Imagine a world where the relationship between humans and machines is interdependent and symbiotic — where a machine’s abilities are sophisticated enough to complement and enhance human work, allowing us to pursue goals that neither we nor they could achieve alone.
This world is not 10, 20 or 30 years away. It’s the world we live in today, a world which impacts our work, our communities and our cities. Across our industry, human-machine interdependence is generally improving outcomes for most stakeholders — owners, operators, clients, employees and communities. But outcomes are not always positive, and there are new challenges on the horizon.
The role machines play in our lives
From asset design to construction, operation and maintenance, machines play an increasingly central role in urban life.
Consider a series of technology layers that enables this central role for machines: a sensor/perception layer for capturing data; a network layer enabling connectivity; a data processing layer that includes computing for turning data into insight. Or an application layer — specific applications built on top to execute specific actions. When we talk about machines’ capabilities, it is helpful to consider these more nuanced frameworks.
Where then are machines and humans working symbiotically today, and how do the use cases relate to our cities?
When we leverage Dynamo, Grasshopper and Generative Components to facilitate development of our designs, we leverage algorithms, computing power and parameters to complete the deliverable. We’re working collaboratively with a machine to produce a design. As we incorporate more complex functional parameters — perhaps at a street, community, or city level — these machines will likely design built environments that look dramatically different.
We rely increasingly on machines to drive efficiency and quality in our inspections of physical infrastructure. For example, computer vision and machine learning help automate the identification of defects in sewer systems, identify cracks and potholes in roads, and compare as-built assets to the BIM to help identify mistakes early and prevent rework.
In our cities, sensors combined with artificial intelligence (AI) and computing power optimize our traffic signals to reduce congestion and prevent accidents. They also optimize parking and utilization of curb space and parking structures, enabling dynamic use and dynamic pricing. Machine learning technology can leverage data from existing camera infrastructure to inform transit agencies, airport operators and retail companies of queue times, throughput, and provide real time alerts and dashboarding.
Meanwhile, sophisticated digital twins are being leveraged in planning new urban landscapes. The Orlando Economic Partnership has teamed with gaming company Unity to produce a 3D model of the 40-square-mile metro region of Orlando, Florida in the U.S. — part of an effort to lure potential investment in the Orlando tech hub. Digital twins go beyond investment pitches. The Singapore Land Authority has built a digital model from two million street level images and 160,000 aerial images, among other data points, to run simulations of policy options and infrastructure projects. And the model is not static; it is being remapped every two to five years.
How new technologies will impact our cities?
The rate of technological advancement is rapidly accelerating. From supercomputers like Oak Ridge National Laboratory’s Frontier and Graphcore’s “Good Computer” to foundation models that flexibly address different types of problems, our machines and their capabilities are increasingly sophisticated and continually encroaching on what were once strictly human capabilities. For more, see “How smarter AI will change creativity,” The Economist, (June 9th, 2022).
The introduction of self-supervised learning coupled with extraordinarily powerful computers has various implications for our cities and people:
Impact on work. Much has been written about the possibility that technology will make unnecessary a large percentage of human work. The more likely outcome is the creation of winners and losers. As with previous waves of automation, some workers will become redundant, including many of those in today’s “knowledge economy.” However, new areas of human expertise will be required to provide oversight, judgment and policing of the increasingly sophisticated AI.
Privacy and data security. Sensors that track vehicles across multiple intersections, cameras and associated AI that leverage facial recognition to identify individuals, search data collected by technology companies — all of these could be leveraged in ways inconsistent with the individuals’ interests. A lawsuit has been filed already based on a claim of violation of privacy laws; the company in question had amassed a database of human faces and provided access to law enforcement.
Extreme vulnerabilities. As we apply AI to management of more infrastructure assets, we subject those assets to vulnerabilities, including bad actors (cyber attackers), poor model training and catastrophic system failure with limited ability for humans to intervene. This is particularly tangible in the case of a power grid, a hydroelectric dam, a water treatment plant or nuclear plant. AI can enhance safety and improve system performance — which is why we leverage it in the first place — but the gradual loss of human control opens the door to a new set of vulnerabilities that require innovative mitigants and contingencies.
Benevolent or malevolent. Today we use AI to improve speed, quality and accuracy of activities previously executed by humans. But is there also a case for AI threatening rather than benefitting human existence. Even if the probability is quite low, the potential impact is massive, and the need for scenario planning and policy development is evident.
Mitigating against unknown threats
The benefits of our increasingly interdependent, symbiotic relationship with technology are often immediately clear; the use cases tangible and the early proof points available. The threats are more complex or unknown, and the protections and mitigants even more complex, undefined and certainly untested. What should be the early steps for decision-makers?
Enhanced focus on data accuracy and data protection. AI algorithms evolve based on the data set on which they are trained. Comprehensive and representative data sets are therefore critical to reduce the probability of algorithmic bias, and potential algorithmic harm. Hand in hand with the importance of providing comprehensive datasets is protecting that data to reduce the risk of breach or theft.
Focus on data and algorithm transparency. Despite the increasing complexity of AI, the basis of specific AI decisions should be discoverable. Certainly, transparency is difficult to achieve, so more attention is required on researching techniques to improve the transparency of AI decision-making.
Conducting a series of premortems is a valuable exercise in exploring the negative potentialities. Where a post-mortem provides insight into the reasons for a negative outcome after the fact, the premortem operates on the assumption that the negative outcome has occurred and requests the team to explain what did go wrong and generate explanations for the failure. This is a first step to developing mitigants and preventive policies.
Foster the continued interdependence and symbiosis of humans and machines. While we view the risks above as emerging from increased interdependence with machines, they emerge from machines acting independently of — or at odds with — human objectives. Abiding by principles for creating safer AI — for example the Asilomar AI Principles, the ARCC Ethical Framework for AI, or Stuart Russell’s three principles to guide the development of beneficial machines, policy makers and the AI development community would be well-served to educate themselves on principles for incorporating human-based values into the development of AI.