How many times have we seen a weather forecast fail completely?
How many times have we had some sort of natural disaster that was supposed to be completely out of the question for that particular place, during that particular time?
What if something could fix these occurrences to the point that we could better predict the coming of phenomenons like out of season hurricanes?
The United States Department of Homeland Security has actually developed a pilot program that at least, begins to address these questions. In it, they are demoing the usage of automated sensors to give better flood warnings in locations like Ellicott City, Maryland. These sensors, however, are less related to AI and more related to an application of the Internet of Things.
This is not to say that AI is not being applied even now, to the same issue. The Chesapeake Conservancy, an environmental group based out of Annapolis, Maryland, has created an AI aided map with the special skill of being 1,000 times more precise than the average map that flood planners currently use to prepare cities for serious natural disasters.
How does it do this?
Reportedly, the map’s usage of AI allows it to show objects as small as 3 feet square as well as in 1,000 times more precise of a scale than such maps without AI can currently do. By all appearances, what this comes down to is that AI allows these specialized cartographers to really zero in on areas that could be the most vulnerable when these disasters hit.
Even so, such a project is not without its’ drawbacks. Creating such a map cost the team $3.5 million up front, which included help from Microsoft and other companies. The partnership with Microsoft on getting the map ready to use was actually where its’ AI system came from. According to a Wired article on the subject, Microsoft assisted Chesapeake Conservancy with training special algorithms that helped it even further in accurately identifying certain objects without human help. In another sense, there was no AI before Microsoft’s involvement.
As to what these objects actually were, the article is somewhat unclear except that there were a large number as well as many variations of them. In order to do this training, however, Microsoft apparently had to employ a special technique called “pixel peeping,” which Wired describes as: “manually zooming in on objects to verify and amend the automated results.”
The way that it does so is really just a matter of repetition. According to Wired, each time that the AI zooms in on an object, it learns more about it. In a more specific sense, the prevailing theory is that the more an AI looks at an object, the better it gets at correctly recognizing it as well as its most important properties. The downside with such a process is that according to the Chesapeake Conservancy, it is far from automatic work. The team around the AI has to engage in pixel peeping manually, inputting training data as it goes, in order to improve the system’s performance. Even so, by all accounts, this is an improvement in terms of how quickly and efficiently disaster preparedness teams can map out potential problem areas before disasters hit.
In the future, it is reasonable to assume that all eyes will be on this system as it develops, especially to see if it eventually engages in what may be called unsupervised learning instead of supervised learning.
Resources:
Primary Source:
https://www.wired.com/story/how-artificial-intelligence-could-prevent-natural-disasters/
Chesapeake Conservancy: https://chesapeakeconservancy.org/
Using AI for Emergency Management:
