenge an engineering design
This might surprise some people, but we fully support clients seeking a second opinion when they have genuine concerns. In fact, one of the best ways to gain confidence in a design is through independent peer review.
We are absolutely in favour of peer review. If you have genuine concerns about an engineering design, the right step is to engage an independent Chartered Professional Engineer to carry out a formal review and issue a PS2 Producer Statement. That means another qualified engineer reviews the design and formally confirms whether it meets the required standard.
If something about a design doesn't feel right to you, a qualified peer reviewer is exactly the right person to put that question to, someone with the training to evaluate the answer, and the professional accountability to stand behind it.
What we would push back against strongly is a non-specialist using AI to challenge the fundamental design philosophy of a professional engineer. Simply because the AI doesn't have the context, the training, or the accountability to be a credible challenger. It will give you an answer that may sound specific and well-reasoned, but the person reading it may not know that it is answering a different question.
So, if you ever want more than reassurance from us, get a second opinion from another engineer. We welcome that process.
This connects to something bigger
We recently wrote an article "Why we're not the cheapest engineers", and the response showed us that many people are thinking about this.
Cheaper engineering often isn't cheaper. It's just cheaper upfront, before the Requests for Information (RFIs), the site surprises, the over-specified materials, and the re-designs start adding up.
We are starting to see the same pattern with AI. Some companies and clients are treating AI as a way to cut corners on the expertise side: generate calculations faster, spend less time on the analysis, reduce the human input.
But the problem is the same one we see with budget engineering generally: you can't take a shortcut through the part that actually requires knowledge. Of course, AI can assist with that, but cannot replace it.
In our article "Why we're not the cheapest engineers ", we explained that clients are not simply paying for calculations or drawings. They're paying for expertise, because the calculations themselves are only one part of the process.
The value is in knowing which calculations actually matter, which risks need attention, which assumptions are reasonable, which standards apply, and which solutions will work in the real world.
It is also knowing which risks can be safely accepted and which ones need to be dealt with properly.
AI can help professionals perform some of these tasks more efficiently, but it cannot replace experience built from real projects, real sites, and real consequences. The best outcomes will come from professionals who know how to use AI effectively while understanding its limitations.
At DTCE, we believe AI is an excellent tool. We use it ourselves, but like any tool, its value depends entirely on the skill of the person using it.
AI does not understand your specific site conditions and doesn't know your council's requirements. It does not know which New Zealand standards apply unless it has been told clearly and correctly. It does not have the practical judgement that engineers build from years of working with Wellington’s ground conditions on real projects and cannot weigh the factors the way a qualified engineer can. A calculator doesn't make someone an engineer. Access to legal databases doesn't make someone a lawyer. Medical websites don't make someone a doctor. And AI-generated answers don't replace professional judgement. |
What this means for your project
If you are a homeowner, developer, or builder, here is the practical takeaway:
Be careful about any engineering advice that seems to come primarily from an AI without a qualified engineer carefully reviewing it. Ask who ran the analysis? Who checked it? Ask whether a CPEng engineer actually reviewed the design and took responsibility for it. AI output that has not been checked by someone with real expertise is not engineering advice. An AI tool producing a contradictory answer is not a second opinion; it's noise without the expertise to back it up. |
Good engineering advice should be clear, checked, and backed by someone who understands the project and is prepared to stand behind their work.
If you have a project coming up, or a technical question you would like to discuss, feel free to contact us.
If you want an independent review of another engineer’s design, DTCE can carry out a third-party design review and provide clear, practical advice.
And if you want an independent review of one of our designs, engage another suitably qualified engineering firm to complete a formal design review and issue a PS2 Producer Statement.
Comments