Implementing AI in structural engineering

You should start by being ready to provide a statement on what your organisation is using AI for. There are three scenarios for the use of AI, you should prepare statements for each of the following scenarios as relevant:

1. If you are developing your own AI tools for internal use, and the direct outputs are to be used internally. Such as tools to help engineers find codes and standards, or tools to help generate finite element models.

2. If you are developing AI tools where the outputs are going to be used directly by other stakeholders. Such as tools to assist in the creation of design reports, creation of BIM models, or specification, or calculation documents.

3. If you are developing your own AI tools for the use of other stakeholders. Such as tools for architects or MEP engineers, to aid the design process, or tools for clients to monitor the health of their structures, or developing free to use/open-source tools for use by other structural engineers.

Be sure when commencing work on any projects that your organisation’s AI policy statement aligns with those of other stakeholders. You can do this by making sure your clients and other stakeholders (such as architects, MEP engineers, occupiers/residents, Building Control authorities, façade engineers, etc.) have a copy of your organisation’s AI policy.

When setting your organisation’s AI policy, focus on transparency, alignment with your goals, and regular reviews. Clearly state how AI supports your objectives and ensure stakeholders understand its purpose and limitations. Periodically review and update the policy to address new risks, technologies, and regulations, keeping it practical and adaptable for your needs, even if full compliance with ISO 42001 is not the goal.

1.1      The risks and opportunities that come from using AI

Any organisation using AI should consider these risks, and how to implement actions into the AI process. As well as evaluating the effectiveness of these mitigation actions. The risk and mitigation statement should be included (either in the main body, or appendix) of your digital policy statement.

You should be performing, conducting, and continuously assessing risks through risk assessments, and showing how these risks can be mitigated. You should include risk and mitigation assessments within your AI policy, and set planned intervals for updating and reviewing your AI policy to reassess the risk—this may be quarterly or even annually depending on the needs of your organisation. Implement a straightforward process to identify and manage risks associated with AI use. You should develop a checklist or risk matrix to evaluate potential issues like structural safety, data security, and project delays.

1.1.1     Safety:

A risk associated with structural safety is that an AI designing structural elements may make an error. This could be mitigated by having a senior engineer review any designs produced.

Storing information from historic projects on a single database increases the risk of that information being hacked. The risk of your business being targeted by malware is potentially higher if you work on projects where knowledge of the layout of the structure could be used by criminals. A mitigation for this would be to not use these AI tools for projects such as nationally critical infrastructure, rail bridges, government buildings, banks, hospitals, main road or motorway bridges, power stations, or projects with non-disclosure agreements.

There is a risk that your dataset could become poisoned (poisoned refers to a dataset that has been corrupted by low quality or inaccurate data) and future results may be unreliable as a result. A mitigation against this would be to ensure that any datasets have a version control system in place so they can be rolled back to a previous working version if poisoning occurs.

There is also a risk to your business’s ability to grow in the future. If you are using AI to do many of the tasks historically done by graduates, you may be deskilling your younger engineers. A mitigation for this would be to ensure that these tools are not available to the most junior engineers until they have developed and demonstrated the competence to understand the process themselves.

If AI reduces the number of people in a design team, this is likely to increase the burden on individual members of staff and reduce the support network available to them. How can you ensure your staff have sufficient support so as not to affect their mental health?

1.1.2     Sustainability:

The impact of poor-quality designs from using historic datasets. If the majority of the structural elements in your dataset have a high carbon footprint, it’s unlikely you will get low carbon footprint design outputs.

The impact of data servers running constantly. Machine Learning AI tools often require huge datasets before they produce reliable results. Question whether the task you are asking AI to do is simple enough that it could be done without it.

This may also include societal impacts. If you are using AI to replace parts of your workforce, such as graduates or junior technicians, what is the impact on the profession?

Consider the broader implications of using AI, such as environmental sustainability, workforce impacts, and fairness in decision-making. Ensure that AI systems are designed and deployed in a way that minimizes harm and maximizes benefits to society. Regularly evaluate the societal and ethical impacts of AI use to align with industry standards and stakeholder expectations, maintaining the profession’s integrity and reputation.

1.1.3     Legal:

There is a short but growing list of prohibited uses of AI. This is dependent on laws of the area you are working in. You should check you comply with the laws in the area you, your project, and your client are based in.

There also needs to be wider consideration on data ownership. If you collate data from all your projects you may be accidentally violating copyright or GDPR agreements. You should be upfront about what information you are collating, what you are using it for, and how long you plan to store it.

If, for example, you are using AI to convert historic drawings to modern formats, you need to consider: are these historic drawings yours to keep? External AI tools might automatically add this information into their own databases—is it yours to give away?

Another example would be if you use AI to analyse structural elements. If you collate the information to expand your own database, you need to be sure that you own that information, and that the ownership does not sit with your client.

A legal requirement in many jurisdictions is that design is carried out by “competent persons.” If you are using AI to undertake design, are you meeting this legal criteria? Consider also AI training data books / reference material. Do you own the copyright to this? Are you invalidating it by uploading it to a large language model (LLM)?

Within your business, the roles and responsibilities should be clearly defined, allocated, and visible points of contact. In larger organizations this may mean having multiple roles and responsibilities for different AI applications. However, in smaller engineering companies this may mean having an AI lead.

A method for reporting and sharing concerns around the use of AI should be put in place, especially regarding concerns around structural safety. These may include ways to report anonymously, but should include mechanisms to shut down the use of unsafe AI tools, and track projects where errors connected to these reports may have not been picked up.

Make sure that you have the resources you need to implement AI successfully. The resources you need are going to evolve over time. This is discussed in the life cycle section below. The resources needed should be documented both now, and with a plan of what is needed in the future.

There are a number of areas to consider including data, software and hardware, and people.

2.1      Data sets:

What information will you need to implement your AI systems? If you are collating your own data, you need to think about how you will do this consistently across your organisation.

If you are using client (or other stakeholder) data, you need to make sure you have applicable data retention and disposal policies. You also need to make sure you have their permission to store this data. You may need to put it into a project-specific repository, especially if there are any privacy or security concerns regarding the use of this information.

2.2      Software and Hardware:

Collect a list of the pieces of software or hardware you will need to acquire, use, and maintain. Define what purposes each piece of software is being used for. If you are creating your own AI programmes, document what algorithms, data conditioning, optimisation, and evaluation methods you are going to use, and keep this list updated. Check the license conditions for any packages or open-source software you are going to use.

Two key questions to ask yourself:

• Are you going to run this software locally or remotely? Do you need new computers, hard drives, or routers?

• Are you going to store the data on the premises, or use cloud-based storage? (e.g. SharePoint, OneDrive, Google, Amazon, etc.)

2.3      People

2.3.1     Privacy and security:

It is important to set out who will have access to the information you collect. Consider who in the organization has access to this information:

• Structural/Civil Engineers, Technicians — companywide or specific design teams?

• Does your company do wider research — is it likely to be used by data scientists or data engineers in the future?

• Does your company partner with academics — will they have access to the data?

You should be clear about who will have access to the datasets. Datasets that have uncontrolled access could allow someone to recreate the structural plans with malicious intent.

2.3.2     Competence and safety:

Is there sufficient knowledge and training available for the structural engineers in your company to understand the outputs these tools produce and, more specifically, how they produce them? (AI algorithms are not typically taught on civil and structural engineering degree courses, so how are you guaranteeing an acceptable level of knowledge?)

Invest in building your team’s understanding of AI through accessible training or educational resources. Identify an internal AI lead or point of contact to coordinate training efforts and address AI-related questions. Encourage knowledge-sharing within your organization to ensure engineers understand how AI tools function and their implications for design safety and decision-making. Continuous learning will strengthen your organization’s capacity to use AI responsibly and effectively.

Approach AI implementation as an evolving process. Define a lifecycle management plan that outlines how AI tools will be selected, maintained, and eventually retired. Periodically assess the suitability of tools to ensure they meet current standards and project needs. By taking a structured and forward-looking approach, your organization can maximize the value of AI while minimizing risks of obsolescence or misalignment.

Like any piece of software, often you will know how long you plan to use it for. AI is no different. You should consider how you will implement it, how you will use it day to day, how you will retire it, and if/when it’s no longer required or needs refreshing.

To support AI implementation, develop internal tools and documentation that address key elements like policies, risk management, and data governance. Ensure these tools are adaptable and scalable to fit your organization’s size and needs. Focus on creating resources that provide clarity and consistency in your approach to AI, helping your team operate efficiently and responsibly.

The organisation should document the responsibilities and development plan for these decisions. This also ties back to managing resources — are they up to date with the current usage, are the resources still fit for the purpose they are being used for?

For example, if you are using a piece of AI software to design structural elements to the current first-generation Eurocodes, this may need to be retired when the second generation of Eurocodes comes into effect.