Extra Precision Considered Harmful
Architects do not make buildings; they make drawings of buildings
— Robin Evans
The design software company Autodesk released the first version of AutoCad in 1982, almost 40 years ago. AutoCad took off, and in 1985 Autodesk went public. It is now the largest company in the world that makes design software for architects.
In 2002 Autodesk acquired a company that offered a new kind of architectural design software, called Revit. Revit seemed to promise a new way of designing buildings. A term was coined to describe this new way of designing: Building Information Modeling, or BIM.
BIM has been part of the mainstream — part of the product stable of the biggest architecture software company in the world — for almost half the time we’ve had any practical architecture software at all. Twenty years from Autocad 1.0 to Autodesk Revit, eighteen years from Autodesk Revit to the present.
So, why does it still feel like we’re waiting for BIM to take off? Revit and similar tools are widely used, of course, but they’re used mostly to produce drawings, which is exactly what AutoCad 1.0 was used for in 1982.
Revit and other BIM tools offer plenty of advantages over first-generation CAD tools, and architecture firms willingly pay for them, but the central idea of BIM, that 3d software models of buildings would replace drawing sets as the medium of communication in the AEC world, just hasn’t panned out.
Why is this?
There’s a graphic that concisely expresses the promise of BIM, sometimes called “the sawtooth chart”:
The idea is that every information handoff in the traditional design process results in a loss of information.
Implicit in the chart is the idea that more information = good. This makes sense, intuitively, but the past twenty years of BIM have demonstrated that the reality isn’t quite that simple. Sometimes extra information is harmful, if it’s the wrong kind of information.
To think about this correctly, consider an analogy to the concept of significant figures. Significant figures are a digital expression of the precision of a number. 1.0000 is a different number than 1.0, because, while they have the same value, the former expresses a much higher level of confidence in that value. 1.0 might, for all we know, actually be 1.0003.
Scientists take great pains to use the correct number of significant digits, as both understating and overstating the precision of a measure can be deeply misleading.
The level of detail in an architectural representation can be understood as expressing a level of confidence in the precision of that representation. If an architect adds a level of detail that expresses an excessive level of confidence, this is bad for the contractor.
Suppose, for instance, that the architect specifies the location of every drywall screw on a wall. What should the contractor do with this information? If the contractor takes the time to look up the location of each screw before drilling it in, the drywall installation process will take much longer. If the contractor simply ignores this information (almost certainly what would happen), then the architect’s time in specifying those locations has been wasted, and more ominously, the possibility of divergence between design intent and design execution has been opened up. The extra precision was, in this case, purely harmful.
This way of thinking about design representations and precision suggests a principle:
Increased design precision is harmful without a way of effecting that precision on the work site.
and a corollary:
The precision of a design representation must be matched to the precision of the construction methods used to execute that representation.
It turns out that traditional construction drawings are already very well matched to the precision of existing construction methods. Architects can, of course, express arbitrary levels of precision with drawings if desired, but expressing unusually high levels of precision in a drawing set requires making extra drawings, and architects only do this when the extra precision is important (these extra drawings might be details drawn by the architect or shop drawings produced by a fabricator in cooperation with the architect). Contractors, in turn, know to look for these extra drawings and understand that when they appear, and only when they appear, are extra precision and the concomitant attention required.
In other words, Revit and Archicad and related software tools did as well as they possibly could have. The only way they could have done better is by altering actual construction methods, not just the way designs are represented — a tall order for any mere software. These software have made it possible for architects to produce what they’ve always produced — construction drawings — more efficiently and effectively, and that is no mean feat.
But it wasn’t a feat that transformed the building industry. In order to accomplish that, a solution must be developed that doesn’t merely propose to improve design representation, but the very methods by which buildings are constructed.
A postscript about the sawtooth chart: is the chart just wrong?
No. I think the diagnosis offered by the sawtooth chart remains fundamentally correct, but only if the Y axis is understood to represent a particular kind of design information. Some kinds of additional information, as we saw above, just bog down the pipeline and lead to inefficiencies across the board. But if the information losses between parties are understood as being losses in comprehensive design reasoning, then I think the point of the chart stands.
The problem is that existing BIM tools are not really designed to to offer additional information about the reasoning of the various parties in the design and construction sequence. My impression is that reasoning is still usually communicated in ad hoc, verbal ways, in meetings and through emails.
A software tool designed to pass reasons, not just decisions, through the sequence, would look very different from anything that currently exists. And whatever such a tool might be, it doesn’t yet exist.