K
Klaus Philipsen
Guest
- Thread Author
- #1
The death of the architect has been predicted before, killed by engineers, by CAD, by parametric design, and now by AI. At the same time we may achieve immortality as our virtual twin. At architecture conferences sessions addressing AI were packed by people seemingly motivated by equal parts of fear and hope. The phrase "If You Are Not at the Table Then You’re Probably on the Menu" is frequently used. The somewhat milder version of death by AI states that AI will not replace humans but humans with AI will replace those without it.
[td]
[/td]
[td]
Old school architectural drafting (Life on Sundays)
[/td]AI in the building industry needs some unpacking, and fast, since AI is progressing rapidly into our professions.
More avantgarde firms like Coop-Himmelblau, MVRDV and BIG have been experimenting with AI for a few years by now, and many mainstream firms are following suit. Ask MidJourney to design a Zaha Hadid building and it will come up with a structure that looks like the late architect would have come back from the dead.
By now AI design tools produce much more than just renderings. They are embedded in Google Earth (Delve) and available on many platforms that allow architects to generate site-specific early massing models, floor area calculations, and unit counts much faster than with SketchUp or similar massing software. Later on in the design process AI can check for embodied carbon, compliance with codes and ensure consistent workable details.
Folks intimidated by all the talk of AI and digital twins should take heart. Most of us have long empowered ourselves with these tools without much consideration, making ourselves smarter in the process, smarter than any of our forebears. Anyone with a smartphone has the entire universe of knowledge in their pocket or purse. The device, thus, has already more knowledge than any human in the entire world, before the term "artificial intelligence" is even needed to explain it and before anyone has used ChatGPT or Apple's new Image Playground. Much of this has to do with "digital twins."
The Digital Twin of the World
Increasingly, the physical world is simulated in digital twins. Weather forecasts based on a multitude of data project quite accurately model a meteorological reality for an entire week. In fact, realistic and dynamic digital models of reality are not science fiction but daily reality. As is often the case, the gamers may be the furthest along with digital worlds, for example, created in Unreal.
Architects have long anticipated buildings in analog models and drawings from which builders created the real product. Then, the drawings were created digitally via CAD, and the model potentially built by a 3D printer. In building information modeling (BIM),
the drawings were enhanced with knowledge about materials, costs, energy, and any data one wanted to apply. Parametric architecture connected materials to form (Frank Gehry, Tim Fu). To become a digital twin of the anticipated product the static models of the past become so infused with data that they become dynamic. Future performance can be simulated with great accuracy, not only in terms of cost, people movements, function, or energy usage but also in terms of behavior in adverse conditions such as storms, floods, and fires. Most importantly, the digital twin allows optimization long before an actual building has to go through the trial and error of real-life conditions but it can also be used in parallel to the physical building to anticipate failures, maintenance needs or to improve performance.
The Digital Twin of Us
Digital twins are not limited to matter, of course, the can include nature and also humans. Thus the computer age added the digital twin to Sigmund Freud's Troika of ego, super-ego, and id. Our virtual alter egos are out there as representatives of who we are in real life, either in real time (digital twin) or for simulations (virtual twin). Our virtual alter egos are part of a new exploited labor force that does the work for the new data-driven economy without either we or our virtual twin getting paid. The latter, and that it happens involuntarily, has made some of us uncomfortable, especially after those digital giants who collect the data, such as Twitter, Facebook, Google, or Amazon, have gone rogue.
The one-dimensional and incomplete digital simplifications of our person feed endless data streams. As the shopper about whom the store knows little in general but everything in terms of purchasing habits. As a driver in a vehicle about whom Google or Waze knows little except location, speed, origin, and destination. As a patient whose data populate not only our health portal, but also that of labs, other doctors we visit, and whatever HIPPA allows. Our daily activities unleash an entire army of virtual alter egos that provide work by allowing precise knowledge about the past and good knowledge about a likely future. Our alter ego riding a bus informs several dozen other alter egos waiting for the same bus along its route with real-time information about where it is and, with "it," where the bus is.
These simplistic forms of us are barely more than our footprints in the snow, yet when overlaid, when data points are continually overlaid over some time, they allow a more complete representation of ourselves that increasingly resembles us in a potentially breathtaking totality with a comprehensive view of an individual’s physical, behavioral, and emotional states. It is easy to imagine how our alter ego can gain complexity until it is almost indistinguishable from our real self, except it may lack spontaneity. Possibly, though, our virtual twin knows better what we will do next while we foster an illusion of free will. In what is, so far, is science fiction, those digital twins could eventually act on our behalf, and in this way, they could also live on after our own death, making us immortal.
[td]
[/td]
[td]
AI image generated upon the prompt to show AI
in our daily life (author)
[/td]in our daily life (author)
Clearly, effective dynamic modeling achieves a higher level of power when the the material twins and people twins are merged. Initially, these digital parallel universes were powered by traditional computing power. With vastly increased capacity to run data, a leap into artificial intelligence is possible, at the very least AI can be used for data processing and evaluation of digital twins. Essentially, this adds learning, reasoning, and problem-solving to the regular data processing, with the prospect that generative AI can train the dynamic models to become ever smarter. The border between stored knowledge and AI is fluid. Your smartphone that suggests activities and optimizes trip suggestions based on how you drove previously has learned not only from all other drivers using GPS in general, but also directly from you and produces custom suggestions.
We Already Live in an AI World
Even before combining digital twins of the world and of ourselves, current digital models far exceed the quality of those which rely solely on human intelligence, creativity, and experience. The reality is that we live already with AI on a daily basis. Most of us are well used to the speed and convenience of AI enhanced models, from the weather forecast to ordering an Uber, getting directions from Waze, passing airport controls via face recognition, or resolving an order problem with Amazon. The world is bifurcated between those who have access to those resources and those who don't. Expectations continually rise. Consumers dealing with agencies and institutions that still operate in the old slow analog world become increasingly frustrated and flock from taxis to Uber or from brick and mortar retail to streamlined online selling.
[td]
[/td]
[td]
The Data, Information, Knowledge, Wisdom (DIKW) pyramid
[/td]This frightens many whose livelihood is designing the future, including architects, urban designers, planners, and engineers, whose project horizons are counted in years during which human design skills are critically important. What we often overlook is that the slow speed is not because it takes so long to design. Instead, the time-frame is determined by the non-creative processing of rules, regulations, submission of permits, and demonstrations to clients. If each of the "dumb" or non-creative steps were accelerated by AI, our intelligence, creativity, and experience could engage more efficiently and at a much higher level. Our brains already function on a level that is adjusted to the digital age.
Imagine how hopelessly overpowered a medieval figure would be if immersed in the modern world, in which we can act without much of a problem. This is not just a matter of the technologies that would baffle someone jumping in from the past. Imagine the manuals of this world had been somehow downloaded into this person before her time travel; it wouldn't help. Her brain would not be able to process the speed of the world flying by at 70 mph, would not digest travel across time zones and climates, comprehend our languages, our art, our music, or our ideas of the universe. That our 2-year-olds can swipe our smartphones in a meaningful way is not because they read the manual.
AI- Opportunity or Threat?
Fear is not a guide for managing the future. Since it is absolutely certain that refined AI will have enormous impacts on almost all aspects of life, we better understand, prepare, and control what is going to happen. This means aligning, and training AI with human values, and controlling the companies that develop and deploy it, so that AI won’t be so easily usable for nefarious purposes.
[td]
[/td]
[td]
AI at AIA International conference in Hong Kong (Screenshot)
[/td]
Opinions about AI are divided between those who believe that modeling with AI will make for a better world and allow the human operators to push their designs far beyond the realm of what was possible in the past, and those who believe that AI will be the end of jobs, creativity, and presents an imminent danger to mankind. Surprisingly, some of the developers of AI belong to the second group. The warnings can't easily be dismissed as uninformed.
Much brainpower is spent on the question of whether AI can be smarter than humans, be conscious, and whether such self-learning AI could ultimately no longer be controlled by humans and instead control us. Computing power has already been more powerful than the human brain when it comes to chess and math, with ever more power in sight through quantum computing. Link this power with artificial general intelligence (AGI), i.e., self-learning AI that is closer to human intelligence; it is clear that AI will be smarter than we are. The issue with consciousness is less obvious, in part because we have no consensus on a good definition for what the term exactly means.
What Architects Can Do and AI Can't
Architects like to debate whether AI will ever be able to design for "emotion," i.e., in such a manner that human feelings can be anticipated and stimulated by spaces built into designs the way creative architects do. The discussion about AI's potential of replacing architects depends on one's view of architecture. The architect and theologian Terrence Curry modifies Sullivan's "form follows Function" in his publication Form Follows Feeling:
By contrast the structuralist architect Peter Eisenmann sees architecture as a process of critical engagement with the syntax of systems, language resulting in a form. He, too modifies Sullivan and says:
AI can be used either way. For example, Mid-Journey can be used by an architect who may not be skilled in turning ideas or intuitions into sketches but is good with words. The software can turn such word commands, for example, whether objectives or emotions via text into a video, emotive images or models. Those can be refined with additional commands, or the results can be blended. Eventually, an objective or even a feeling turns into something that resembles a structure. What this type of image generation usually lacks is the analytical thinking that in the traditional process informs initial design concepts. Good architects create value by responding to particular site and program conditions with a unique design that is derived from those. Mid Journey can potentially do this, but only with the appropriate human input, since the software is unable to go out to the site, see the views, the terrain, hear the noise, or determine the styles of surrounding buildings.
[td]
[/td]
[td]
Midjourney image creation (via Philippe Morel)
[/td]Benefit and Risk are Intertwined
The history of technology shows that rising utility and benefit bring with them a rising risk of unintended consequences and potential dangers. Nuclear fission has brought high energy potentials and the atomic bomb. The energy comes with the unintended consequence of waste that will remain dangerous for millennia. It won't be different with AI. The promises include better design, better products, better foresight, and lots of energy and expense saved from running simulations in the virtual realm.
[td]
[/td]
[td]
Increasing complexity and narrow expertise can bundle up to
achieve "super-intelligence" (Graphic: Philippe Morel)
[/td]achieve "super-intelligence" (Graphic: Philippe Morel)
The risks and unintended consequences include loss of control, mass unemployment, rogue uses, and vast energy consumption from running the necessary computers and chips all to the total loss of control. For design professionals, it may be comforting that the risk of rogue abuse seems not very high in our field and that AI being truly better at creating quality spaces than the most creative minds is at least as difficult as it for AI to create better poetry than a very good Poet. In both cases, we are not yet at that point, and some say we never will since AI will never have empathy, emotion, or sentience. Given how different brain wiring is from computer networks, this may be true. However, I believe that in a process called computational illusion machines will eventually perfect a simulation of emotion that could fool us just the same. It remains to be seen if the disadvantage of unembodied AI can be overcome with AI powered robots that learn from experience in the physical environment.
AI and the Architect - Practical Uses
For now, a variety of alternative models of a proposed building program can be generated in seconds by defining a building parcel, placing it in the context of a map, entering a floor area ratio, lot coverage, setbacks, parking requirements, and proposed uses. The resulting building blocks and their arrangements can also show sun angles, shadows, and be seen from all angles.
[td]
[/td]
[td]
[/td][td]
[/td]
[td]
AI generated massing model embedded in Google Earth with
unit count and floor areas (author)
[/td]unit count and floor areas (author)
with the AI software.
The AI-generated first cast replaces the time-consuming steps of traditional design with clay, cardboard, or SketchUp, all of which visualize what a given program under certain rules would look like. To get it right, assumptions about floor plans and unit layouts had to be made, combined with calculations to arrive at floor heights, building depths, and dimensions. For mixed-use or multi-family uses, AI-based programs use standard apartment layouts to arrive at the dimensions of the proposed blocks. Of course, it can also be done for offices, factories, or sports facilities as long as there are sufficient precedents on which the AI software has been "trained."
With the help of AI, arriving at a pleasant composition, good placemaking, and an appropriate response to the surrounding areas would occur much faster. Admittedly, speed is not quality, and the time needed to build a terrain model and place cardboard volumes in it via the traditional method could assist the architect in getting a better feeling for a place, a haptic sense for the massing, and thinking about creative arrangements while performing rather basic activities like cardboard cutting and gluing. In reality, though, these banal steps towards a design were frequently performed by an intern and not by the design architect in charge. The use of AI, by contrast, will bring the actual designer into play from the beginning and allow the utilization of experience and creativity at an earlier stage, applying it to more variations because these can be created quickly and efficiently.
What applies to these early steps of schematic site design applies to every step of the architectural process. AI can assist in achieving code compliance, material and finish design while gauging embedded carbon and projected energy use every step of the way, as well as ensuring that only best practice details are applied for masonry, curtain walls, windows, roofs, etc., using a vastly larger repertoire of details than even the best-organized firms could ever have at hand in their detail library. Once the design is defined, it can be rendered in traditional rendering software or AIA rendering models.
Outlook
The current market has an array of software models for each phase, which is a bit confusing since at each step the previous result needs to be imported, and the new model must be trained on this base. So far, the underpinning of larger architectural projects is still BIM, which catalogs all building elements and allows the integration of architecture, structural, MEP, and interior design into the same model.
Architecture is a product of culture and prevailing influences within a given culture. AI would reflect on what precedents and information it has been trained on. While those limitations can be overcome in the long run, they are still a problem. Current AI design models are likely to suggest some swooping, futuristic, curvy massing reminiscent of the late Zaha Hadid, rather than solutions that resemble RIBA Gold Medal winner Yasmeen Lari's recent creations, such as the Women Center in Darya Khan, Pakistan, a cylindrical bamboo building raised on stilts to avoid flooding. Lari described it as a key example of her "second phase of life" as an architect—when she focused her attention on humanitarian work. (Dezeen).
[td]
[/td]
[td]
Yasmeen Lari Women Center Darya Khan, Pakistan
[/td]AI is prone to amplify current cultural biases until it has been trained to do otherwise. But even if the cultural bias could be overcome, current AI suggests that training on precedent results in a reproduction of precedent, even if it is with variations. While one could argue that human creativity also operates on the basis of precedence, one can certainly name genial minds that created entirely new concepts. Could AI do that, too? Many assume yes.
As it has been known since Vitruvius wrote his "10 Books on Architecture," the matter of architectural design is not just a matter of function, structural integrity, and materials, but is also a highly emotional matter, even happiness (Walter Gropius: "Architecture begins where engineering ends"). Good design affects people inside and outside buildings in many ways to the point that architecture can directly influence well-being. The best architects merge intuition, empathy, creativity, and precedents into their designs.
Some architects are sure that AI will never achieve the emotional synchronicity that a talented architect can accomplish. (See this Reddit discussion). For now, AI seems to be most useful in efficiently achieving variations of principally possible design solutions, creating visualizations and renderings and screening plans for obvious code or engineering defects. But to see AI as just another tool that accelerates the design process is flawed, per Bill Gates.
In theory, this allows the talented designer to begin the refinement earlier and at a higher level than the traditional approach would have allowed. But could the human designer be entirely replaced at every step of the way? It is unclear where AI could take us. BIG's design technology manager Oliver Thomas even speculates about some re-emergence of "artisanal architects" as a countermovement to AI.
[td]
[/td]
[td]
Hand sketch of the unbuilt monument to the
Third International by V. Tatlin
[/td]Third International by V. Tatlin
An AI ecosystem of generatively learning AI agents and robots could produce phase transitions, quantum leaps, or the type of emergence known from swarms. In emergence, it isn't the individual agent that has intelligence, but the interconnection of agents creates emerging intelligence. We have already seen such leaps when language models such as ChatGPT were suddenly able to translate between languages, an aspect they had not been trained for but that "emerged" from having been trained on several languages. It is entirely conceivable that a generative, i.e., self-learning AI design tool that has been trained in various specialties of architecture and many precedents could generate its own phase transition in which something entirely new, functional, code-compliant, and appealing could emerge that would check off all the boxes of the Vitruvian catalogue.
Thus, one day we could lose our jobs while our digital twin lives on long after our death until somebody kills it, too. If, how and when AI would replace lies in our hands as long as we humans still have control we better use it.
Klaus Philipsen, FAIA
You can subscribe to this blog by clicking the "subscribe" button the right toolbar
For this article in German see here
Resources mentioned in the article:
How MVRDV uses AI (Video)
AI and the future of architecture, Neil Leach (Video) "AI is not about form it is about data"
Opinion Piece in Dezeen (2023)
DigitalFuturesWorld
Machine Hallucinations (Co-op Himmelblau)(AD)
Refic Anadol Machine Hallucinations
LookX
The Digital Twin: Immortality Through Virtual Existence
AGI and Design Video
Philippe Morel (video) (video)
(the formal basis of architecture)
Embodiment intelligence and cognition
fei fei li spatial intelligence
Enscape: AI in Architecture: 7 Benefits and Examples
Autodesk: Digital Twins
The Guardian: It’s already way beyond what humans can do’: will AI wipe out architects?
Architizer: 15 Top AI Tools for Architects and Designers
Reddit Architects: Case Study: Can AI Architecture Design Replace Architects' Work?
ArchiStar: Building Permit Compliance Assessment
Tim Fu: Building our Future with AI
Max Tegmark (“We need a kind of FDA for AI”)