Dovie Maisel, the VP of operations at nonprofit Israeli emergency response network United Hatzalah, is really excited, and I quickly understand why.

Working with graduates of the Israel Defense Force’s 8200 signal intelligence unit, the organization has developed an AI technology that can predict where and when medical emergencies in Israel are about to occur.

Wait. What?

It sounds like science fiction. Minority Report, perhaps, only this time instead of predicting future crimes and tracking down the criminals about to commit a crime, this technology predicts future medical emergencies, enabling first responders from United Hatzalah to arrive at likely problem spots, even before the heart attack, or road accident, or any number of other emergencies have happened.

He gives me an example. When he came on shift one evening, the dispatcher told him to go to the mall in Rishon Lezion, because the AI system was showing that something might happen there.

Even before he had pulled up at the mall, his app went off saying there was an emergency call just 400 meters away — a patient suffering chest pains. Maisel arrived at the location before the caller had put down the phone, surprising both of them, and managed to save the patient’s life.

“I couldn’t believe it,” he says. “It took me one minute. This is the dream come true. It’s the epitome of saving lives.”

Over a three month pilot, the organization says the AI system increased United Hatzalah’s accuracy rate by 85% in urban areas and response time to 90 seconds. 

Necessity is the mother of invention

As with most new inventions, the decision to incorporate AI into United Hatzalah’s operations came out of harsh necessity. 

When October 7 happened last year, almost overnight United Hatzalah lost 1,000 of its 7,000 volunteers to reserve service in the army.

“It massively affected our response capabilities, and I was trying to think of ways we could bridge the gap. To train more volunteers would take anywhere between six to nine months, and that was just too long,” he explains.

He started sending out requests for help on WhatsApp groups, saying he wanted people who could help him forecast where the next call would come. Some former graduates of 8200 answered his call and volunteered their experience in AI predictive modeling. 

Using 18 years of data on United Hatzalah calls and considering weather, time of day, season and other factors, it took the developers a month or so to get the first model prepared, and another few months to present a usable and effective system that would enable them to start moving ambulances and ambucycles into place.  

10-minute predictions

Maisel shares his screen with me on Zoom to explain. Suddenly, I’m looking at a bunch of bubbles — red, orange and a greeny yellow, overlaid on a map of Israel. 

The red bubbles represent life-threatening calls, the orange less critical emergencies, and the yellow ones are locations where the AI predicts a call may come. The new bubbles opening up on the right are calls coming in. 

This is the United Hatzalah system live and in progress. It’s fascinating and a little bit scary, too. It’s only 2:30pm on a sunny Tuesday, and already there have been 1,100 calls. 

The AI divides up the number of alerts by sections of the hour, and makes predictions for the next 10 minutes or so. The predictions change as the hours change and the day wears on.

We look at Ashdod, Jerusalem, Pardes Hanna – yellow bubbles appear and then red or orange ones often appear nearby, sometimes within a few hundred meters. 

“The crossover is unbelievable,” says Maisel. “You can see the prediction, and then where the call is actually happening and it’s a two minute drive away.”

We look at Hadera. There is a prediction, and a call pops up nearby. In this case a car accident. 

“Look at this, it’s even better,” says Maisel, pointing his cursor on the screen “Look where the prediction is, and look at the call. Across the street. Wow. Can you see that? Cardiac arrest. Odds are we have an ambulance or a car, or a motorcycle right there, right now, because of the prediction. This blows my mind.”

Today, all United Hatzalah volunteers have this app on their emergency phones, but Maisel is quick to point out that the system is most useful for people on shifts rather than casual volunteers going about their daily lives. 

Shift workers can check the phone to see where the AI predicts a call may come in, and head in that direction or wait for dispatch to send them. 

85% accurate

The AI isn’t always right. Around 15 percent of the time it flags incorrect locations. Moreover, responders have no way of knowing what emergency it might be; they just know that based on historical experience, something is likely to happen.

“It doesn’t matter if the call is a cardiac arrest or a choking victim, all that matters is that you need a medical professional right there,” says Maisel.

The system also can’t predict the unexpected. “I can’t predict where a missile is going to fall or where there’ll be a terror attack. We are scanning and researching general trends of total emergencies,” he explains. 

“We can’t guess at floods because of climate change, but can take into consideration general weather conditions that are consistent over time and affect traffic.”

In the future, however, he believes the system – which is still developing — will be able to provide more information about car accidents, which often happen at black spots on the roads.

Faster response time

So how does the new AI system impact United Hatzalah’s work?

“First of all, it minimizes our response time. We’re much more effective. An ambulance doing a shift now doesn’t need to drive 15 blocks to get the call. They’ll be driving one block. Which means they’ll be getting there faster, and they’ll be crossing less intersections.

“It means they run less red lights, which makes it safer for the public and the volunteers themselves. It’s a win-win. We get to patients faster, and are more effective at responding with existing resources. And in the meantime, the system keeps learning and getting more precise.”

In addition, it allows the organization to be more accurate about its own resources. “We learn, when demand is highest, what our weak links are, and where we need to recruit and deploy more volunteers.”

A lifetime of emergency response

Maisel, 49, has been a paramedic almost all his life. When he was nine years old, he was sitting at a bus stop outside his school in Jerusalem when he saw a little girl run out from behind him and get hit by a bus.

“I was in such shock,” says Maisel. “I just remember people running around, screaming, yelling, calling to the buildings next door because we didn’t have cellphones then. They called for an ambulance and nobody knew what to do. This poor little girl is on the ground.

“I just got up eventually and ran away. I didn’t tell my parents. And next day when I came to school, they gathered us together for a talk because it turned out this little girl was one of the teacher’s daughters. I watched her get killed.

“I just remember the helplessness of everyone there. And I said to myself, when I grow up, I’ll do something; I won’t be helpless.”

By the age of 14, he was already working as a volunteer at Magen David Adom, and later became a paramedic in the army. 

Then in 2006, he and Eli Beer founded United Hatzalah as a neighborhood-based volunteer emergency response organization that could respond faster than MDA ambulances. Beer is now the president of United Hatzalah.

Over the years, Maisel has responded to emergencies all over Israel and the world, leading disaster-relief missions to Haiti and Nepal in the wake of earthquakes.

United Hatzalah currently relies almost entirely on donations. 

Though Maisel only started his bachelor’s degree in disaster management two years ago – he has now completed it and has begun his master’s degree in the same topic — he has always loved technology. Aside from this new AI system, was instrumental in bringing in GPS tracking to United Hatzalah to speed up response times, and in developing the ambucycles – motorbikes with ER response equipment — for areas of heavy traffic. 

Where next for this new system?

Developing the new AI cost about $1 million and will require constant updates.

Though it was meant as a way to cope with reduced staff, ironically in the difficult 14 months after October 7, the number of volunteers in the organization has risen by 1,000, and now stands at 8,000 volunteers. The new system makes them more effective, according to Maisel.

He believes that other emergency services around the world may be developing similar systems, yet none is doing it at the scale or precision of United Hatzalah. 

“We are looking at sharing this globally as part of a package of communications, dispatch, control and predictions,” he says. “We’d like to be able to make income on it to support the efforts of the organization and create sustainability.”

Already the organization is talking to a few different US companies. 

“I have no doubt that this will be the gold standard because it won’t make sense not to use it. This is an opportunity for emergency medical services to create a paradigm shift on emergency response, period,” he says. 

And can it be adapted to other emergency services like the fire service or police?

“It can be applied to every emergency service that exists,” he says.