An international group of researchers is warning against the creation of AI-led autonomous cities, outlining fears that diverse AIs may eventually perform social and managerial functions in an unsupervised manner without human input. 

Research concerns

  1. Competition between humans and AIs for control.
  2. The unsustainable nature of such cities due to energy intensive supply chains.
  3. The potential for racist and sexist characterisations.
  4. Whether AI and human ideals will align (eg in assessments of justice and equity).
  5. Whether the slow pace of urban planning can keep up with rapidly developing AI technologies.

The research, led by Dr Federico Cugurullo from Trinity College Dublin’s School of Natural Sciences, has recently been published in the journal Urban Studies. The article can be read on the publisher's website

“Autonomous cars, robots, city brains and urban software agents are technologies that already exist and populate urban spaces, and their impact on urban planning, urban governance and urban living is transcending well-known practices of smart urbanism. This is not the usual smart city. This is an unprecedented type of urbanism. One that we are losing control of, given the growing agency of AI,” said Dr Cugurullo.

“Urbanism driven by urban AIs is already solving some city problems but it is also giving – and will continue to give – rise to new practical and philosophical challenges. By transcending the smart city, the rise of AI urbanism might lead in the future to the formation of the autonomous city in which humans do not supervise decisions, and governance is in the hands of AIs.

“It might sound like science fiction fantasy, but AI urbanism is not far away.”

The seeds of the autonomous city have already taken root in experimental urban projects like Neom in Saudi Arabia where a robot was granted citizenship in 2017, but the future emergence of this model of urban development remains an open question.

Enormous generators of data

Urban AIs need cities to improve their intelligence through the acquisition of data. In terms of quantity, cities are enormous generators of data, because they are the primary locus of human societies, where intense and manifold human activities occur.

And in terms of quality, cities offer the best possible data for AI to learn from: real-life data, as opposed to carefully curated and cleaned datasets that are modelled on a computer.

Dr Cugurullo added: “Urban AI learns in the wild by observing the messy dynamics of real-world urban environments. The life of urban AI intersects with the life of cities because these are AI technologies designed and programmed to interact with humans and to mediate socioeconomic activities that occur primarily in urban spaces. This includes algorithms that calculate if you qualify for a home mortgage, or who is likely to commit a crime in the city.

“Ultimately, human agency might be overshadowed by the agency of urban AI, which points towards a distinctly different type of city than exists today and raises questions about the opportunities and drawbacks presented by the contemporary roll-out of AI and its radical impact on urban futures.

“What we are saying is that AI needs the city, but we question whether or not the city really needs AI.”