Written by Naysan Saran | 2020
Artificial Intelligence is Everywhere
Do you take pictures with your phones? An algorithm is used to optimize the settings. Are you thinking about a career change? Your next interview will probably be conducted by AI. Each day, new models are being developed to optimize fields as varied as health, finance and the arts. Moreover, the book “Deep Learning”, one of the bibles of deep learning, was translated from English to French… mostly using a deep learning algorithm. There are certain areas where AI’s progress is slower. Among these, is the management of our most precious resource: water.
When it Comes to Water
According to the Financial Times, water will be the petroleum of the 21st century. However, invisible problems are rarely a priority: between 30-50% of this precious resource is lost before even reaching our faucets. Here’s why: in big cities, the majority of water pipes that were installed more than 70 years ago, are reaching the end of their lifespan, causing more and more leaks and pipe breaks.
According to the EPA (Environmental Protection Agency) the required cost for replacing these pipes is up to 500 billion dollars, which will primarily be financed by governments, and thus will come out of our taxes.
People need to get ready to pay this enormous bill over the course of the next few decades. To avoid increasing our taxes more than necessary, it is crucial that we use the available data to optimize the maintenance on these water networks.
Keeping in mind that the majority of cities already collect data on their infrastructures, what would slow down adopting artificial intelligence in this field?
Case Study
Since 2017, CANN Forecast has used city data to help it best manage its water resources. Last year, our team created the first pancanadian research project which aims to apply artificial intelligence to optimize pipe replacement for the drinking water network.
In addition to McGill University and INRS, nine municipalities across Canada joined us in this challenge. Though the end of the project is planned for summer 2020, a few trends have already appeared. In fact, even some of the results are very promising, certain issues must be addressed in order to allow the algorithms to reach their full capacity, both in terms of data and interpretability
What About the Data?
According to current practice, data scientists spend about 80% of their time formatting data. This is definitely the case when it comes to water management as the data manipulated comes from a wide variety of sources with very little standardized formatting (Esri database, proprietary software, PDF maps, free-hand surveys…).
In the mean time, we are developing an automatic data processsing tool based on AI in collaboration with the Centre d’Expertise en Infrastructure Urbaines of Quebec.
The Interpretability Question
This question is applicable in two ways: the first is model interpretability. Often, with AI, new models are developed with the sole purpose of having better results than their predecessors. In this race for higher performance, few researchers take the time to understand what is happening inside of their blackbox.
However, with regards to water management where every bad decision can have crucial impacts on the health of citizens, few institutions are ready to trust opaque systems. Interpretability of algorithms is a domain that is relatively new to AI, and we must accelerate it’s expansion.
The second factor is about the interpretability of results. Data scientists estimate the success of their algorithms in a jargon that only they speak: accuracy, train/split, cross-validation, etc. On the other hand, municipal institutions want to know how many years it will take to make up for their infrastructure deficit. In order for these two to communicate, performance metrics need to be translated into a measurable ROI for municipalities, which is not easy to do in terms of risk management.
Water management is a complicated challenge, and recent progress in artificial intelligence could clearly help with part of this. But even today, some reactions remind us of Philip Thicknesse’s reaction when faced with the Mechanical Turk. However, realistically speaking, we are in the AlphaGo era. In conclusion, even if AI has come a long way, there are still several challenges to address. This article presents some food for thought for us to be able to fully take advantage of the data revolution to optimize one of our most precious resources: blue gold.
