In spring of 2011, the headlines in western Finland blared; “Man bludgeoned with hammer by four thugs and left in the forest to die”. Against all odds the victim was found by a woodsman and miraculously made it to the hospital where he was stabilized. It wasn’t a random incident; the man knew his attackers and the police were quickly able to take the four into custody. The police then coaxed the location of the weapon out of the suspects and found the hammer where the suspects said it would be—in a river, which would be walled off by a dam in ten days.
Seems open and shut, the men could be charged, right? Not so. Each man was blaming the other as the sole criminal and unless the actual bludgeoner could be identified they would all be released after their three-day hold period had expired. Ideally DNA testing, the gold standard, would be run on the assault weapon and compared to the four, but it would take weeks to months to get a result. By then the criminals would be out and potentially have fled the area. What could be done?
Enter Finland’s own scent ID K9-unit headed by Ilkka Hormila, who is also an instructor of cadaver, accelerant and scent ID dogs at Finland’s Police Dog Training Centre. Dogs from the scent laboratory in Hameenlinna, Finland could be used to compare the scent on the weapon to that of the suspects in a scent line-up. If all went well, in one to four hours the police would have an answer that would allow them to identify the true bludgeoner and keep him in police jail.
How the scent line-up works
As I entered the inconspicuous building that served as the Finnish scent ID lab during my trip to Finland in November of 2011, the first thing I noticed in the kennel area was that the four dogs—German Shepherds and Malinois— were all calm, relaxed, and friendly. I was visiting as a guest and the scent ID unit had prepared a demonstration of their testing process and the training steps their dogs go through currently, compared to methods they used when they first started.
Except for the fact that we had to remove our shoes and change into plastic sandals (Crocs®) to avoid tracking in many outside scents, the inside looked like any other office, except for the testing room. The testing room was lined by two long low platforms that extended down the room—one against the wall on the right and one against the wall on the left. Each platform had seven round metal holders equally spaced for holding glass jars and each holder was labeled with its own number from 1–7. Narrow metal cylinders with human scent would be placed in the glass jars—one from the suspects and six from the reference people or foils; adults not associated with the suspect. The foils generally included people from the jail where the suspects were being held and those from the trail where the suspects have been (some of the reference persons are from the same background scent area as suspects and some were from different background scent areas, like a fire department, school etc). All seven people have held eight of the 10 cm stainless steel tubes each for two minutes. The scented cylinders are then stored for testing and brought to the lab. Each person’s tubes will be stored in their own glass jar.
An officer in a room outside the test area rolls a set of dice that determines the order for the line-up. The chosen line-up order tells him where to place the suspect’s scent and where to place a control person’s scent. Using metal tongs, the officer carefully places each metal cylinder (individual odor from 7 different people), into the glass jars on the platform on the left side of the room. They are placed in jars to help keep the scent within a discrete area so that it won’t waft over and intermingle with the scent from the nearby samples. The officer then steps out.
A minute later, a handler and her German Shepherd enter. With her back to the line-up the handler asks her dog to lie down calmly on a small carpet and then has her sniff a control person’s object, which carries the scent of one of the designated foils (not the suspect). She rewards the calm sniffing with a click from a clicker to signal that the behavior is correct followed by treats—the dog’s kibble. She repeats this several times to be sure that her dog has had a good chance to detect the scent. Then with her back still to the line-up she releases the dog to sniff the platform on the right. She’s purposely facing away from the dog so she does not influence the dog’s choice. The dog sniffs the jars one by one and then lies down at jar 6. A green light on the wall in front of the handler tells her that the dog has chosen the correct control scent. The handler clicks, walks over and rewards the dog multiple times at jar 6 while the dog is lying down. She’s done a good job.
Then the handler has her dog lie down on the carpet again and after a short period of relaxation on the carpet, they repeat the procedure with the row on the left side of the room. If the dog identifies the odor of the control person again and does not show a particular interest in the tube of the suspect, then the dog can proceed to the suspect identification stage. This control step is important because it’s possible for a dog to choose a scent for reasons other than identification. There could just be something about the scent that he’s attracted to. In fact, three times in the eight years that Finnish police have been doing this, the dog has chosen the suspect scent over the control scent he just smelled. If the control stage had not been performed first, the results could have led to an incorrect conviction.
To start the suspect identification, the handler and dog leave the room and the assistant resets the tubes as needed. Then handler and dog return and this time the dog is rewarded for sniffing the assault weapon. The procedure is repeated and if the dog responds to the odor of the suspect in both rows, the police conclude that the scent evidence object and the suspect share odor similarity. The demo dog again gets the test right and in this case matches the item to the correct person both times.
In the case of the 4 criminals the scent ID led to one person being identified as the bludgeoner (his scent was on the weapon) and that person could then be held in jail, allowing for an even stronger case to be built. The entire process has been recorded on video for review.
This demonstration during my visit was only a reenactment, but these dogs in the unit practice every day. Since 2003 they have run 1200 searches for a total of about 3800 line-ups. Dogs have identified positive suspects 97 times—meaning they’ve first identified the control person twice and then identified the suspect scent twice.
Human scent line-ups are a relatively new idea for law enforcement agencies in the U.S.; however, human scent ID line-ups have been used in Europe for 100 years and are accepted in courts as circumstantial evidence. More recently, programs have been discontinued due to inability to hold up in court. However, dogs in Finland’s lab have an over 95% accuracy rate as tested in blind tests, where the handler doesn’t know the answer, and double-blind tests where neither the trainer nor the immediate tester knows the results (but the chief who plans the test and labels the scents does). Because of the strong track record, the results for this scent ID unit hold up in court. Says Hormila, “Defense lawyers never question the dogs these days. They trust our results. Instead, lawyers look for other sources of error that might get their clients free—such as the evidence collection methods.”
How the Methods Have Evolved to Positive Reinforcement-Based Techniques
Finland first started using their own dogs for scent ID in 2003, and it’s been successful in part due to help from other countries. It all started in 1999 due to a collaboration with the Danish National Police in Denmark and the Netherlands (Dutch) National Police Agency.
“We sent about 10 cases to them—five to each unit— and got good results and because of it our police Board of Directors decided we should embark on a 5-year project to start our own lab.”
Hormila selected three other trainers and they trained with and got certification with the Netherlands (Dutch) National Police Agency while also receiving a lot of help from the Danish National Police. “The Dutch and Danish helped us by analyzing our video and providing feedback. We went to train at their training centers and their instructor visited at our place. They allowed us to use as much of their system as we wanted in order to develop our own system,” says Hormila.
Each country that uses scent ID dogs runs their program a little differently, and Ilkka’s unit has made many adjustments to their system that helped them achieve such good results. First off, the training is very different from the traditional heavy-handed training that is prevalent in European patrol dogs. In Finland, all patrol dogs must also be trained in a specialized scent task—to detect narcotics, arson accelerants, or cadavers, and the training style that is used for patrol dogs tends to bleed over into the scent training.
Hormila’s group, however, does not use the strong force techniques. “We do not use choke chains, pinch collars, or electronic collars,” says Hormila. “Very rarely we may use a harsh voice and body language, but we do not want to create fear in the dog.”
It’s a far cry from the methods Hormila used when he first started as a patrol officer and scent detection dog handler 40 years ago.
“When I started training in the 1970s, many people were using a lot of punishment. And it got worse when the sport dog training philosophy came from Europe.” But he found soon that the punishment didn’t work long-term.
“We started dogs and had problems so our instructor showed us how to use electronic collars and I found it was disaster.”
“For example, the dog would find criminals or lost persons and starts to bark but then start to bite because they had been bite trained. To fix this, the trainers would put the dogs on long leash with pinch collar and give corrections which might cause the dog to bark more aggressively and the biting tended to stop, but only temporarily. When we took the leash away the dogs attacked much more powerfully.”
Hormila continues, “Then we went to having a helper hit the dog with heavy stick; more speed and more biting. Then we switched to the magic machine; the electronic collar. We found now we had almost destroyed the dog. The dogs were so mad and aggressive we could only use them with a muzzle.”
Ilkka quickly found he needed to look for another way. “A small group of us started to change the training method. It took a long time. We made exercises easier instead and used something positive, like toys.”
In those early years Ilkka found that punishment was also a problem in the scent trained dogs.
“When they are punished the dog start to worry about making more mistakes and is less confident to search. So they look to the handlers to see what to do.”
Some tracking or trailing dogs who were punished when the handlers thought they weren’t trying hard enough learned to keep their head down, as if scenting, when they had veered off the track that had been laid out for them. These dogs had lost the scent but were afraid to be punished so they keep their head down as if they could smell.
To prevent these side effects of punishment, Hormila trains his students to avoid harsh punishment. Instead of using force when the dogs have messed up, the worst that happens is that his trainers remove the expected reward. Of course, for this to work they are sure to repeat each of the training steps enough times so that the dogs quickly learn that correct behaviors work well and incorrect behaviors get nothing and may even end the session.
With this approach they have also been able to train the scent dogs to tell them when their search comes up negative. For instance, in a line-up where there’s no match to the scent on the item from the crime scene or when dogs are sent into a room to search for narcotics that aren’t present, dogs can get stressed because they have no way to communicate this information. Hormila trains the scent ID dogs to signal a negative find by coming back to the handler and lying down. This has been so successful they will now be training all other scent detection dogs—used for narcotics, explosives and accelerants—to give a “no-find” cue when the search is negative.
So how do you keep the dog from being lazy and coming back without checking carefully for the item it’s looking for? In fact, during the training phase, dogs do test to see if they can cheat by just coming back to lie down. “We just ignore them and they learn that this false behavior will not earn rewards.”
Another big difference with Hormila’s training now compared to when he started is that in the past they wanted “high–drive” dogs for all police work. They encouraged high energy and excitement over the reward they were using, which was a toy. Says Ilkka, “We used positive reinforcement, but the way we were training we never got the dogs to relax well.”
Hormila quickly learned that this high-energy behavior was undesirable. “My first patrol dog would bark in the car any time he saw people and would bite the windows. One night, after whole night with the dog barking we had a search case near the end of the shift at 6:00 am. After the 8 hours of barking he was supposed to search for a missing person in the forest and he was so tired and he couldn’t.”
This type of situation occurred several times. Since then Hormila has trained all of his dogs to remain calm in the car and that the car is a place to relax. Many trainers then and even now like for their dogs to guard the car, but Hormila has a different philosophy: that because the dog is in a cage in the car, someone could steal the car even with the dog barking. So it’s better to use keys and lock the door.
While Hormila had been successful in training many dogs to remain calm in the patrol car, he did not have the same success in the scenting situations. The result was that dogs often got more excited the more they were trained, especially since he was encouraging excitement over the reward, a toy. The problem here, says Hormila, is that “With high drive their nose works but their brain does not. Dogs start to make more mistakes and have incorrect alerts in training and also in crime cases.”
Then in 2008, after first having read some books on clicker training and attending a seminar in Finland by Bob Bailey, they sought help from Finnish trainer Tommy Wiren, who studied under Bob Bailey. Wiren lectures across Finland and helped the Finnish Assistance Dog Trainers develop more efficient programs in 2000. Hormila attended Wiren’s operant conditioning camp where chickens were used as a training model. After that, two additional trainers attended and then three more. Hormila also arranged a training weekend with Wiren for 15 police dog handlers around the country. “What we got most from this was the importance of timing the rewards exactly right and also the use of food in training, the direction of reward,” says Hormila.
They started rewarding dogs at a high rate for lying down calmly in the training situation until the dogs were consistently calm and then decreased the amount of rewards needed. They also spent time in the room just relaxing with the dog and not training.
The improvement was dramatic. Says Wiren, who was hired to also consult on a regular basis with the department, “When I first started working with the scent ID unit, the dogs were pacing anxiously during their searches instead of focusing carefully on the task. Now, as I had witnessed in the demonstration, the dogs work in a calm manner. There’s no whining and pacing. If dogs start to get too excited, they are given a break to calm them down and are taken out again about 10 minutes later.” Says Sergeant Harri Sojakka, a handler who had been working in the ID unit from the beginning of 2003.
Training with Wiren also focused on carefully understanding canine and handler body language and taking data to determine whether the training should be going faster. Since Hormila’s group had already been videotaping their training sessions they were able to make important discoveries. For instance, they were having a problem with one of the arson dog handler teams. The handler was calm and didn’t appear to be giving cues to the dog, but he moved his eyes and the dog picked up on that cue when making decisions.
“Since then we start as soon as possible to perform exercise blind where the handler doesn’t know where the scent is,” says Hormila. Because of these findings they also make sure they are facing away from the dog in the scent ID line-ups so they can’t see what the dog is doing and accidentally cue him.
The New Techniques Have Increased Efficiency
Overall, the reward-based training, improvements in timing, data collection and rewarding of calm behavior improved the training of dogs in all of the scent areas immensely. Say Hormila, “We can cut our training of narcotic dogs down by 1/2.”
It used to take several weeks to train the first 5 scents. Now they can train the first 3 in just five days. After the handler and dog know how to work with this method, to teach a new pure scent will take just 2 days, equaling about 120 correct positive reinforcements. After that they still have to train the dog to search in distracting situations, but this is relatively easy once they’ve done the initial indoor training. As a result, it used to take from 4 to 6 months and now it takes only 2 months of training full time. The human scent ID training is going faster too. Training time took from 20 to 24 months at first and now it takes about 12 months.
Because of the success of the reward-based training and of the scent ID unit, the Police Dog Training Centre is now making it policy that all the scent dogs—narcotics, accelerants, explosives, cadavers, currency and human scent ID – be trained using the same positive reinforcement techniques that Hormila and his scent ID unit have been using. It’s a big change from what other trainers in the school may have been doing, but they hope it cuts down on training time and provides the more consistent, long-lasting performance that the scent ID lab is getting.