AI specialist Alan Mosca employs the help of ChatGPT to explore the background and potential uses of generative AI.
DISCLAIMER: I used ChatGPT to help me write small parts of the introduction (I reviewed its accuracy and added my own views) - I wanted to see what it knew about itself!
The recent rise of generative artificial intelligence (AI) has been nothing short of remarkable.
The best example is ChatGPT, which has made many of us realise that AI is here to stay.
From answering complex queries to creative writing, ChatGPT is a useful tool for businesses and individuals alike.
However, the impact of AI isn’t limited to the world of online chatbots and virtual assistants.
This technology is revolutionising entire industries, including construction and engineering, and its potential continues to grow every day.
Generative AI can create new content, such as text, images, music, and even plans and designs, by learning from existing patterns extracted from large datasets.
Unlike other types of AI trained to recognise existing patterns, it generates entirely new and unique content.
This technology has wide-ranging applications, from assisting in creative tasks like writing and design, to automating complex tasks like predictive maintenance and optimisation.
Generative AI before ChatGPT
ChatGPT isn’t the first example of generative AI to emerge.
In 2014, Generative Adversarial Networks (GANs) were introduced as a generative AI that could create new images from scratch by learning how to generate realistic fake data.
Variational Autoencoders (VAEs) were introduced around the same time. These work by learning underlying “latent” (meaning invisible or implied) data that can be used to generate new similar content.
Transformers, used in natural language processing, have also been adapted for generative purposes, enabling the creation of coherent, high-quality text. GPT is a transformer.
Large language models (LLMs) like BERT and GPT have taken generative AI to new heights of popularity by producing human-like language on a massive scale.
Types of Generative AI
Generative AI has many ‘modalities’, each describing the type of data used as input and output of a model.
- Text-to-text: recently popularised by language models like ChatGPT
- Text-to-image: models like Midjourney, Stable Diffusion, and DALL-E that generate images from text prompts
- Image-to-text: sometimes referred to as ‘image captioning’. Google Brain runs a long-standing competition for models that can caption a predefined dataset of images.
- Multimodal: models that receive or generate several different types of data as input. For example, SDFusion will take a 3D model and a description to generate an image that combines the two (see below). A future version of GPT-4 will allegedly operate with image and text inputs alike.
AI for construction and engineering
It’s clear that AI will impact all aspects of daily life.
Looking at the long history of the ICE, part of its role has been to lead the charge towards the future.
We think it is plausible that a “generic construction AI” will emerge that is more specific and has dedicated plugins and we could ask it all sorts of questions after uploading our current project data or documentation.
Additionally, here are a few examples of how AI can change construction and engineering over the next two to five years.
AI will allow designers and architects to create ambitious visions much faster, by reducing the time needed to produce alternative designs or alter an existing one.
Commands such as ‘make the windows larger’, ‘move the cafeteria’ or ‘change the bridge to a suspended cable design’ could be given to AI that will quickly outline these new options.
Designers that incorporate this technology will be more productive and creative as a result.
During construction, it’ll be possible to generate solutions on the fly when there’s a problem.
For example, if a construction project runs into material supply issues, it will be possible to ask AI to generate several new options to no longer use a certain material.
The AI could generate alternative drafts of schematics, schedules, bills of materials, cost estimates, and scope documents.
It’ll still be necessary to have certified professionals review the output and add the level of detail required to make the options realistic, but having several alternatives readily available for review will help the decision-making process.
Similarly, during the engineering phase of a project, several ways of constructing something can be quickly drafted.
An AI could help generate many options for utility and power distribution, structure, or how to combine offsite fabrication with traditional construction.
This could feed into an additional AI that will evaluate rough cost and time estimates of each option.
Project management, planning, risk management
When planning a project, the AI could suggest schedule alternatives to fit a particular set of constraints.
Risk registers could be automatically created from designs and schedules to ensure that all possible discrete risks are being thought of.
When a project is delayed or runs into financial trouble, alternatives (schedule, construction methodology, or design of certain components) that improve the capital or time efficiency can be generated.
Operating assets and commissioning
The commissioning and operating of an asset can be helped by AI by ‘pre-stressing’.
AI could suggest weaknesses over an asset’s lifecycle before commissioning begins.
Then, it could create several commissioning and operating plan drafts that look out for these potential weaknesses.
Pathway to more efficient, ambitious and creative projects
AI will be exciting to follow over the coming years.
If the recent pace of development is maintained, the ideas above are only a small underestimated set of what will emerge.
Those who embrace these new opportunities will gain a significant advantage that will ultimately allow them to do their jobs more efficiently, and build more ambitious and creative projects.
All the while, it’s important to remember that AI will automate several operational functions, but we don’t consider it likely that AI will replace many categories altogether.
Rather, AI will disrupt how those categories operate, in the same way that other technological developments have enhanced human capability in the past and elevated each individual to be tens or hundreds of times better.