Construction complexity, causal opacity, and why it's more important to 'make sense of' than to control information.
Plus: Ingesting and querying unstructured data, robots, and the UK's plans for AI datacentre buildout.
What I’m thinking
Causal Opacity
Construction projects are complex systems-of-systems — not just because of the number of moving parts, but because of how those parts interact. Trades, disciplines, companies, contracts, and technologies collide and cooperate across time. And it’s the interactions — not just the components — that determine outcomes.
But within such a system, cause and effect are not easily paired. A defect may appear months after a decision was made, triggered by a combination of overlooked details, misaligned assumptions, or unspoken expectations. A delay might stem from paperwork as much as procurement, and a safety incident may be traced not to one act, but to a cascade of quiet signals that went unread. In this context, we aren’t dealing with simple chains of cause-and-effect — we deal with webs of influence and feedback loops.
Information is what binds this web together. It is the medium through which action happens — decisions are made, instructions are carried out, warnings are raised, trust is built or eroded. It’s also the substance through which we perceive what’s happening. In short: if construction is a living system, then information is its nervous system.
Here lies the trap that you can fall into. If information is what drives the system, then surely by controlling information you can gain control over the system itself. This is the illusion. Because in reality, information is not linear, nor fully knowable. Some of it is formal — schedules, contracts, RFI responses, models, documents. But so much more is informal and ephemeral — a conversation on-site, an email not cc’d, an assumption not challenged. The visible tip of the information iceberg is managed. The vast bulk — undocumented interactions, tacit knowledge, cognitive bias — remains submerged.
Trying to control information, then, is like trying to consciously manage every function of your body — your heartbeat, your digestion, your micro-movements — you can’t. You can only nudge, interpret signals, and respond. The same is true in construction: success lies not in tight control, but in sense-making. In tuning in to the weak signals, aligning the rhythms of different actors, and creating systems that are sensitive and responsive, not brittle and rigid.
“For complex systems are, well, all about information. And there are many more conveyors of information around us than meet the eye. This is what we will call causal opacity: it is hard to see the arrow from cause to consequence, making much of conventional methods of analysis, in addition to standard logic, inapplicable.” Nassim Taleb, Antifragile
In complex systems, the arrow from action to outcome is foggy at best. This is Taleb’s idea of causal opacity. Standard logic falls apart because reality isn’t a neat series of dominoes — it’s a network of interdependencies where small nudges echo unpredictably.
So what do we do? We accept the opacity. We design for it. We move from command-and-control to sense-and-respond. We build ways of seeing and listening — dashboards, data stories, knowledge systems, team rituals — not to control information, but to make sense of it. And in doing so, we increase the system’s ability to adapt, not just comply.
The Case for an Intelligence Layer in Construction
In complex systems like construction, success depends less on knowing everything and more on sensing the right things early. But our systems aren’t designed for that. They are built to record, track, and report — not to understand.
I envisage an “intelligence layer” which sits on top of project and enterprise platforms — Common Data Environments, planning tools, finance systems, health & safety platforms — and doesn’t try to replace them. It listens. It senses. It looks for weak signals across the system:
Friction points in workflows
Repetitions and rework
Gaps in data handover
Emerging risks
Patterns of silence as much as patterns of activity
Just as our nervous system detects signals our conscious brain might miss, this layer gives construction leaders a kind of augmented awareness — a way of seeing not just what has happened, but what wants to happen if the current trend continues.
This isn’t about dashboards and metrics alone. It’s about building a layer of sense-making — using knowledge graphs, LLMs, and contextual reasoning to trace how signals relate to one another, across phases and disciplines. It’s not BI. It’s not just RAG reporting. It’s an evolving map of how information flows through the living system of a project.
The intelligence layer doesn’t impose control. It offers insight. It embraces causal opacity, not by trying to eliminate it, but by creating conditions where subtle causes can still become visible — and acted upon in time.
What I’m consuming
I’m constantly scanning the news, social media, and industry for case studies, innovations, and generally interesting construction related content. In these very brief articles my goal is to share these with you.
GroundX: Data Extraction from Complex Documents
Capturing data from unstructured documents has been a challenge. Often knowledge is locked inside pdf and word files. Historically, Natural Language Processing (NLP) has been valuable in extracting this information, such as sentiment analysis, in order to unlock the data at scale. But with the advent of LLMs more value is on the table, and better tools are available.
This week I’ve been playing around with EYELEVEL AI’s GroundX which is a specialist tool to extract, store, and exploit documented and unstructured information.
I built two quick tools using python, OpenAI APIs, and GroundX APIs to show the value of being able to capture and interact with information derived from complex documents with a mixture of text, tables, and figures/imaged. For the documents I used the standard details drawings from West Berkshire Council.
To ensure fairness and consistency I used the following system prompts and messages for both tools:
Prompt: “You are reviewing a civil engineering standards document containing detail drawings and technical notes. Respond based only on the information in this document. Please answer clearly and concisely for a technical audience.”
System message: “You are a civil engineering assistant with expertise in interpreting standard detail drawings and technical notes. You help users understand the purpose, materials, specifications, and usage of construction elements like foundations, kerbs, drainage, and concrete. Base your answers strictly on the document content provided."
I also standardised the user prompts to test the tools with two questions:
"Summarise the purpose and scope of the civil engineering standard detail drawings and notes in this document. What types of projects are these details intended for, and what kind of work do they cover (e.g., drainage, footways, roadworks)?"
"According to this document, what concrete grades, mixes, or specifications are required? For which construction elements or standard details (e.g., foundations, kerbs, channels, footpaths) should each type be used?"
The table below compares approaches and results:
Okibo Robot Painter
<Robotics & Automation News: Okibo>
Okibo, a construction robotics developer, has launched its US headquarters in New Jersey and introduced its EG7 robot to the American market. The EG7 is an autonomous, AI-guided robot designed for painting and drywall finishing, capable of covering 1,000 square feet (185m^2) per hour. Operating independently without cords, pumps, or Wi-Fi, it uses AI for navigation. This compact and lightweight robot aims to improve safety and efficiency on construction sites, addressing the construction industry's labour challenges.
“Unlike systems that depend on external markers or BIM tools, the EG7 uses a patented AI-driven 3D scanning and real-time modelling algorithm for navigation and execution.”
Robotics are getting better and better. There is a bit of a competition between robots which are specialised such as this painting and drywalling machine, and humanoid general robots which can do multiple tasks. In an industry with a skills challenge which is only getting bigger and a drive to for safety and wellbeing there is more than enough room for further mechanisation. The tasks for the innovators and manufactures is to make them cheap enough, with proven performance, to encourage contractors to invest.
Building the UK’s AI Powerhouse
<Data Centre Knowledge: UK 500 MW AI Cluster>
“The UK government wants to build strategic partnerships with AI developers to work on shared AI and AI-enabled priorities. The government wants to work with data center developers and energy solutions firms to establish an AI infrastructure cluster of at least 500 MW by 2030. Officials are particularly interested in strategic proposals beyond data center developments to contribute to the UK’s broader AI ecosystem, including research, innovation, skills development, and energy solutions.”
The question is whether the power grid infrastructure can adequately supply the significant energy demands of such a large-scale data centre development. Issues such as high energy costs, planning restrictions, and slow grid connections are potential obstacles to meeting the AI infrastructure goal and cementing the UK's position as a leader in the tech industry.