Smart cities, where digital technology and data are used to enhance economic efficiency, quality of life, and sustainability, are becoming the focus of urban debates. Artificial intelligence (AI) is at the forefront of these discussions, with policymakers and urban planners seeking technological solutions to address challenges related to urbanization, such as climate change and mobility. The use of big data and AI-powered urban technologies holds the promise of improving the efficiency of urban infrastructure and services, thereby helping to mitigate these issues. AI has numerous applications in urban areas, including automated decision-making processes, infrastructure assessment, post-disaster reconnaissance, autonomous mobility, urban analysis, and robotic and chatbot services.
However, despite government efforts, many of these initiatives have failed to generate ethical and sustainable solutions. Current practices are often reductionist and technologically driven, overlooking social and urban complexities. Iconic projects like Toronto Sidewalk, Masdar, and Songdo have disappointed in delivering on the promises of smart urban transformation. In response, an innovative approach is proposed - green AI - which moves away from immediate solutions to focus on an ethical and sustainable practice that promotes a fair and sustainable urban future.
The adoption of intelligent and innovative digital technologies has become common to address urban crises related to climate, pandemics, natural disasters, and socioeconomic factors. AI offers opportunities to increase infrastructure efficiency and improve the quality of life in smart cities. However, AI poses risks such as opaque decision-making and privacy violations, leading to discrimination and lack of transparency, not to mention the crucial challenges related to data. AI-related incidents, such as algorithmic biases and privacy breaches, undermine public trust in AI solutions in smart cities.
Many examples highlight the issues related to the use of AI in urban and public contexts. For instance, in Pittsburgh, AI provided incorrect diagnoses of child abuse, while Amazon's recruitment program showed a bias towards men. Additionally, algorithms used in American hospitals have been accused of discriminating against people of color. Equally problematic were the Clearview AI scandal and the controversial Robodebt program in Australia.
It's clear that the development and integration stages of AI are critical for the success of public services. As highlighted by Pasquinelli, it's important to consider the various types of biases that can influence AI, such as those related to the concept of a "just world," data quality, and the algorithms themselves. Integrating such biases into public services can inevitably lead to failures or dissatisfaction, undermining citizens' trust in AI and its potential benefits.
Human activity has caused severe damage to the environment, resulting in effects such as chemical pollution, biodiversity loss, and climate change. These threats, along with pandemics and weapons of mass destruction, highlight the global failure to prevent risks. This problem is the primary cause of the failure of AI solutions, which often focus on economic productivity rather than addressing environmental threats. Humanity has thrived due to favorable planetary conditions, but now it's necessary to examine how AI can contribute to ensuring a balance between humans and the environment in the Anthropocene. This topic is of growing academic interest.
Goranski and Tan have examined how AI can accelerate the progress of the SDGs, showcasing its potential to generate targeted data and reduce waste. Although AI offers opportunities for the SDGs, significant investments and international collaboration are required to ensure governance and security. Academic literature has documented an increase in AI applications to address environmental and social challenges, such as air pollution, biodiversity protection, and natural disaster prediction.
Although the benefits of AI are manifold, it's essential to consider the most common negative effects on the environment, which include increased electricity usage (energy consumption for computations and transmission) and resulting carbon emissions, along with errors in critical decisions due to user and data biases. Given the exponential growth in global technology adoption, the impact of these externalities is expected to be significant. For example, in recent years, cryptocurrency mining has led to increased energy consumption globally. According to Cuen, the energy usage of Bitcoin, as measured by the University of Cambridge, is significant, corresponding to approximately 0.6% of global electricity consumption. This is equivalent to the CO2 emissions of developing nations like Sri Lanka or Jordan.
These undesired effects underscore the need for a sustainable approach to AI, which includes green technological infrastructure.
Dobbe and Whittaker have put forward suggestions for a climate-conscious technology policy, which takes into account the entire technological ecosystem, addresses the impact of AI on climate refugees, and restricts the use of AI for fossil fuel extraction. Making AI sustainable requires a bias-free and responsible approach aimed at addressing development challenges in an eco-friendly manner. This would not only benefit the environment but could also catalyze the transformation of cities towards a smarter model.