More than 70 years have passed since the first
experiments with Artificial Intelligence and, for at least twenty years,
parametric design, generative algorithms and, more generally, computation have
introduced post-human elements in the way of conceiving and practicing
architecture.
However, the
renewed interest in Artificial Intelligence differs from previous explorations
thanks to the concurrence of two fundamental factors in technological progress:
the power and speed of the latest generation of supercomputers and the
availability of an immense amount of data that has no precedents in the history
of mankind.
The flow of
data processed in real time controls many aspects of our lives: from bank
transactions to geolocation technologies, from the consumer price index to air
traffic and weather forecasts. The term Big Data refers to very large sets of
data which, when analyzed by computers with large computing capabilities,
reveal patterns, trends and associations that would remain incomprehensible if
partially analysed. In essence: if a fact (data) occurred in the past, was
recorded, filed, and can therefore be processed, under similar circumstances,
that fact will re-occur in the future. It is based on this simple principle
that the extraordinary predictive and optimization capacity determined by the
combined effect of Big Data and Artificial Intelligence can be determined.
However, a problem must be quantifiable to be
optimized. If, on the one hand, some aspects of architecture can be quantified
and therefore processed by machines, on the other, architects still rely on
knowledge deriving from experience, intuition, example or imitation of models:
in other words, they make use of that tacitknowledge described by Polanyi as «knowing more than we can say». It is no
coincidence that architects learn the secrets of their trade by frequenting the
'workshop' of the most experienced masters. In fact, not all architectural
problems are quantifiable or algorithmically solvable.
The split
between engineering and architectural skills has meant that the mathematics of
architectural design gravitated from aspects related to structural design
towards issues related to schemes, topologies, geometries, orientations and
networks. Thus, the adoption of an increasingly quantitative approach -
modularity, proportional relationships and environmental considerations, etc-
has led to a departure from an empirical approach to architecture which lacked
objective criteria and had historically characterized its indeterminacy. The
development of this measurableknowledge, algorithmically reproducible and therefore transferable into the software,
has led to a greater understanding of architectural design and its effects.
Today, the Generative Adversarial Networks (GAN) are
systems capable of processing images (with similar textual description) by
extrapolating common archetypal characteristics. In this way, GAN systems
create original images by imitation of a reference archetype. For centuries, in
the history of art as in architecture, there has been discussion about the
relationship between the original and the copy, about how to interpret la bella maniera
degli antichi, or even about the concept of style understood as a
common trait and sensibility between a group of works. The current AI-powered
software can produce, in a matter of seconds and by inputting any text, new
archetypes thanks to to the hybridization of millions of data. Jean Baudrillard
would called them simulacra: copies without
originals. Old concepts, perhaps, but some around which, in the era of
synthetic intelligence, designers may need to develop a renewed critical acumen
and unprecedented creative abilities.