AI and Competency
Over the past few days the RIBA has published the results of a survey on the use of AI within architecture - https://www.architecture.com/knowledge-and-resources/knowledge-landing-page/artificial-intelligence-in-architecture
At the same time and following changes to Building Regulations as a result of the introduction of the Building Safety Act, DLUHC has also published guidance on the Dutyholder requirements that apply to all buildings. This places requirements on clients, principal designers, principal contractors, designers and contractors. -
From a client perspective, there is a duty on the client to ensure that the all reasonable steps are taken to appoint designers and contractors with the necessary competence and organisational capability to fulfil their roles. But what does this mean when some 41% of Architects are reported to have adopted the use of AI in some form. This is before the consideration of other relevant design and construction related disciplines.
The RIBA suggests that AI within Architecture is generally being used in 4 key areas:
Design Ideations
Concept Designs
Text Generation - using models such as ChatGPT / Bard
Practice Archiving - where information on older projects may be being used to train AI systems on legacy projects, information libraries and stylistic methods.
While the use of AI systems is largely in their infancy, over the coming years this is likely to grow and if not carefully managed could create significant issues for not only the relevant duty holders but also occupiers and asset owners/managers.
Here we offer some headline thoughts:
Clients
Do you know how AI is being used in relation to your project?
It would be advisable to seek clarity on what and how AI technologies are being used at each work stage (RIBA 0-7) as your project progresses so that the risks can be properly assessed and managed. Given most construction projects run for many years the issue of changes in AI systems adoption and use should also be adequately captured, managed and agreed.
Do you know how AI information is being checked, updated, reviewed and approved?
What is the procedure in place for checking, reviewing, updating and correcting any AI generated information to prevent errors, omissions and mistakes from occurring. All design information that has been generated by AI Technologies should be properly identified and checked by a suitably qualified and competent person to ensure it’s accurate, buildable and compliant with the relevant regulations where that building is to be built. This could be a significantly involved task that is likely to have an impact on both programme and associated fee.
Information Ownership and Restrictions
With the advent of practice archiving are there any elements of the design that may need to be kept confidential or which are proprietary to your business such as unique reference designs, process information, GDPR related information or similar. If so how is this to be controlled? There should be provision in any appointment documentation that expressly establishes how AI is used and covers what can and cannot be used in the future by AI large learning models, including whether these are practice specific or wider global AI systems.
How realistic is a design
Where design ideas and concepts are substantially generated by AI how realistic are they to the clients needs and budgets. While it might look seductive on paper, Clients and their team should be unafraid to question the source of concept design information, its buildability, compliance with the brief objectives, future maintenance, repair and likely Regulatory compliance before committing too much potential resource to a potentially abortive project.
Designers and Contractors
If you know your using AI in the design and construction process be open about it with your clients -
It’s important that any use of AI in building design and construction is transparent and that anything generated by AI is appropriately recorded as such. As the RIBA report sets out, insurances and liabilities run with individual organisations and not AI systems. Under the Building Safety Act failures by duty holders can now result in unlimited fines and possible prison sentences. The audit trail and source of information will be key if any issues arise.
Do you have the right processes and skills to ensure information is suitable and fit for purpose.
Where the designers role may move more into checking, commenting, updating and issuing information consideration should be given to a robust process, dedicated resource and sufficient programme. Additional regular skills training/orientation may also be necessary. While some AI systems may produced designs relatively quickly these review processes may take significant time which may give rise to significant fee requirements.
What is the model being used and how is it trained.
As Google has recently found with its AI image software, Gemini, systems can show bias and create other issues depending on the information that they have been trained with. https://www.theguardian.com/technology/2024/feb/28/google-chief-ai-tools-photo-diversity-offended-users
For example in Building design if a system has been trained on largely North American Building designs, details and specifications its unlikely to be compliant with UK Building Regulations. As another example where a product or system specification may be AI generated what processes are in place for verifying the accuracy of that information particularly to ensure there are no errors in the data or questions asked to generate the response, the compatibility of products, their use and their associated certifications to show that they are fit for purpose.
Both of the above examples present a significant risk for all parties for all types of building. As identified above under the Building Safety Act failures by duty holders in relation to Building Regulations can now result in unlimited fines and possible prison sentences. In addition poorly designed and questionably compliant products and works can also now give rise to Registered Building Inspectors requesting that works are opened up and/or the issuing of non compliance or stop notices which can lead to cost and time issues and possible financial claims.
Confidentiality
In all cases where information is digitally stored, questions should be asked around what access third parties may have to that information and how it is controlled. For example where something is stored by a ‘cloud’ based service could it be quietly being used to train AI models without a practice being aware. How is confidential and proprietary client information kept secure and does that information potentially come with any GDPR implications?
The Future
By way of a bit of crystal ball gazing here are some thoughts on how AI may change the industry in the next 5-10 years:
Standard drawing referencing
ISO 19650 establishes the framework for the organisation and digitisation of information about buildings. With the advent of the Building Safety Regulator and many different individuals and organisations involved in producing information for the design, construction and management of buildings of all types its entirely possible that ISO 19650 may become a mandatory requirement for all Building Control related design, construction and management information.
Standard Classifications
Standard classification systems such as Uniclass may also become mandatory. These classification systems are already in use by significant construction clients as they permit the rapid identification of information of the same type regarding assets, characteristics, systems, products and spaces. This could aid management of a significant portfolio of different buildings.
Automated Building Control approvals
In theory, with both of the above and significant library of properly catalogued, regulatory compliant material available, plan review and approval could become automated giving a compliant/non compliant position on submitted information.
Product Rationalisation
In time and given the above there may be the opportunity for products and interfaces to be rationalised. This type of approach may comfortably sit alongside the Platform approach that is supported by the UK Government Construction Playbook.
Automated compliance checks and auditing
Systems could potentially be used to check images from works against drawn and approved design information to ensure that what is built accurately represents the approved information.
While AI can provide some useful benefits to the construction and real estate industries there needs to be an understanding of how to manage the use of this technology. While the above is only a snapshot of potential challenges that exist, with the increasing adoption of AI systems within building design we have to be both careful and mindful that all buildings should be safe to construct, to occupy and maintain and comply with the relevant regulations given the severe penalties that now exist for all Dutyholders.