Standards, Regulations and Earthquake Research (TS/EN/TBDY) — Artificial Intelligence in Civil Engineering

Gains:

Ability to reduce a design question to relevant standards and regulations and use AI as a research map
Ability to verify the item numbers, coefficients and values cited by AI from the official and current text
Ability to recognize the risk of item and value hallucination of the language model and confirm each standards-based output from its source

Civil engineering is based on standards and regulations: TS 500 (reinforced concrete), TS EN 1993 (steel), Eurocode series, TBDY (Turkish Building Earthquake Regulation), TS 498 (loads), Ministry of Environment and Urbanization unit prices and hundreds more. An engineer spends a significant part of the day searching for answers to the questions "which substance, which coefficient, which limit value?" AI can be a powerful accelerator in this research: reminding what standard a topic is covered in, explaining a concept, summarizing the relationship between items. But there is a critical danger here: language models misremember or make up (hallucinate) item numbers, coefficients and values. In this unit we will see how to use AI as a “probe map” and verify each value from the official text.

Using AI as a Research Map

The real value of AI in standard research is not to give you the answer, but to show you where to look. A good use case works like this: you ask the AI about the topic, it tells you the relevant standard, possible clauses and concepts; You go and read those articles from the official text and confirm them. Think of the AI like a librarian: it points to the right shelf, but you read the book and verify the quote.

The most productive questions are:

"Which standard/regulation regulates this issue?" (wayfinding)
"What does this concept mean, what parameters does it depend on?" (learning)
“What is the difference between these two approaches?” (comparison)
"Can you explain this item in plain language?" (interpretation)

The riskiest questions are those that directly ask for numbers and item numbers; Never use these answers without confirmation.

Another powerful use is to give real text to AI and have it interpret it. For example, if you copy the official text of a standards clause and say "explain this in plain language, but don't change any of the values in it," the AI will interpret the correct text in front of you instead of producing a hallucination. In this approach, since the source comes from you, the risk of error is much reduced; AI only serves to understand and summarize. Same logic; It also applies to tabulating a section of regulation, comparing two clauses, or breaking a complex sentence into steps. In short, instead of "making AI ask the value", it is a much safer strategy to "make AI interpret the correct value".

Caution: An item number provided by AI (such as "according to Article 7.3.2 of the TBDY...") may be real, made up or false. The fact that the item number appears correct does not indicate that its content is correct. Each item and value must be read verbatim from the official and current regulation text.

Recognizing and Breaking Hallucinations

Hallucination (AI producing non-existent information as real) is the number one risk of standards research. Typical symptoms:

Very specific but unconfirmed item numbers.
"Current" version information (AI's training data may be outdated; regulation may have been updated).
Values without hesitation, confident but without resources.
Confused coefficients between standards (such as attributing the TS 500 value to Eurocode).

The way to break the hallucination is to not burden the AI with resource responsibility. Take the "where should I look" from it, take the "exact value" from the official text.

Weak Prompt / Strong Prompt

WEAK: "What is the building importance coefficient according to TBDY?" (AI can make up a single number; there is no context and confirmation.) STRONG: "What subject does the building importance coefficient I depend on within the scope of TBDY (Turkish Building Earthquake Regulation) and in which section is it regulated?- Explain the concept in plain language - Explain in general which building classes it changes according to - Tell the name of the relevant table/section so that I can read and confirm it MYSELF from the official text. Do not make up the exact numerical values; Show me where to look. Also note that your training data may not be up to date."

This prompt keeps the AI in a guiding position and reduces the risk of hallucinations.

Verification Workflow

For each standards-based AI output, follow this workflow:

step

to be done

Purpose

Get standard name and relevant part from AI

wayfinding

Get the official/current text (published standard)

Reaching the source

Read the item and value verbatim from the text

Confirmation

Check if the version/date is up to date

Currentness

Apply the value to the account and check the order

Consistency

If critical, confirm with second engineer

Approval

This flow leverages the speed of AI while tying the risk of error to the engineering discipline.

Tip: Be sure to pay attention to the current version of the standard. Regulations change (e.g. earthquake regulations have been radically updated over the years). The version the AI "knows" may be outdated. Verifying the current official version is mandatory for the approvability of your project.

Three Mini Cases

Case 1 – Made-up substance. The engineer asks the AI for a minimum reinforcement limit. "According to TS 500 Article 11.4.7, the rate is 0.3%," says AI. The engineer opens the standard: the relevant item number is different and the value varies depending on the section condition. AI gave both the number and the constant value incorrectly. Official text confirmation avoids the error.

Case 2 – Old version. An engineer requests an approach from AI for earthquake load calculation. AI explains the logic of the old regulation that has been repealed. The engineer realizes that the approach has changed according to the current regulation in force and calculates it according to the current text. Freshness check saves a project that was about to be rejected.

Case 3 – Correct use. The engineer does not know where to look for a fire resistance requirement. AI tells you which title the issue is under in the relevant standard and regulation. The engineer goes directly to that section and reads the value from the official text. AI saves time, the decision is made from the source.

Case 4 – Standard confusion. An engineer asks AI about material coefficient in a reinforced concrete calculation. AI mixes a coefficient of Eurocode 2 with the representation of TS 500 and gives it. Since the security philosophy and coefficient definitions of the two standards are different, this mixture leads to errors. The engineer determines from the beginning which standard he works according to and confirms the value from the text of that standard. Lesson: choose a single standard and run the entire account consistently within that standard; Particularly audit where AI crosses standards.

Copiable prompt templates

STANDARD WAYFINDING PROMPT: "Which standard and regulation regulates the following subject in Türkiye/internationally: [subject]. - List the relevant standard names and general section headings - Explain which concepts/parameters the subject is related to - DO NOT give the exact article number and number; show me where to look - Remind me that the training data may not be up to date."

CONCEPT SIMPLIFICATION PROMPT: "Explain the following standard/regulation clause in plain language, but DO NOT CHANGE any numerical value or limit in it; just explain its meaning. Mark where you are unclear, do not make it up. Text: [paste the clause]"

UPDATE / COMPARISON PROMPT: "Specify the date on which the version you know of [standard/regulation] may be dated and the current version in force may be different. Tell me what may have changed at a general level between the old and new approach; I will confirm the exact values MYSELF from the official current text."

Common mistakes

Using the item number and value given by the AI without confirming it from the official text.
Not checking the current version of the standard; Calculate according to the old regulation.
Substituting the confident language of AI for resources.
Not noticing the coefficients that interfere between standards (TS/EN/EC).
Directly asking for the "exact number" and paving the way for a hallucination.
Basing a critical value on a single source (AI) and not making a second confirmation.

In summary

Use AI as a “research map” in standard research, not an “answer machine.”
AI is powerful for navigation, learning, and comparison; risky for exact item and value.
Hallucination is the biggest danger; Verify each item and value verbatim from the official, current text.
Be sure to check the current version of the standard.
Follow the verification workflow (standard name → official text → confirmation → recency → rank → approval).
Do not base critical values on a single source; Confirm with second engineer if necessary.

Application task

Choose a standard/regulatory question from your own work (a minimum limit, a coefficient, a load value). Ask the AI this in “research map” format: which standard, which section, ask for concept explanation; Tell him/her not to make up the exact number. Then: (1) find the part the AI points to from the official text, (2) confirm the value verbatim, (3) verify that the version is up to date, (4) compare whether the AI matches what the official text gives. If you find a discrepancy, save it.

checklist

[ ] I used AI for navigation/learning, not for exact number.
[ ] I determined the name and section of the relevant standard.
[ ] I confirmed the item and value verbatim from the official, current text.
[ ] I checked the current version of the standard.
[ ] I compared the AI output with the official text.
[ ] I checked that there is no coefficient confusion between standards.
[ ] I confirmed the critical value with a second source/engineer if necessary.