Winning the digital race starts with intentional application over chasing trends, argues Naismiths’ Gareth Parker.
As can often happen when a buzzword enters a debate, the nuance around AI can be lost, especially when it comes to its practical applications. AI has been thrown almost immediately into the mainstream consciousness as something that is accessible without a traditional engineering or programming background.
This is similar to how the database became omnipresent when it first came onto the scene as a way of opening up a whole range of opportunities. However, with this accessibility comes a slightly homogenised view of AI that centres around generative approaches and the likes of ChatGPT, which are essentially predictive models that infer context.
While this generative AI is useful for chatbots, its uses in construction are more limited. The homogenised outlook on AI must change if the industry is to harness its full potential.
Not right for the task
Take the kind of work that Naismiths does. If we’re looking at construction data to work out the costs of a set number of projects over a set number of years, AI isn’t right for that task. You’d write an algorithm to calculate it rather than use AI to predict it.
This race to include AI in products and services is partly driven by funding, given that it is the area that is getting the most attention from investors in a tech industry that – like many others – is struggling. Real thought needs to be put into its application, particularly in the construction industry. But value can be found if that groundwork is done.
Using AI’s ability to process both vast quantities of data and natural language, one application that could be of tangible benefit is the processing of existing reports and bringing together multiple datasets to produce answers to queries. You might want to know, for example, the average completion time for all projects in London since 2010 that were five-storeys or more and received a green on their risk assessment.
An AI application could process this natural language query and produce outputs based on a range of information pulled from reports and location data within seconds.
“Recognising that things can be both similar and different, and being able to apply it in practical circumstances could be a gamechanger in helping to expand the context of searches.”
Vector embeddings
That approach to using AI is becoming increasingly common, but one exciting area that is emerging and taking these models even further is the use of vector embeddings. Simply put, a vector is a similarity of context.
To use an example of image scanning, an AI application will scan two images of faces, and determine that one person has blue eyes and one has green eyes. The AI will use vectors to database blue and green as colours and put a number against them to determine that blue and green are more similar than, say, blue and dog, because it knows that blue is a colour and dog is an animal.
That is a simplified example, but how that technology can be applied to the construction industry is where it gets interesting. Recognising that things can be both similar and different, and being able to apply it in practical circumstances could be a gamechanger in helping to expand the context of searches. It adds some grey between the black and white.
The technique used to do this is called retrieval augmented generation, or RAG. Applying RAG to our earlier query around project completion times can help counter human error in the language input, or simply expand the number and detail of responses returned.
Back to the future
The companies that create and write these types of RAG pipelines are going to be the ones that really harness the overall power of AI in the construction industry.
What is interesting about this is that a human is still involved – dictating and checking the context – and the AI is assisting by optimising queries and double-checking the input, offering suggestions and expansions if needed (known as Corrective RAG).
“Rather than replacing people, this takes away a large chunk of the “less fun” elements of our jobs, making the most of a technology that is accessible, but often misunderstood.”
These RAGs have been the talk of the tech industry in recent months, and we foresee a scenario in the not-too-distant future where there are multiple branches of RAGs that are very good at specific tasks such as location-based searches and risk assessments.
Rather than replacing people, this takes away a large chunk of the “less fun” elements of our jobs, making the most of a technology that is accessible, but often misunderstood.
Gareth Parker is chief technology officer at Naismiths
Don’t miss out on BIM and digital construction news: sign up to receive the BIMplus newsletter.
