If AI is going to help construction learn from its own knowledge, the industry should take an active role in shaping how that knowledge is governed and used, argues Zacharias Fotos MCIOB.
A typical construction project produces several thousand drawings, specifications, schedules, reports, RFIs, meeting minutes, calculations, technical submissions and certificates. Collectively, these documents contain far more than project information. They contain professional knowledge, judgement, liability, experience, regulatory interpretation, coordination decisions and lessons learned over many years.
For a long time, much of that knowledge has remained fragmented. Some of it sits in common data environments, health and safety files and O&M manuals. Some of it sits in consultants’ archives. A great deal of it sits in the heads of experienced professionals who remember why decisions were made, what alternatives were rejected and which risks were accepted.
AI could be one of the most useful tools construction has ever had for unlocking that knowledge.
The industry is already using AI tools to search, summarise, generate and interrogate project information. These tools can compare drawings, read specifications, identify gaps, flag inconsistencies and support compliance reviews. Used well, this could reduce manual checking, help teams find information faster and support better decisions earlier.
“Construction should not be afraid of AI. If anything, we should be far more ambitious about it. But ambition should come with clearer governance.”
That is why construction should not be afraid of AI. If anything, we should collectively be far more ambitious about it. But ambition should come with clearer governance, because there is a question beneath the productivity story: who owns construction knowledge in the age of AI?
Not just a copyright issue
This is not simply a copyright issue, and I do not think it should be reduced to a purely legal debate. Construction information is different from most digital content because it is produced in an environment of accountability. A drawing is not a generic image. A fire strategy is not generic text and diagrams. A programme is not just a sequence of dates.
These documents are created by trained professionals, under appointments, contracts, regulatory expectations and real-world consequences. In some cases, people have been asked to defend their decisions in court.
A construction drawing is therefore never just a file. It is part-instruction, part-record, part-contractual artefact and part-professional commitment. Behind it sit assumptions, coordination decisions and professional judgements formed over months, sometimes years.
That matters when AI enters the picture.
There is a clear difference between using AI to review information for a live project and using project information to improve or train a commercial AI system. The first may be a natural extension of existing project workflows. The second is more complicated.
Lessons from the Napster era, not a repeat of it
To be clear, this is not a like-for-like comparison. Drawings are not songs, and construction projects are not music libraries. But it is a useful reminder that technology can move faster than professional norms, commercial models and regulation. Once information becomes easy to copy, search and reuse, the question of who benefits from that reuse becomes much harder to ignore.
This is not only about where information goes. It is also about where value goes. If proprietary project information helps improve a commercial AI system, then the organisations and professionals who created that knowledge may be contributing value without visibility, attribution or benefit.
“The industry has spent decades trying to capture lessons learned, reduce rework and improve coordination. AI may finally give us the tools to access and reuse that knowledge at scale.”
A simple opt-in to a ‘data partnership’ may therefore be a much bigger decision than it appears: a project team could be allowing the reuse of information it may not fully own, to improve systems that may later reshape how that expertise is delivered.
That does not mean construction should resist AI. It should use AI to improve quality, reduce waste, strengthen compliance and make better use of the information it already produces. But adoption should not happen without greater transparency about how these tools are developed and how project information is treated.
Interrogate vendors
Clients, contractors and consultants should start asking more direct questions of software vendors. Is project information used only for that project? Is it retained after upload? And is it used to improve or train the model? How is confidential or commercially sensitive information handled? What happens to the information once the project ends? Some vendors already offer clear commitments on these points; others do not, and the difference matters.
Questions like these should sit alongside discussions on information management, cybersecurity, quality assurance and professional responsibility. They should form part of organisational AI policies, not be left to project teams to work out under pressure.
The industry has spent decades trying to capture lessons learned, reduce rework and improve coordination. AI may finally give us the tools to access and reuse that knowledge at scale. That is exciting. But if that knowledge is valuable enough to power future systems, it is valuable enough to govern properly.
We do not need to answer every legal question immediately. But we do need to start asking better questions now. If AI is going to help construction learn from its own knowledge, the industry should take an active role in shaping how that knowledge is governed and used.
