Design for Life

Unlimited Complexity WITHOUT Marginal Cost

Many people are now familiar with AI-generated art from systems like DALL-E 2. Algorithmic design works similarly, using rules, constraints, and aesthetic themes to generate 2D and 3D designs -- a wind turbine blade, an architectural facade, a building layout fit for purpose.

Paired with additive manufacturing (3D printing), both design and manufacture collapse into one automated process: incredible complexity at no extra marginal cost. In the 1900s we still had a fashion for intricate design, since replaced by stark, utilitarian, mass-produced simplicity -- inoffensive and generally timeless, but also dull and soulless. We may soon see a renaissance in which plainness becomes passé in a world where beauty has become next to free.

In urban spaces, algorithmic design applies to interior and exterior building design, construction sequencing, town planning for current and projected needs, and warehousing and logistics. Some systems already predict the varying rental yields from placing apartments, shops, or offices on a given floor. These capabilities can greatly reduce the time and cost of construction.

But algorithmic design has pitfalls. Many government and financial systems built in the 1960s assumed a Dr. must be male, that married persons are always of the opposite sex, that people never change gender, that gender markers will always be M or F. Every one of those assumptions has since broken, creating immense challenges for legacy systems. The changes were unimaginable at the time, and never accounted for -- least of all under the scarce computing resources of the era.

The lesson: build flexibility into system architecture so components can be modified as necessary. Every design process has assumptions baked in, from fire regulations to the expected size and weight of human beings. Most parameters vary across time, culture, and geography, and that inevitability should be accounted for with tolerances and maintenance in mind. Machine learning models compound the problem: trained on datasets from a particular moment, they carry temporal biases -- old-fashioned impressions of a world that has since moved on.

Most of us know someone who ran afoul of a content moderation algorithm for a harmless remark -- banned for discussing a chess game of black against white, or saying they ‘shot themselves in the foot’. Sometimes an unscrupulous engineer deliberately interprets a statement in the least charitable way that suits their worldview. Where context is lost, by accident or design, justice and truth cannot prevail. Nuance must be present in all forms of deliberation, and above all in non-transparent algorithmic processes with the power to abuse us with Kafkaesque petty tyrannies.

These challenges matter especially in urban design. Machines are learning to navigate environments, but they cannot know the experience of doing so. We must never sacrifice the feel of an urban landscape to efficiency, nor disrupt any person’s enjoyment of a shared resource. The greatest question of AI is not whether we can do something, but whether and how we should -- in a manner we can trust. The actions of a system must serve human needs, not the needs of the system. The struggle to teach machines to recognise, acknowledge, and respect our values will define the decade ahead.

If we can indeed achieve that, the future seems a little bit brighter.