Full House

It’s only recent history but, before the use of image datasets became commonplace, architects used language prompts to tell an AI system what they wanted. The problem was that computers and people don’t speak the same language. If you say to an AI system “Design me a beautiful building!” it won’t know what you mean – because we don’t know what we mean. If your AI system was fed internet images of buildings tagged “beautiful” you might receive a mashup of Fallingwater, the Taj Mahal, Rouen Cathedral and a selection of houses from Home Beautiful. This post describes the same aesthetic framework I’ve posted about before. It uses six attribute names and sixteen different aesthetic effects resulting from combinations of three types of architectural ideas. Together, they describe and classify 16⁶ = 16,777,217 aesthetic signatures with alphanumeric identifiers to facilitate search and retrieve. This is a model for how human intelligence indexes and retrieves architectural aesthetic information. It just happens to be AI-compatible.

The worth of any model is in the amount of known phenomena it can describe and the amount of unknown phenomena it can predict. This latest attempt of mine to describe it begins with the writings of Charles Jencks, not because I believe everything he wrote but because I don’t, and so I’ve looked at his writings more critically, particularly those relating to what he called the iconic building. He wrote that there were four conditions a building needs to satisfy if it is to be iconic.

The first was that a building look different in its surroundings. This was the only tangible condition. Iconic buildings stand out, he tells us. The second was that a building be known to be new, or original, or novel, etc. This is no surprise. Much architectural endeavor is for the purpose of making something seen as original or at least novel. Form chasing has long been seen as the goal of architecture. The third was that the building have some conceptual association with its surroundings. This gives us bird-shaped airport terminals and shell- and sail-shaped buildings near bodies of water. We’re used to this. The fourth was that the building appear to be something not a building – or at least not the building it is. This last types negates the identity of a building as a building. It’s not just about making a building look like a duck or a flower. Ideas of Negate have the power to make us question whether or not something is a building and hence force us to reconsider what it is a building can be. They can be powerful, especially in combination with the other two types of idea. I’ve given these last three conditions the names of Ideas of Separate, Ideas of Unite, and Ideas of Negate.

Buildings that Jencks identified as iconic often inspired sketches showing us how they evoked associations of things that weren’t buildings, and that often had some sort of association with the location. Bilbao Guggenheim has many watery and fluid associations, St. Mary Axe has associations with dynamism and energy, and Beijing’s CCTV has associations with Chinese gates and joinery. Jencks even once mentioned the term “moon gate”. However, a building may evoke associations in the mind of a viewer but architects have limited control over what, if any, associations are evoked. To paraphrase philosopher Douglas Hume, these associations do not exist in the buildings themselves, but in the mind of the person that contemplates a building”. This makes sense. Nobody believes in absolute beauty anymore.

Later, Jencks was to write that buildings that produce these free associations had what he called the enigmatic signifier. In essence, these were just a combination of an Idea of Unite that made a building seem conceptually fitted with its context, and an Idea of Unite that made a building seem like something not a building. Therefore, all a building needed to do to be iconic was to have an enigmatic signifier and be new at the same time.

On the left, below, is a model of those four conditions as Jencks stated them, with SEPARATE being the universe of buildings that look different from their surroundings. On the right is a model of Jencks’ actual area of concern.

Only three modifications are necessary to expand Jencks’ model into one that describes and classifies the aesthetic effect of ALL buildings. Firstly, a building attribute such as the shape of a building doesn’t always have to look different – it may have a tangible similarity with its context. This is another completely separate universe of possibilities. The model can now describe twice as many buildings as it did previously. 

Secondly, a building attribute night not evoke all three types of architectural idea. It could evoke, two, one or even none. There are now sixteen possible conditions for the Shape attribute of a building. The model can now describe sixteen times as many situations as it did before. The names of the effects describe the components of the effect and are for the benefit of the user. The respective numerical identifiers have no mathematical meaning but may assist computation.

Thirdly, buildings have attributes other than Shape and that also carry meaning. There area also visible attributes of Colour, Pattern, Position, Alignment and Size. All buildings have these six mutually independent attributes. The model can now describe 16⁶ times as many situations. 

You can think of this as a slot machine with six wheels, each of which can have one of sixteen positions. It will have 16⁶ = 16,777,216 unique combinations of attributes and effects. This is the mechanism that translates aesthetic information into computational information that can be indexed, searched and retrieved.

Each attribute can have one of the sixteen different effects.

The first two are 0 SEPARATE and 1 UNITE. These have no architectural ideas and we use words such as “charming”, “naïve”, “vernacular”, and “primitive” to describe them. Above left is an example of Colour to 0 SEPARATE and below it an example of Colour to 1 UNITE. 

Six more effects have one type of architectural idea.

2 DETACH and 3 ATTACH both have an Idea of Separate. In the top image above, the Pattern of the building is different from is surroundings, while the building below it has intentionally been made to look similar. In both cases, the Idea of Separate is “artificial” or “man-made”. Detach is a strong effect because the visual difference is reinforced by an idea of difference.

The next two are 4 Extract and 5 Combine. The oil rig in the upper left doesn’t look positioned with respect to anything but we know it is probably where it is because of the possibility of oil or gas below. The position of the lighthouse has a one-to-one physical unity with the island, but we know it is there to warn of hidden dangers nearby. Combine is another strong effect because the visual similarity is reinforced by an idea of similarity.

The third set of effects with one type of idea are 6 Disguise and 7 Merge. The upper building has the shape of a duck when we can’t see any ducks, while the lower building has the shape of a wave when we can see other waves. Both buildings have shapes of things that aren’t buildings. You can think of Disguise as, say, a building being shaped like a mountain when there are no mountains around, and Merge as being shaped like a mountain when there are.

Six more effects have two types of intangible association. 

These first two have no Idea of Unite and thus it is impossible for these buildings to be considered a part of their surroundings. The Size of the building in the image on the left has no way of fitting in but, in the image on the right, it appears to have found a friend.

This second pair have no Idea of Separate and this means building attributes exhibiting this effect can never be truly novel. The upper building has a visually different shape, and the Idea of Separate being of something fluid and the Idea of Negate being of something fish-like. The lower building is similar in shape to the sail of the boat characteristic of the region, and is also something that is not a building. 

The third pair of effects have no Idea of Negate. Building attributes exhibiting this effect can never really challenge our ideas of what a building can be. The alignment of the building on the left varies but we know it always faces the direction of the wind. The alignment of the building on the right is fixed but towards the prevailing wind. The novel idea of a building funneling wind to turbines is the Idea of Separate.

Only two effects have attributes evoking three types of architectural idea. The Sydney Opera House looks different, was known to be different at the time, and is evocative of waterside things such as sails and shells that are not buildings. The Museum of Cave Paintings at Niaux has a shape that “fits” the shape of the mouth of the cave and seems like an animal protecting the mouth of the cave in which it lives.

As the number of ideas evoked increases, the effects become increasingly unstable as it becomes difficult to sustain different ideas for any length of time even when only one person is involved. Moreover, if the notion of novelty is the Idea of Separate, then this notion will fade over time unless something is so novel that it will always be remembered as having been novel once. For example, The Empire State Building is no longer the tallest building in the world. This is an idea of Separate. However, it is still remembered as having been the tallest building in the world and this is still an Idea of Separate that applies to only a very small set of buildings. Without novelty as an Idea of Separate, buildings that aimed at being iconic become lame signifiers instead.

• • • 

We would not expect a system that organizes subjective architectural responses to be simple but this model is not particularly complex either. Its value lies in the amount of information it organizes. First of all, we need to be certain of what we’re actually seeing before we can have an opinion about it. Uncertainty, Tension or Discomfort occurs when an attribute of a building is not clearly seen as SEPARATE or UNITE. Mirror glass often does this for Colour, making it difficult to tell if the effect is Colour to Disguise or Colour to Merge. 

All buildings have six attributes, each of which will exhibit one of sixteen effects. Most of these signatures will not follow any particular pattern and effects that Separate will conflict with or cancel or confuse effects that Unite, However, distinctive signatures occur when there is synergy between attribute groups. The following are some I’ve identified. Consistency is when the three Surface Attributes of Color, Pattern and Shape exhibit the same effect. There are 16⁴ = 65,536 aesthetic signatures like this. The signature component 222___ usually denotes a Modernist building as the three Surface attributes each evoke the Separating idea of ‘artificial’. For the lighthouse, the surface attributes are all visually different – you do not want your lighthouse to blend in with its surroundings. With this lighthouse, red and blue are the opposites of green and yellow but together, the four colors form a single palette of primary colours.

Importance is when the two Placement Attributes have a visual unity with surrounding landscape features. It typically represents power and authority and is manifested by symmetry and axes. Importance is exhibited by 16⁵ ÷ 2 = 524,288 aesthetic signatures.

Strength is when each of the three groups of Surface, Placement and Size is internally consistent. There are 16³ = 4,096 signatures featuring this.

Emphasis is when only one characteristic out of the six displays a different effect. There are 16² = 256 aesthetic signatures like this. In both of these examples, the emphasized effect is color.

Finally, Beauty, for want of a better name, is when all six attributes exhibit the same effect. 16¹ = 16 aesthetic signatures have this. 

And here they are. Having sixteen different types of beauty explains why very different buildings can be seen as beautiful – in their own way. We can recognize the beauty of technology, of the vernacular, of the exquisitely contrived, of the weird and wonderful and also of the contextual and the poetic.

• • •

This model organizes subjectivities but offers no guarantee that all people will even see the same thing, let alone have similar ideas evoked by it. Different people can see the same building, or the same building at a different time, and feel differently about it. It’s what happens. The two buildings below will not be universally seen as examples of aesthetic signature CDDDDD below show how buildings with identical signatures are doing the same thing aesthetically even if manifestations of that signature differ. This is what allows this model to be used to index and search an image database to retrieve a more coherent set of images for whatever AI process comes next. As long as you specify what type of beauty you want, it is now possible to ask your AI system to “Design me a beautiful building”. It might never know what you mean by that, but it will at least be able to retrieve and process examples in line with what you wanted the attributes of your building to do aesthetically.

• • • 

• There still remains the problem of indexing all these images. AI-assisted figure-ground pattern recognition can deal with the tangible judgments of Same or Different but although this model can organize subjectivities we still need to know what those subjectivities are. In theory, statistical data could be gathered for a large number of people and a large number of images but I don’t hold much hope for a Wikipedia-like universal databank organized according to this model and the statistical prevalence of Ideas of Separate, Unite and Negate. Instead, we seem to be moving in the opposite direction with isolated AI systems trained on image banks pre-loaded with subjectivities in the form of selection bias. If AI training data is not systematically organized, then architectural AI will only get better at producing the increasingly tropey Zaha Hadid buildings it is now.