now on bbc news, our world. operator: 911, - state your emergency. man: itjust sounded like there were fires of gunshots. - how many shots? 15 or 16. on average in america, 50 people are killed every day with guns. do you feel safe here? no. in california, we have the worst gun violence. it's just terrible, man! one company, shotspotter, says its gunfire locator technology can help tackle this huge problem. shotspotter is the reason why we were able to apprehend him and stop that killing spree. but some have questioned how often the company gets it right. so, is that like a typical type of call out from shotspotter? yeah, we look, don't find anything, can't find any evidence. and when things go wrong, what are the consequences? to be injail and not know what you're there for, emotionally, you're a wreck. does this technology work as well as the company claims? and what role can it play in the american fight against gun crime? you can tell a lot from a waveform. how loud, how high, what frequency. train a computer algorithm and it can, in theory, work out what it might be listening to. the's the idea behind shotspotter — that a computer can detect a gunshot and then tell the police the location it was fired from. gun fire. small mics are placed around a neighbourhood. when a gunshot is fired the mics pick up the sound, how long it took to reach the sensors and from which direction, effectively triangulating the shot. shotspotter was founded on the west coast of america in the mid—90s, in that cradle of american innovation, silicon valley. the company has agreed to give the bbc access to show me how the system works and why they believe it's needed. what makes the system so compelling, we believe, is the fact that in many communities that suffer from ongoing persistent gunfire, people for a lot of reasons don't call 911 which means there's no police response which means that communities begin to distrust police. you can see we've got some proximal incidents right here. this is shotspotter�*s incident review centre. ginger is one of the analysts. wavy alarm sounds. oh, there's an incident right there. that's the sound we hear when an incident comes up, yep. when the computer thinks a gunshot has been fired, herjob is to review it and decide whether to dispatch police officers, all in under 60 seconds. this is a waveform which if you do any kind of audio work, you've seen this before. so we're listening to that, we're looking how far does the sound travel, we're looking to seeli how many sensors picked it up, if the sensor has made a directional pattern because, in theory, a gunshot can only travel in one direction. to me it doesn't look like it's gunfire, but that's why we also listen to it. wavy alarm sounds. possible gunfire pops. that sounds like gunfire to me. in this case, i think you might be right — we've got a directional sensor pattern. i'm going to call it gunfire. i'm going to submit it and i'm going to send it over to our agency, and you can look up there, itjust popped up right now. so now, it was that quickly, there's going to be police going to that location right now? correct. the people in this tiny room are listening to cities across america, there are over 100 of them, and their decisions what is a gunshot and what isn't a gunshot is absolutely crucial. it could be the difference between life and death, between finding a criminal and not. i'm travelling to fresno because i've been told of a shooting incident which proves shotspotter�*s effectiveness. in 2017, kori ali muhammad was on the run, wanted for murder. he had a hatred of white people and on the 18th of april he decided he wanted to shoot as many white men as he could. operator: 911, - state your emergency. man: itjust sounded like there were fires of gunshots. - how many shots? 15 or 16. fresno police officer bill dooley describes what happens next. he comes up to the passenger side where the first victim is and lifts up his weapon and fires four shots and kills our first victim. operator: 911, - state your emergency. man cavme: my i partner's been shot! this is when our very first shotspotter activation comes out. although there were plenty of 911 calls, they were delayed and unspecific. shotspotter, however, was alerting police to his route, one shot at a time. he starts heading west on mildreda, he sees a gentleman coming out of his house on the south side of the street so he fires two shots, misses the individual who came out of his house... gun shots. he continues to travel down towards fulton and that's where he sees his next victim. so he's fired his first two, he's running up to the victim who now is down on the ground. gun shots. he fires his next two shots. muhammad then runs towards a group at a bus stop and chases an older white man towards a car park as he reloads. and that's where we have our last shotspotter activation. gun shots. operator: 911, do - you have an emergency? woman: yes, there's a guy out here that's bleeding, _ he got shot. ok so the suspect, do you know where the suspect is? he's runnin' around out here. three people were killed in the space of 90 seconds that day. police identified where muhammad was going and arrested him as he headed towards busy downtown fresno. shotspotter gave us the clear path that he was taking. because of all the 911 calls that are coming in, information is good but it may not be up—to—date. the up—to—date information was shotspotter. so for you shotspotter in this scenario was essential. oh, absolutely. there's no doubt in any of our minds that additional lives would have been at risk or lost if we had not apprehended him as quickly as we did. shotspotter is the reason why we were able to apprehend him and stop that killing spree. there are clearly examples of shotspotter system working, even saving lives. and for years it's been used by police forces pretty uncontroversially. that was �*till lawyers around the country started to ask questions, difficult questions about how the tech actually works. they started wondering whether it's as accurate as the company claims, and if it isn't, what the consequences are for americanjustice, and those questions have been loudest here in the city of chicago. so talk me through, where are these actual shotspotter sensors here? you see the round ball thing, the black thing at the very top, that's the sensor up there. how do you feel when you see them? it makes me angry. michael williams was released from jail two months ago, having been accused of murder. on the 31st of may last year, michael decided to take a drive to buy some cigarettes. it was a night of protests over george floyd's murder. there was crowds of people, they was all over the place. that's what made me turn around. someone he recognised from his neighbourhood asked him for a ride back home. he agreed. i stopped at this light. he wasn't even in the car two minutes before another car pulled up and opened fire, struck him in the head, and i was hollering out to my passenger, you "are you all right right, you ok, are you hit?" and he wasn't saying anything and when i glanced up at him from the floor i saw the blood coming down and ijust put the pedal all the way to the floor. michael took the man to hospital where he later died from his wounds. two months later, the police knocked on his door. he was arrested for first degree murder, accused of shooting his passenger in his car. brendan max, michael's lawyer, talks through the case against him. the police had no witness who said that they saw michael shoot anyone. they had no weapon, they had very little evidence in this case other than a shotspotter alert which directly led to him been charged and incarcerated in this case. we had two reports from shotspotter. one was signed by one of their expert witnesses. the shotspotter alert placed the gun shot around the location of michael's car. it was used as evidence in a case against him. to be in a four corner concrete room for 11 months... ..with 39 other people, and you know you haven't done anything, it does a great deal to you, mentally. ijust couldn't function like a normal human being. it got so bad for me in there, i was on several medications. ijust saved them up, i was hoping that it would have put me into a deep enough sleep to stop me from breathing. while michael considered taking his own life, fortunately, he didn't go through with it. after 11 months injail, and despite shotspotter giving the police a detailed forensic report, or dfr, attesting to the location of the gunshots, the prosecution suddenly withdrew their evidence. the case was dropped. it is well—known on a dfr and also our standard contracts that we don't warrant or stipulate to detecting what's called suppressed gunfire, and suppressed gunfire are gunfire that happens indoors or a gunfire incident that happens in a car. it became very clear at some point in time that with that explanation or that description of what we can warrant to and what we would testify to, that was in conflict with the prosecutor's theory of the case. spotspotter evidence withdrawn, charges dropped. brendan max, however, doesn't buy that response. we showed him shotspotter�*s explanation. ..suppressed gunfire are gunfire that happens indoors or... what mr clarkjust said is two things. there was a gunshot here, so shotspotter worked correctly. mr herring was shot, so our report was right. and on the other hand, he's saying don't trust our report because the shot may have come from inside a vehicle under one potential theory of this case. those things are contradictory. brendan max found something else — that the algorithm had initially classified the gunshots in michael williams' case as a firecracker, but a human had reclassified it as a gunshot. however, that detail was not in the forensic report shotspotter gave to the police. and in fact, when i visited shotspotter, the algorithm disagreeing with the human analyst was fairly common. this one is probable gunfire. popping sound. i think it's a firecracker. we are not under no obligation to go with what it tells us. it will suggest, but we're looking at so many other things in such a short period of time. honestly, i don't even look — i am so busy looking at the sensor patterns. like, all of the sensors got skipped. i don't think it is anything. and you can hear there's the little pulse and then there's the big pulse right here. popping sound. i think it is a firecracker so i'm gonna go ahead and dismiss it as such. during the process of defending michael williams, brendan max began to suspect something. shotspotter claims their system is 97% accurate. brendan max began to wonder was that really true? police radio chatter. i want to see how shotspotter is used on the ground and fresno police have invited me along on a night patrol. i'm out with officer palomino. popping. so that was the activation. oh, that was it right there? yeah. radio: shots fired. came in at 2252, about four minutes ago, so our helicopter is above the address that the spotspotter pinged at. we went to the location where shotspotter had detected a gunshot. so we think there were shots fired on the street somewhere? yeah, that's where the shotspotter activation was pinpointing it. this is where it was — this is where the shotspotter was indicating. so it's literally underneath this tree where it thought the shots that happened? yeah, that's what they said. 0k. so you have had a look for casings. mmm—hmm. have not found any. right. what — do you think there was a gunshot here? i mean, you can't really say, but if there is no evidence of a gunshot or... so was that like a typical type of call—out from shotspotter? yeah, that's pretty normal. yeah. yeah, interesting. we look, don't find anything, don't get any — can't find any evidence. no house is struck, no vehicle struck, no other callers. the police tell me this kind of scenario is common. when a shotspotter alert is created, it's often hard to know when a gunshot has actually been fired on the ground. that makes trying to work out how effective shotspotter is tricky. we wanted to really get at the question of the operational value of shotspotter, so we looked at about 50,000 shotspotter alerts over a period of 1.5 years and we found that in only about 9% of those shotspotter alert responses does the police department record having developed some evidence of a gun—related crime. did that surprise you? well, a 9% success rate in developing evidence of a gun crime does not look to me like an operational value that necessary outweighs the cost and the risks of the use of shotspotter. the inspector general of chicago said that they believe only 9% of dispatches there is any evidence of an actual gun crime or even a gunshot. how do you respond to that? i've read this report a few times and what they are stating is that there is a 9% reported evidence of something being found. right. it does not mean that there wasn't evidence found. but that leads to the next kind of obvious follow—up question — if it's so difficult to work out whether a shot has actually been fired on the ground, how does shotspotter know with such confidence that it's 97% accurate? on our ride along, i saw for myself. gun shots. shotspotter says this figure is accurate because the police officers themselves are able to listen to the noise recording and feed this back to them — something they call ground truth. and that's part of the feedback that you are basing that figure on? absolutely — yeah, that is the feedback. we rely on ground truth from agencies to tell us when we miss, when we miss detections or when we misclassify. other people have said about the general stat of 97% stat that it is not rooted in scientific rigour — i.e., it has not been scrutinised, it hasn't been peer reviewed. do you think that is fair? it is so certainly a factor that there has not been an academic peer review of the service, but i would push back in saying that it is not been analysed — it has been analysed for 20 years across 100—plus customers that are using it every single day. here's the problem with that — often, it's pretty hard to tell whether a gunshot has been fired by simply listening to the audio. so it's actually pretty difficult to know whether there was a gunshot or not. hmm—mm, yeah. i mean, it definitely could have been and it'sjust we don't have anything to prove that it was, so we just clear up the call and that's pretty much it. right, just from listening to it, did you think it was? yeah, it sounded like a gunshot. yeah? sounded like it, but who knows for sure. right, and also, you are not an acoustic expert. yeah, exactly. ..analysed for 20 years across 100—plus customers that are using it every single day. what mr clark had to talk i about there was marketing, it was not science, and to hear the ceo of this corporation say that that is the basis - for the claims they've made for many years, for— the evidence that has been used not only against michael but against hundreds . of defendants around - the country, it's outrageous. crowd chants: i can't breathe! following the death of george floyd in may last year, the use of shotspotter became even more controversial as trust in the police was at an all—time low. we need to steer them away from technology and to the people. in chicago, activists argue the tech is disproportionately deployed in black and brown communities, so could be dispatching police to neighbourhoods where they think a gunshot has been fired, guns drawn, when, in fact, there has not been. like, we might not get the answers that we want. alyxandra is overseeing today's meeting and wants chicago's shotspotter contract to be cancelled. it's sending police in to, like, these situations that they expect to be hostile, they expect there to be a gun. because, again, of like, where this is deployed, they are expecting a black or latinx person to be holding the gun. people are being stopped and frisked, harassed by police, stopped by police because they happen to be in an area where shotspotter is actually deployed and where the alerts are actually going off. why are we paying millions of dollars for something that sometimes works? people almost want to say well, if the real problem would be officers showing up trigger—happy, that is something we immediately do a dfr on. we have to do a dfr on that, because those are really big things. we don't see that. we we — don't see that. it's not in the data. it'sjust not in the data. police across the country do believe shotspotter is effective. fresno is looking to expand shotspotter into other areas. this expansion is going to be close to a million dollars a year the city of fresno is paying for this service. it's got to be worth it. it's gotta be worth the money. and if we ever find out it's not worth the money or it's inaccurate or it's actually hurting our efforts instead of helping our efforts, then i'm going to look into getting rid of it. but right now, it is proving to be very useful. on our ride along, we came across a group in an area that did not have shotspotter. do you know where the gunshots were coming from? it sounded like by the orchards. theyjust heard shots being fired. do you feel safe here? no! in california, we have the worst gun violence. i hear every single day there is another article about someone getting shot. it's just terrible, man! let me tell you something, bro. most of the problems out here in fresno are due to the manpower that the fresno police department has. you know, they can't have one crew here and two crews over there. so you want more police? well, yeah! yeah, i honestly do! clearly, there are people that feel unsafe in their own neighbourhoods for whom shotspotter could be a welcome piece of technology. and yet, if it is not so accurate, this could have huge knock—on effects for the criminaljustice system. in the last four or five months, i'm aware of dozens of chicagoans who have been arrested based on shotspotter evidence. i'm sure that has played out in cities across the country. shotspotter needs to be dismantled because it's doing more damage to us than anything. i mean, many of us is gonna end up in the countyjail? but shotspotter can get police officers to gunshots more quickly. the technology does save lives. what if shotspotter only saves one life in a given year? is it worth $1 million? in our view, it is, because you cannot put a price tag on the human life. ultimately, though, until shotspotter�*s reliability has been independently established, there will continue to be questions about whether this surveillance technology should be used so widely. hello there. it's staying pretty cold now for the rest of this week. there will be quite a bit of sunshine around by day, but it will be chilly at night with frosts and fog in places. now, through the day there will be quite a bit of sunshine in central and southern areas, but also some showers. this mainly affecting coastal areas. got low pressure to the north of the uk, this is where we are seeing the strongest of the winds today and weather fronts enhancing the shower activity. although most of them will be affecting coastal areas all around the country down into west wales and the southwest, northern parts of northern ireland and much of the eastern side of the country. some of the showers across the northeast, east of england will push into the midlands. so a bit cloudy here, more than we have had the last few days. some of the showers heavy with some hail and thunder mixed in, and they will be wintry over the high ground, particularly across northern scotland with the strongest of the winds. further south the winds will be lighter. it's going to be a chilly day wherever you are, highs of six to eight in the north, nine to 11 further south. through wednesday night it stays quite blustery in the north, further coastal showers and many central areas will turn dry with lengthy clear skies again. winds will be lighter here. but stronger winds across the eastern side of england, more cloud. not quite as cold here as what it will be further west, and across the north we will have some frost and also some patches of fog. now as we move towards thursday into friday we start to see a change to the weather. this area of high pressure begins to build in from the west. it kills off lots of the showers, but what it is also going to do is cut off the arctic air supply as our wind begins to veer more west and southwesterly direction. so that will bring milder air back to our shores as you can see here from the orange and yellow colours. thursday, then, another cold start with frost and fog around. and then it is bright. plenty of sunshine around, more sunshine around on thursday. still a few showers across the eastern coasts again down into the far southwest. later in the day, thicker cloudy, patchy rain pushing into the northwest of the uk as we start to pick up westerly winds. temperature slowly rising here, otherwise for most again it is a chilly day. as we move out from thursday into friday we start to see an area of high pressure toppling down towards the south of the country, and that will allow this area of low pressure to move in across the north at the start of the weekend. so it will be turning milder towards the end of the week, particularly as we head on into the weekend. low pressure will start to bring wetter and windier weather to northern areas, and will tend to stay drier and brighter the further south and east you are. this is bbc news — welcome if you're watching here in the uk or around the globe. i'm david eades. our top stories: progress at the climate change summit in glasgow but the cop26 host warns of false hope in the race to limit global warming. the eyes of the populations of the world are on in the eyes of the world are on in the eyes of the british government and all the british government and all the other governments care about this, are on our negotiation and we have your numbers. as the race to be the next governor of virginia hots up, we report on what the outcome could tell us about the us, a year after president biden�*s victory. relief in australia, as a four—year—old girl, who was taken from a campsite more than two weeks ago,