Hurricane Patricia is one of the strongest storms ever recorded. In October 2015, it grew from a tropical depression into a category five hurricane in a matter of hours, leaving businesses and cities unprepared and individuals vulnerable to harm.
However, roughly 150 miles inland, weather prediction models based on artificial intelligence (AI) provided advanced warning to an IBM production center in Guadalajara. The system had analyzed huge volumes of disparate information, including weather data, social feeds and news reports to get a comprehensive view of the storm's trajectory. And even though the system accurately forecasted that the cyclone would track far enough north to avoid the plant, the company opted to evacuate the site as a precautionary measure. Early warning gave site officials the crucial time they needed to act.
Life-and-death decisions like this are never easy, but cognitive computing can combine machine and human capabilities to give officials greater insight as they make these difficult calls. Early adopters are already using these new tools to combat epidemics, manage disasters and fight crime. Through better situational awareness, our protectors can protect us more effectively. However, as these solutions permeate the public safety world, they’re also sparking important conversations about privacy, policy and trust.
Protecting public health
Public health officials are constantly on the lookout for the next pathogen that could cause an epidemic. As population, urbanization, travel and interaction between humans and animals rises, there’s more opportunity for diseases to spread and become public health threats.
Generally speaking, health agencies typically have a huge amount of data surrounding the spread of infectious disease, including incident reports, lab data and pathogen analysis. However, since this data has different sources and is collected differently, it’s difficult to integrate and often goes unused. Complicating matters, there can be a significant lag between when symptoms are first reported and when agencies can identify a problem and take action.
To better forecast and mitigate outbreaks, government agencies and health organizations are turning to cognitive systems. The Shenzhen Center for Disease Control and Prevention (CDC) in China is a perfect example. Roughly 500,000 people worldwide die each year from complications caused by the flu, making the seasonal flu a major health concern in many countries, including China. The Shenzhen CDC is bringing its disease data together along with meteorological, air quality, geographic, economic, population and population mobility information into a single big data platform to forecast outbreaks. Through an ensemble of machine learning models that adapt as the data changes over time, the Center is gaining new insights to better control the spread of disease and provide health advice to citizens.
Foodborne illness is another public health risk that cognitive capabilities could help thwart. According to the U.S. Centers for Disease Control and Prevention, one in six Americans (roughly 48 million people) get sick from food poisoning each year, resulting in roughly 3,000 deaths. To identify a widespread foodborne illness, public health analysts traditionally look at data that bubbles up from doctor visits, which can take a significant amount of time. With the help of sophisticated detection systems that can analyze and understand unstructured data from real-time social media posts and local news, experts can pick up signals and see patterns, accelerating their ability to pinpoint sources.
Scientists now estimate that 75 percent of new or emerging infectious diseases in people come from animals, making the study of zoonotic diseases critically important to protect human health. The Cary Institute of Ecosystem Studies is harnessing the power of artificial intelligence to predict the spread of diseases that originate in wildlife and make the leap to humans. Zika virus is the focus of a collaboration with IBM's Social Good Program. Using data from disease-carrying primate species in Africa, machine learning models have been trained to identify primate species most likely to harbor Zika in other regions of the world, such as South America and Asia. By pinpointing areas vulnerable to Zika outbreaks, health organizations can plan interventions to help prevent the spread of disease.
As humans encroach into the wild, there are many more interactions and opportunities for diseases to infect humans, and once they’re in humans, to spread widely. These last few years have shown how important it is to accelerate disease ecology research.
– Kush Varshney, Research Staff Member, Manager and Co-Director of Social Good Program, IBM Research
Mitigating and responding to emergencies
When dealing with complex emergencies, it’s extremely difficult to plan and execute rescue and relief efforts. As any emergency responder will tell you, the last thing they want is to go in blind. They need an accurate and detailed understanding of the situation on the ground. Drones equipped with cameras and cognitive visual recognition and analytical reasoning capabilities can give command centers an “eye in the sky” to help optimize response efforts.
When a first responder is on the way to the scene of an emergency, they have all these inputs they're getting from disparate agencies. It's this big, convoluted puzzle of data. Cognitive technology can pull all of this together and make sense of it to provide much richer situational awareness.
– Dr. Gary Nestler, Global Public Safety Segment Leader, IBM
As an emergency unfolds, the ability to glean information from people on the scene is important, however, dispatch centers are often overwhelmed with high call volumes. "If there's a train derailment, emergency lines might get 500 or more calls flooding 911," Dr. Nestler says, "but you only need two or three to get the necessary information.” Cognitive systems with speech and natural language processing capabilities can help dispatchers manage calls more efficiently so that critical information gets passed through faster to first responders.
When paramedics, police officers and firefighters are in the midst of managing an emergency, they don't have the luxury of diverting their attention from their environment. Their eyes and hands are already engaged. That’s why advances like voice recognition, tone analysis and gesture recognition are so critical. They allow emergency responders to more seamlessly interact with cognitive systems through equipment they’re already wearing – like microphones, earpieces and watches – without having to stop and look at a screen or type in a response.
In addition to helping public safety officers respond to emergencies, cognitive technology can assist with averting disaster in the first place. For example, DRONEBOX manufacturer H3 Dynamics is aiming to use a cognitive Internet of Things platform and vision capabilities to help safety personnel monitor high-risk environments like dangerous chemical storage sites, nuclear facilities or off-shore oil drilling platforms.
Dormant drones in a box are like those fire extinguishers that sit in a glass cabinet: you ‘break the glass’ in case of an emergency. You now have an always-on drone that’s sitting near or within a sensitive facility and can be deployed within seconds.
– Taras Wankewycz, CEO, H3 Dynamics
To help reduce the odds of railway accidents, a railroad company is piloting a visual recognition solution that monitors trains in motion. Through stationary cameras and machine learning algorithms, it analyzes images as railroad cars pass to help spot potential equipment problems before an accident occurs. In addition to helping prevent accidents that could harm the general public, companies are using cognitive capabilities to reduce job-related risks. For example, Aerialtronics uses cognitive-enabled drones to help eliminate the need for dangerous tasks like manual cell tower, wind turbine and oil rig inspection.
On average, landscape fires claim the lives of 340,000 people each year. With improvements in prediction models, officials can better anticipate where fire risk is highest and take action to help prevent loss of life. For example, researchers have developed adaptive data-driven models that can help mitigate the impact of wildfires by predicting which areas are at greatest risk of a fire spreading. These new models use spatial-temporal patterns, weather data, and terrain and vegetation knowledge to add greater detail to threat ratings. They also bring in data like season and time of day to calculate risk in context.
Today’s fire-risk ratings are available only at a very coarse level, typically for an entire town. Data-driven machine learning models can pinpoint risk to a particular property. "In Australia, a single region can be hundreds of kilometers wide," says Anna Phan, IBM Research Scientist. "You might have very dry farm land in the same region as alpine areas with snow.”
Citizens need specific information about how much danger they're actually in if a fire were to spread. This depends on many more factors than just the region they live in. “If you could find out what the risk is for your own particular property on a particular day,” Phan explains, “you could also figure out how to reduce that risk." Fire-fighting agencies could also use this type of information to plan evacuations and optimize the placement of crews and equipment.
In law enforcement’s decision-making process, cognitive technology offers advantages that can help compensate for human limitations. It doesn't sleep. It has no emotion. It isn’t bound by volume or memory. It can read millions of documents in seconds to make connections and expose patterns that are difficult for people to discern. It also allows us to scale human expertise, by bringing together the collective knowledge of an entire police department or government agency.
Situational awareness is remarkably important in policing. When officers first arrive at a scene, they observe and orient themselves before deciding on a course of action. In these initial moments, cognitive systems have the ability to piece together disparate information that can provide law enforcement crucial information. For example, it could warn them of past violence or drug presence at this location—or that a house in this neighborhood is actually a learning center full of small children.
The holy grail for national security analysis is to understand the intent and motivation of a threat entity. Not only does it help to anticipate what's going to happen next, it provides clues to the appropriate response and countermeasure.
– Juliane Gallina, Partner and Director, IBM Cognitive Solutions for U.S. National Security and Justice
In addition to providing situational insights to officers in the field, cognitive capabilities can assist investigators working to solve criminal cases. For example, the Law Enforcement Analysis Portal (LEAP), composed of a coalition of U.S. law enforcement agencies, is piloting a cognitive-enabled service that helps officers identify probable locations where suspects might be found.
In the past, when officers created search warrants, they had to rely on information searchable in the department’s structured database. But through cognitive capabilities, LEAP is now able to tap unstructured data from dispatch systems and officer notes from speeding tickets, domestic violence incidents, arrests and other interactions with the police department. In addition, the solution searches open sources, such as credit agencies, social media, and vehicle history and license plate image databases. With access to more data, officers can assemble a richer, more accurate view of where they might locate a particular individual. For example, notes from a traffic stop might indicate that a suspect is married, or a credit report could show that this person lives with an extended family member.
The value of cognitive technology lies not just in making us more effective but also in enabling us to do things we’ve never done before.
– Randy Hunt, LEAP Director
In future phases, the solution will help crime analysts and officers find non-obvious relationships across these data sources. For example, the system might discover that the address of a suspect is the same as a 63 year old woman who owns a particular car and that this car was spotted recently in a specific neighborhood. “To identify potential threats or suspects in a case, investigators have to research an overwhelming volume of material,” says Richard Varos, VP of Global Government Market Segments, IBM Public Sector. “If a cognitive assistant can read and form hypotheses, it can help investigators get to high probability pursuits more quickly.”
Privacy, policy and public debate
Policies and laws must strike a fine balance between protecting individual privacy and keeping the public safe. This affects how cognitive solutions are designed. For example, in the United States, intelligence data cannot be comingled with law enforcement data, meaning official facts from legal records (e.g., arrest reports) cannot be combined with subjective, open source data uncovered in an investigation (e.g., clues from social media accounts). As a result, investigators may need to look at two separate sets of analysis rather than one.
With cognitive capabilities, though, it’s not always an either-or choice between personal privacy and public welfare. "The beauty of cognitive tools is their ability to discover and characterize patterns and relationships,” explains Gallina. “We combine these tools with analytic approaches to identify threats and vulnerabilities while still protecting the privacy of individuals. The idea is to give government and law enforcement officials the tools that help them anticipate events, so they can intervene and support individuals and communities before threats manifest themselves.”
The public safety officials who use cognitive technology must also trust it. Ultimately, it's the emergency responder, health official or criminal investigator who must live with the decisions he or she makes, regardless of who or what informed those decisions. “Just because a machine is telling you to do something, that doesn't mean you should,” Dr. Nestler says. “We need to first be sure we absolutely trust the integrity of the data, and the tool itself.”
I trust humans. Making sure we have that combination of human and machine is the ideal approach. I still feel more confident knowing that a human is there alongside the computer. That's what augmented intelligence is all about.
– Richard Budel, Government Leader, IBM Cognitive Solutions
It’s somewhat ironic that the media so often focuses on the potential dangers of AI and the threats it may pose to society when, in reality, this technology offers so many new capabilities to help keep the public safe. Cognitive capabilities can be a force multiplier, extending and augmenting what our human protectors can do on their own. “Whether we’re creating algorithms to model the spread of flu or determining how weather patterns might impact disaster response activities,” says Diane Melley, Vice President, IBM Corporate Citizenship & Corporate Affairs, “this technology is affecting the lives of people in a very positive way.”