Artificial intelligence (machine learning, more correctly) has not proven creative (not only) when it comes to architecture; even the pioneers in the field are abandoning the projects whose ambitious visions have not been achieved and reducing their efforts to pragmatic parametric tasks. Intoduced in section (2) of this paper, models like Stable Diffusion, Midjourney, and ControlNet represent state-of-the-art AI-deployment in architecture and built environment development shortening the way from a simple sketch to materiality-rich rendering, and easing and enhancing the process of visual conceptualization and image representation in phases from the (client´s brief) ideation through sketches drafting, CAD BIM (or better an into VR/XR extended working platform) model development, over rendering to final design. Currently, whatever of the multiple applications deployed, AI is a "clever", more sophisticated pencil in an architect´s hand.

Being right and wrong at the same time, Chaillou reveals the core of the misconception when concluding his paper [329]: ... our belief is that a statistical approach to design conception shapes AI’s potential for Architecture. Its less-deterministic and more holistic character is undoubtedly a chance for our field. Rather than using machines to optimize a set of variables, relying on them to extract significant qualities and mimicking them all along the design process is a paradigm shift. Then, nesting models one after the other will ultimately allow ... to encapsulate relevant pieces of expertise. „Relevant pieces of expertise“ is the keyword, indeed. However, it is not the expertise hidden in input-output pairings but pieces of expertise representing steps and milestones of the (design) process as the AI-driven-robotics benchmarks of Figure-1, GR00T, and others introduced within the state-of-the-art (2) section suggest. The statistical approach and supervised-learned GANs, on the contrary, appear doomed to remain just an etap of the history of the deployment of AI in architecture that was treading a dead end. In this field, supervised-learned GANs exhausted their potential before they could deliver qualified results.

AI is a super-parrot: it is superb in repeating what it has learned, explains Tomas Mikolov in a chat with Dan Vavra [105]. In other words, for a Generative Pre-trained Transformer, the magnitude and comprehensiveness of the training dataset is the starting point, the algorithm is the method or, running on an artificial neural network, the tool respectively, and the computational performance is the limit.

Together with the boom of the development of "practical" AI applications offering (though not always delivering) contributions to industries, the development of the once-promising technology of virtual and extended reality (VR/XR) has met its plateau. Leaving aside entertainment and social media in the metaverse, the promises for industry, design, engineering, planning, and architecture have been deferred and the development has slowed down as money - immense money dedicated originally to XR technologies has found a more promising investment target - AI.

So far, slowly, AI has been showing helpful in architectural practice regarding imagery and parametric analyses - and failing when it comes to spatial creativity. However, what else is at the heart of architecture but physical spaces? Previous sections of the paper strived to lay a base to elaborate interconnectedness of spatiality (physical, three- or more-dimensional, when time and other-than-sight senses are included) that the Introduction features tentatively in terms of first, architectural space as such and its embeddedness in public space, second, the computer technology of virtual and extended reality that provides an unprecedented opportunity to tackle the space instantly (not only through its representations as has been the rule so far), and third, the recently disclosed essential spatiality of thinking (and of "intelligent computing" thus). Now it is time to zoom in on the third area.

Consciousness … is not merely the process of learning. It is not, strangely enough, required for many rather complex processes. Conscious focus is required to learn to put together puzzles or play the piano. However, after a skill is mastered, it recedes below the horizon into the fuzzy world of the unconscious. As Jaynes [284] saw it, a great deal of what is happening to you right now does not seem to be part of your consciousness until your attention is drawn to it. What evokes consciousness is an impulse from - or induced by contact of "a self" with the external world. Perhaps surprisingly, a physical substance proves herewith to be a prerequisite of consciousness; and not only any physical substance, but also a substance that shows a degree of independence of the world of its existence [285], but, on the other hand, it can perceive the world around, and can and wants to react to it. Posessing an individual, specific history of being in the world and experience, any human performs individual and specific relation to the world of his existence - a specific, individual consciousness. And a computer? The world is not only World Wide Web and digital data; by far.

Today, the early-adopting professionals are witnessing AI performing well concerning parametric aspects of diverse materializations of architecture (such as construction, energy efficiency, daylighting, or noise in buildings and neighborhoods) and failing to be effective and productive in conceptual architectural designing. Rather than a temporary swing in the performance of particular efforts, it is a consequence of AI applications´ developers failing to grasp and follow the starting points and the workflow of creating architecture, be it buildings or areas of the built environment.

The previous subhead demonstrated big expectations the field of AI lives with: nothing similar refers to architecture and the built environment development. As annotated in concluding the overview of AI production ecosystems, the scale of insight must change to examine this field.

Obviously, training dataset quality in terms of size, comprehensivity, and relevance is of critical meaning for an AI application´s performance. Not always but often, the objects that assemble the set represent the intellectual property of respective authors. The authors feel mishandled and affected if someone - no matter whether a human or an AI - takes and compiles their creations (or their digital representations) to put them on display or to submit them individually.

Also, VR introduces processes of AI-driven 3D model creation: HTC Viverse offers 3D model generators for the VR/XR community. AI enables text to be turned into VR 3D models, 2D images transformed into dynamic 3D models, and even extract intricate models from videos [162].

There are ecosystems of natural language processing, image and video processing, voice processing, and code or software processing and development, further robotics, and expert systems or business intelligence [99, 100], altogether represented by DALL-E (DALL-E3 newly), ImageGPT, InstructGPT and ChatGPT, Bard or Gemini, Ernie Bot, Tongyi Qianwen, Sense Time SenseChat, Bedrock, and many other tools by OpenAI, Microsoft, Google, Baidu, Alibaba Group, Amazon, also MidJourney (that released version 6 recently), Stable Diffusion (currently released version 3, which demonstrates unmatched performance on the ControlNet network, designed to control diffusion models in image generation, and LayerDiffusion that introduces latent transparency, which allows the generation of a single transparent image or multiple transparent layers, combined into a single blended image [101]), Gong.io, Tellius, OPENNN, Theano, and many other tools by multiple producers.

An artificial neural network is a collection of connected units or nodes called artificial neurons designed to model loosely how the neurons in a biological brain have been supposed to look and work. Like synapses in a biological brain, each connection can transmit a signal to other neurons. A deep neural network is an artificial neural network with multiple layers between the input and output layers; in a shortcut, a deep neural network makes machine learning deep learning [59, 60]. In essence, two computing principles apply in artificial neural networks today: feedforward computing and backpropagation. The goal is always to train the models generated to cope with the criteria typically inserted by vast sample datasets. Feedforward computing refers to a type of workflow without feedback connections that would form closed loops; the latter term marks a way of computing the partial derivatives during training. When training a model in the feedforward manner, the input “flows” forward through the network layers from the input to the output. By backpropagation, the model parameters update in the opposite direction: from the output layer to the input one. Backpropagation, a strategy to compute the gradient in a neural network, is a general technique; it is not restricted to feedforward networks, it works for recurrent neural networks (to be introduced soon) as well [61].

Since around 2010, young enthusiasts combining information technology and architecture background have been trying to turn the attention of well-doing, mostly global-star architectural studios towards AI’s potential contribution to architectural design or, better to say, disclose where such a contribution might stem from and what it might consist of. In 2020, DeepHimmelb(l)au - a video of a journey through an imaginary landscape of Coop Himmelb(l)au-like building forms - came into existence. The result of the elaboration of datasets comprising reference images of geomorphic formations on the one hand and actual Coop Himmelb(l)au projects on the other by CycleGAN and other forms of AI technologies provided "machine hallucinations" [10, 11, 12] - represented prevailingly in two dimensions, substantially lacking both spatial comprehensivity and the for architecture inherent interconnectedness of the experiential (poetic, in other words) and material attributes.

Architectural practices and the planning and management of built environment development lag in adopting artificial intelligence (machine learning techniques, more correctly), and the sparse tools and approaches implemented provide only marginal contributions. A contrast reveals not only comparing to other industries and creative disciplines but to the opportunities at hand. The paper evaluates the situation both in the context of the AI field and in the sector of architecture and the built environment, points to the causes of the sector´s current setup in terms of the starting points of creativity, the technologies used, and approaches to their development, as well as in terms of the economic, social, and political framework, subsequently introduces the opportunities to overcome the falling behind, and outlines the paths. Across the paper, the critical review applies three fundamental perspectives: authentic, poetic creativity that passes and precedes parameterization and algorithmization, second, novel, in architectural designing not yet applied learning strategies and training approaches, and third, concurrence of the fundamental three- and more-dimensional spatiality of both architecture and recently developed virtual reality technology, as well as the new theory of human thinking and intelligence that waits for implementation in machine learning (together with other novel computing approaches). Given the coincidence of the three aspects, a singularity is predicted for the next development of architectural craft and field.