AI to the rescue A cacophony of tweets
(New Scientist Via Acquire Media NewsEdge) How to react?
When disaster strikes, call in the AI
In the chaos of large-scale emergencies, artificially intelligent software can help direct first responders where they are most needed
In the chaos of large-scale emergencies, artificially intelligent software could help direct first responders
9.47 am, Tavistock Square, London, 7 July 2005. Almost an hour has passed since the suicide bombs on board three underground trains exploded. Thirty-nine commuters are now dead or dying, and many more are badly injured.
Hassib Hussain, aged 18, now detonates his own device on the number 30 bus?- murdering a further 13 and leaving behind one of the most striking images of the day: a bus ripped open like a tin of sardines.
In the aftermath of the bus bomb, questions were raised about how emergency services had reacted to the blast. Citizens and police called emergency services within 5 minutes, but ambulance teams did not arrive on the scene for nearly an hour.
As the events of that day show, the anatomy of a disaster?- whether a terrorist attack or an earthquake?- can change in a flash, and lives often depend on how police, paramedics and firefighters respond to the changing conditions. To help train for and navigate such chaos, new research is employing computer-simulation techniques to help first responders adapt to emergencies as they unfold.
Most emergency services prepare for the worst with a limited number of incident plans?- sometimes fewer than 10?- that tell them how to react in specific scenarios, says Graham Coates of Durham University, UK. It is not enough, he says. "They need something that is flexible, that actually presents them with a dynamic, tailor-made response."
A government inquest, concluded last month, found that no additional lives were lost because of the delay in responding to the Tavistock Square bomb, but that "communication difficulties" on the day were worrying.
So Coates and colleagues are developing a training simulation that will help emergency services adapt more readily. The "Rescue" system comprises up to 4000 individual software agents that represent the public and members of emergency services. Each is equipped with a rudimentary level of programmed behaviours, such as "help an injured person".
In the simulation, agents are given a set of orders that adhere to standard operating procedure for emergency services - such as "resuscitate injured victims before moving them". When the situation changes?- a fire in a building threatens the victims, for example?- agents can deviate from their orders if it helps them achieve a better outcome.
Meanwhile, a decision-support system takes a big-picture view of the unfolding situation. By analysing information fed back by the agents on the ground, it can issue updated orders to help make sure resources like paramedics, ambulances and firefighters are distributed optimally.
Humans that train with the system can accept, reject or modify its recommendations, and unfolding event scenarios are recorded and replayed to see how different approaches yield different results. Coates presented his team's work at the International Conference on Information Systems for Crisis Response and Management in Lisbon, Portugal, last week.
That still leaves the problem of predicting how a panicked public might react to a crisis?- will fleeing crowds hamper a rescue effort, or will bystanders comply with any instructions they receive?
To explore this, researchers at the University of Notre Dame in South Bend, Indiana, have built a detailed simulation of how crowds respond to disaster. The Dynamic Adaptive Disaster Simulation (DADS) also uses basic software agents representing humans, only here they are programmed to simply flee from danger and move towards safety.
When used in a real emergency situation, DADS will utilise location data from thousands of cellphones, triangulated and streamed from masts in the region of the emergency. It can make predictions of how crowds will move by advancing the simulation faster than real-time events. This would give emergency services a valuable head start, says Greg Madey, who is overseeing the project.
A similar study led by Mehdi Moussaïd of Paul Sabatier University in Toulouse, France, sought to address what happens when such crowds are packed into tight spaces.
In his simulation, he presumed that pedestrians choose the most direct route to their destination if there is nothing in their way, and always try to keep their distance from those around them. Running a simulation based on these two rules, Moussaïd and his colleagues found that as they increased the crowd's density, the model produced crushes and waves of people just like those seen in real-life events such as stampedes or crushes at football stadiums (Proceedings of the National Academy of Sciences, DOI: 10.1073/pnas.1016507108). The team hope to use their model to help plan emergency evacuations.
Jenny Cole, head of emergency services at London-based independent think tank The Royal United Services Institute, wrote a report on how the different emergency services worked together in the wake of the London bombings. She remains "sceptical" about these kinds of simulations. "No matter how practical or useful they would be, there's usually no money left in the end to implement them," she says.
For his part, Coates says he plans to release his system to local authorities for free as soon as it is ready. n
In the chaotic moments after disaster strikes, people often turn to Twitter for information. But making sense of a flurry of Twitter posts can be difficult.
Now Jacob Rogstadius at the University of Madeira in Portugal and his team have developed a system that sorts updates from Twitter by keyword - for example "Japan" or "earthquake" - and places them into an event timeline, without the need for hashtags.
In the next phase of development, people will look at tweets clustered in this way to judge the pertinence and reliability of different sources of information, or request more?- pictures of the area, for example - to create a virtual "incident room" as the crisis unfolds.
(c) 2011 Reed Business Information - UK. All Rights Reserved.
[ Back To TMCnet.com's Homepage ]