AGI Can Make Smart Cities Even Smarter

The city is not just a collection of buildings, neighbourhoods, and roads.

It is a “living thing,” ever-changing, constantly growing and re-forming. Its population – both numbers and composition – is constantly changing, which means that the services residents need have to constantly be adjusted. As cities have gotten bigger and more complicated, administrators have increased the use of technology to manage them – and that includes utilizing artificial intelligence to manage trash collection, traffic flow, energy usage, and much more. AI turns cities into Smart Cities, enabling administrators to manage them much more effectively and efficiently.

But machine-learning based AI can only do so much. Like any living organism, cities change on a moment-to-moment basis, and by the time administrators parse the gathered data, the policy changes they might implement based on that data could easily be outdated or irrelevant. And because machine learning-based AI analysis systems need previous data in order to learn how to cope with a situation, sudden new events, like an unprecedented storm, mean that machine learning-based AI systems may not provide much insight into how to cope with the situation.

In order to manage situations like that – as well as the countless other novel issues that can crop up at any time – city administrators should step up their AI implementation, deploying systems that are capable of Artificial General Intelligence. Unlike standard machine learning-based systems, which analyze data in terms of patterns discovered once-and-for-all during a singular training phase, AGI systems continue building and developing those patterns on the fly, while ensuring that the patterns they create “make sense.” Thus, as new data is input into the system, AGI-based systems can adjust their understanding of the new situation – and respond accordingly.

Goals are described in a specialized format used by the AGI system to make its decisions. In a smart city scenario, the goals could include dozens of parameters, such as preventing traffic jams, ensuring efficient trash collection, deploying of police at specific times and places, identifying available housing units, providing insights into usage of electricity and other resources, pedestrian traffic in specific areas, the most opportune times to shop or commute to work, and many more. The AGI controller considers all the resources available and matches them to goals as they arise – while constantly learning based on results how best to meet the goals. As it learns, the controller develops a plan laying out the specific steps needed to achieve its goals. The advantage in an AGI system is that if the model and/or goals change in real-time, the controller can adjust its use of resources in order to ensure that the goals are met. The owner can change the description of the goals at any time, with AGI capable of “work on command.”

Thus, if there is a bus accident at the scene of a major sports event, with dozens of people injured and traffic snarled – an unprecedented event that is unlikely to be contained in the model – the controller will be able to automatically reassign resources, including rescue personnel, traffic officials, traffic light changes, and much more, in order to lessen the impact of that emergency and ensure as safe a situation as possible.

While the incident itself may be unprecedented, specific aspects of it – traffic jams, overcrowding, the need for emergency medical treatment – will be familiar to the models that were utilized to enable the AGI system to achieve its goals of being able to develop responses to novel situations. This process, called compositional reasoning, enables AGI systems to automatically develop responses for unprecedented events and situations that don’t appear in a training set. The system compares the “primitive causal elements” of these situations based on a newly-generated combination of previous elements, and composes a new, consistent set of such elements, thus building a model of the new situation.

Another important area where AGI can help cities is managing energy use, providing insights and taking automated action in response to fluctuating demands. AGI controllers can manage deployment of electrical resources – usage of the grid, drawing electricity from other power districts when needed, and other factors – in order to ensure that power levels remain robust enough to meet demand. Here, too, AGI systems examine and deconstruct previous models in real-time, composing new models in response to the streaming data.

AGI systems can even be programmed to choose the best sources of power, including “deciding” to first utilize power from sources that have the smallest carbon footprint, and only after that drawing from other sources. It all depends on the goals set by the owner of the AGI system, who can change those goals at any time, via a control panel that even personnel not trained in AI R&D can utilize.

AGI can accomplish all this thanks to categorical cybernetics, an advanced field of mathematics which studies processes interacting bidirectionally with both an environment and a “controller.” A constant exchange of information between the human-owned controller and the owner, including novel inputs and changed goals, ensures that the system remains “on mission.” Inferences are translated into actions achieving the goals set forth by the owner (a “coherence condition“), with models for AGI based on human thought patterns.

Smart cities are seen as the Next Big Thing in urban living – with advanced technology ensuring a high living standard for growing numbers of residents. Like with many other things, artificial intelligence can help achieve that goal. By gathering and analyzing data, AI systems can ensure that city services are delivered in the most effective and efficient manner possible. But AGI systems enable all of that to happen in real-time under constant and unforeseen change – guaranteeing that residents get the most out of their urban living experience.

About the Author

Ralf Haller is Executive Vice President at NNAISENSE. NNAISENSE leverages the 25-year proven track record of one of the leading research teams in artificial intelligence to build large-scale neural network solutions for superhuman perception and intelligent automation, with the ultimate goal of marketing general-purpose AIs. The company is an outgrowth of the internationally renowned Swiss AI Lab, IDSIA, headed by Dr. Jürgen Schmidhuber, which has been at the forefront of scientific breakthroughs in artificial neural networks, deep learning, reinforcement learning, artificial evolution, and general purpose AI, since the early ’90s.

Featured image: ©Zaleman