There’s been a flurry of articles voicing fears of AI in creative fields such as art, graphic design and, to a lesser extent, poetry and literature and, to a lesser extent still, architecture.
As soon as computer processing speed and power increased, it was inevitable that this would be showcased by computers challenging humans in “computational” games such as chess or the Japanese shogi. It became progressively easier for computers to have, use and scan a database of all consequences of all known moves a player could make. Nobody bothers playing computers anymore but people still play chess for all sorts of reasons and I venture their main pleasure comes from the interaction of minds and how it includes elements of surprise, elegance, daring, fun and, occasionally, stupidity.
My first encounter with language translation algorithms was in the mid-1980s in Tokyo when I was doing work for a translation agency that claimed to have the world’s first automatic translation software. Text in the source language had to be rewritten into grammar the algorithm could understand, and the output text had to be rewritten into more natural language but it was the beginning. Importantly, rewriting the input and the output could be performed by persons not paid by the word as human translators were at the time. Much of the work a translation agency used to do is now performed by increasingly sophisticated programs that produce increasingly convincing language. We routinely use translators on our mobile phones but translations of literature are another matter.
Approximations of the human voice and speech patterns came next. Who’d ever think we’d look back with fondness to the days of call centers manned by humans? Nowadays, if you bother requesting help at all, you have to flummox a chatbot to force your enquiry to be diverted to a human who even then might answer from a list of copy and paste responses. I’ve had both good and bad experiences. Humans have to be positively incentivized by things such as pay and working conditions but there are also negative incentives such as performance indicators or the threat of being replaced by an algorithm.
It turns out that most of what we think of as natural sounding AI speech was really just the arrangement of words in expected patterns. And that anything that fits our expectations of an answer will be mistaken to be the product of human thought processes when it’s really just scanning a huge database for the statistically most likely arrangement of words in response. I’m not forgetting that communication between humans can involve the repeating of certain words and phrases in certain situations or that, if we overdo this or use them when something more is expected, we’ll get a reputation for being either boring or insensitive. It’s probably possible to get through a day without a single instance of creative use of language but this isn’t something we should aspire to. It’s heartening to know that certain skilled interviewers can ask speech algorithms a series of leading questions to force them to say stupid, random, or offensive things. To use a human analogy, these interviewers force the algorithm out of its comfort zone of databases and statistics.
It takes a while for humans to learn to speak and, by the time they are adults, they’ve built up their own databases of knowledge and experience to draw upon to create new communications suited to the situation. Humans don’t communicate by first sucking up all the knowledge and data in the known world. The data scraping model for AI might produce approximate results but this is how computers work. It’s not how humans work.
Wordwatch 1: The word “artificial” can mean something that’s a good substitute for something real, as with “artificial heart” or “artificial limb”, etc. It can also mean something that’s a poor substitute for something real, as in “artificial flowers” or “artificial land”.
There are those of us who see Artificial Intelligence as a poor substitute and there are those of us who see it as good enough for certain purposes. The fields of translation and customer service fall into the poor substitute that is “good enough” for some purposes or, probably more accurately, for the purposes of some. The field of medical diagnosis could be an example of a good substitute but I’d still prefer my symptoms be input by a competent doctor. I’m not yet convinced by self-driving cars and I’m definitely not ready for airplanes without pilots up the front.
Wordwatch 2: The meaning of the word “Intelligence” is also fuzzy. I have an “intelligent refrigerator” for example, but even if the refrigerator of the near future sensed I was running out of milk or soda water and order some in, I’d still know it was just some sensors doing their job. We say “Oh what an intelligent dog!” when a dog has just done something we want it to do. An “intelligent student” could be one who analyzes and interprets things they learned and arrives at their own conclusions, or it could be one who merely knows what is expected and how the system works. There are different types of intelligence. “Emotional intelligence” is an instinct for saying or doing the appropriate thing and is not about databases and statistics.
Intelligence and creativity are both applied to problems for which the output is determined only loosely. The terms and language used in technical and legal text have fixed meanings and can probably now be automatically translated almost perfectly, classical Chinese poetry less so. It looks like graphic designers and illustrators are next in line as much of their work involves the creative assemblage of known imagery to form new illustrations or graphics. In the fields of publishing and especially digital publishing, it also needs to be done to tight deadlines and this only lowers the bar for “good enough” and makes AI look more attractive. For graphic designers, the creativity exists in 1) “knowing” what source images or graphics to pick in terms of their graphic potential and the associations they might evoke), 2) “deciding” what a desirable outcome would be, and 3) “combining” the source imagery in a way that produces the desired outcome. However, “Knowing” could be intuitive, learned, accidental, or the contents of a database. “Deciding” could be dogmatic, inspirational, guesswork, or derived from requests such as “A sea otter in the style of Vermeer”. “Combining” could be collage, mashup, pick-and-mix, or the most statistically likely. Money and time set the dividing line between “creativity” and a “good enough” approximation of it.
And so to architecture. It’s been a couple of years since I last saw some supposedly state-of-the-art application of an algorithm to architectural design using a data set and parameters to arrive at some desirable and to some extent predetermined outcome. There was once heated debate about the differences between parametric design and algorithmic design. There are those who claim one or the other is the shape of the future but architects have a habit of bandwagon jumping for any new technology that looks as if it will 1) save time and money, 2) increase profits and 3) make them look cutting-edge if not avant-garde. Remember how Gropius threw craftspersons under the bus in 1923 when he realized that design for manufacture by machine was the shape of the future? Or how post WWII architects rushed to design prefabricated metal houses made by aircraft manufacturing facilities no longer operating at full capacity?
The field of architecture has some easily automated tasks such as the laying out of a housing subdivision, or the arrangement of medical equipment inside a hospital room. Both cases have sufficient specifications to draw upon. It’s possible to automate the design of apartment buildings and to design the apartment building designing software to incorporate parameters for the curvature or overhang of walls (if that particular subset of possible outcomes is what you want), and then for the optimization of apartment layouts in the spaces created. Once again we need to be careful with words. Cost is the most important parameter affecting the curvature or overhang of walls and it would be a service to the world if architects could have immediate feedback of the total cost consequences of any design decision. It’s just number crunching pure and simple, and is what computing power does best. All we need is somebody to create and continually update a global database of materials, quantities and the cost of using them for a particular place and time. This would be more infinitely more useful than a picture of a sea otter in the style of Vermeer and might even work to discourage certain design decisions being made in the first place.
Architectural intelligence that goes by the name of creativity is more problematic. Certain problems can be framed in such a way that solutions can be generated by an algorithm. Or, to be more precise, if we reframe the problem as how to frame the architectural problem so that an algorithm can generate a solution, then offices around the world won’t have to pay such huge sums to their employees anymore. ZHA is on the case.
This “Data-driven, algorithmic housing” project continues to amaze me but mainly due to the fact somebody thought it was exhibition worthy and somehow represent the advancement of architecture. Four elevators for, on average, five floors of, on average, 50-60 studio apartments per floor eh? I’m not going to say anything against the use of light-wells as I’ve been exploring this myself but 5m x 5m would probably be okay if there weren’t beds right up against those windows and as many as eight windows around a 5m x 5m light-well. [Column 4 from the left, Row 4 from the bottom.] The plan and model below show that situation for what is probably only levels two and three so it’s not as horrible as if the light-well had been seven floors deep. BUT. It’s pointless having light-wells and windows if there’s not going to be any audio or visual privacy. Three windows are adjacent to corridors. [e.g. Column 2 from the left, Row 5 from the bottom.] If an algorithm makes a misjudgment of appropriateness then it’s the fault of the embedded values of its creator. But if it makes a more fundamental error then it’s a case of some obvious facts about people and buildings not being given to the algorithm. TWO UNITS ARE ACCESSED VIA VOIDS FFS! [Column 4 from the left, Row 3 up from the bottom.]
In fifty years we haven’t progressed that far from shuffly windows. We’re still locked in a phase where self-similarity and variation are understood as representing creativity. Randomization (within parameters) is something that can be easily computerized and produced with minimal time and labour. We need to decide if this is artificial creativity or whether we are redefining or being asked to redefine creativity as what the available technologies of the time can accomplish. The above example of data-driven design has a an apparently random sprinkling of balconies, but only in places where the obstruction of light matters less. The building is higher towards the north but contrivedly and irregularly so. The floors are non-identical to allow more light but again, contrivedly irregularly so. Despite these nods to live-ability and current notions of what’s currently pleasing to the eye and mind, it’s fairly easy to see what were chosen as parameters for optimization and what wasn’t. We can’t blame AI for those decisions.
The only question is what we want from architectural creativity. Is it still all about shape making or, as some say “form giving”? Or is it still all about success in branding oneself while one brands one’s clients? It’s always been about problem-solving but the problem is that the nature of the problem is a moveable feast. If we’re not clear about what architectural creativity is then we’re in danger of falling for representations of it or, worse, for crude approximations of what we think it is. We’re prone to do this anyway, with or without AI. AI could just be another instance of architecture aligning itself to technologies that look like ther future. Remember, we’re still waiting for factory-produced prefabricated houses to revolutionize housing supply. 3-D printing may have revolutionized the field of medical prosthetics but it’s had zero impact upon the construction of buildings and how we live.
Sooner or later it will be possible for an architect-person to say DESIGN ME A 100M2 FISHBURGER RESTAURANT IN THE STYLE OF FRANK GEHRY CIRCA 1985. Too easy? OK then. DESIGN ME A 10,000M2 MAXIMUM SECURITY DETENTION CENTRE IN THE STYLE OF ZAHA HADID CIRCA 2000. Both are database-driven framings producing deepfake solutions. Would they be violations of intellectual property? Probably, as many contemporary artists are discovering. I’m not going to end by claiming the human brain is superior and that creativity is some mystical thing impossible to comprehend. It’s just that the human brain is a constantly reorganizing database of learning, memories and experiences and we have to use placeholder words like inspiration and creativity to explain how that data is selected and combined to create a desired output. This all happens in our brains that are STILL A BLACK BOX and, as long as they are, we cannot expect anything but crude approximations from AI, even if they prove good enough for the task they’re given.
• • •