Loading summary
Transcript47 lines
- [00:12]
Katherine Levin
I thought we'd do something different with the POD and dig down into a single topic and technology seemed a good place to start. Tech Perspectives then is a short POD series brought to you in partnership with the Emergency Tech show and the Police Digital Service. Although we won't just be talking about the police. I'm Katherine Levin, editor of Emergency Services Times, and in this episode I talk to Al Murray, the new Director of Police AI. I don't know about you, but I see AI everywhere. It's in the news, it's in my work, it's in my home life, and of course, it's in the emergency services. AI is finding its way into policing in a big way. So it makes sense to start with Al Murray. The timing is good for his new role. He has been the National Police Chief's Council lead for AI for quite some time. He's a police lifer, as he described his career to me. He has moved on to Convent from the National Crime Agency to take on this AI role which sits in the College of Policing. Prior to that, he was Chief of West Mercia Police and spent a lot of his career in the Met. He has a curiosity about doing things differently. He talks about evidence based practice being important to how he works and thinks. We start this conversation by asking him where his interest in AI comes from.
- [01:44]
Al Murray
All the way through my career, I've been interested in how the service needs to be at the frontier of development. I'm profoundly interested in evidence based practice, which is a real driver for innovation. So how do we experiment and do things differently to better serve communities? And so much of that is now in the technology space and in the nca, I was director of Threat Leadership and across things like child abuse, organized immigration, crime, cyber crime, technology runs through it as a hugely important factor. In fact, the National Strategic Assessment this year, when we look at crime and serious organized crime, makes quite a good point where it says technology has ceased to become an enabler for crime, it is now a driver for crime. So in the old days we used to say, yeah, people want to commit crime and they use technology to do it. And now we're saying actually technology is so profound in everyone's life, it actually drives crime in some areas. So I've always been interested in technology and yeah, I'm not a natural born technologist, but. And I have been a police officer all my career, but luckily with the team I've got, I'm surrounded by technologists. And that nuance between both policing technologists, criminologists, it all sort of works quite well, I think.
- [03:04]
Katherine Levin
Excellent. And I think there's something about a curiosity, isn't there, about wanting to kind of get underneath the skin of all this? And you said in a recent post of LinkedIn where you talked about your new role giving you a sort of sense of behind the curtain. And you told us, you know, you talked about it being a steep learning curve and the challenge of building the train while it's on its journey. What did you mean by that?
- [03:25]
Al Murray
Yeah, this is not like other AI functions around the world, I think, in that we are squarely in the face of delivery. Our job is to deliver responsible AI into the hands of frontline staff. So quite a lot of AI office around the world are quite thought orientated. You know, they'll commission a lot of research. We're not in that space. We're even probably not in the proof of concept space where a lot of people operate. You know, does this AI work in this area? But we're in the space of understand what works, test and evaluate it, scale it and engineer it into police forces around the country. So we've got an expectation, rightly from government to save, for example, 500,000 hours worth of police time by effectively using responsible AI. So I need to get out there and make it happen and I can't afford to recruit a whole team that's going to do that and then say, okay, how are we going to deliver AI? I need to do it at the same time. So that's the metaphor of sort of like building the train whilst it's on its journey. I will be recruiting and contracting in lots of people to do work whilst we're building our team, because the need for policing to embrace responsible AI is right now and can't wait.
- [04:45]
Katherine Levin
And you've got, I think, just over £100 million for this. Is that the annual amount or is that, is that for staff? What will that achieve? It's not very much when you're thinking about all the things you've just set
- [04:57]
Al Murray
out there with 111 million in the white paper, not all of that goes to police AI, so quite a lot of that goes to some existing work streams like live facial recognition. So we've got £75 million over three years, and the majority of that absolutely will not go to building a team in police AI. The majority of that will go into building and scaling capability and rolling out into policing. I will have a very small team. And we will be buying in solution architects, we'll be buying in business analysts, we'll be buying in buying in people who are capable at building and engineering these tools and then giving that capability to forces. And it's good. Having been an ex chief, when you see money not going to a police force, it's concerning, but the reassuring thing is that the money is coming to us to package up a capability, make sure that it's responsible, make sure that it does what it says on the tin, and then help a force plummet into their systems and their workflows so that it works effectively.
- [06:04]
Katherine Levin
And within policing, you already have some systems which help you with that. In terms of thinking about Paul Taylor and the Scientific Advisors unit and the STAR program where they give out money for innovation and also the accelerated capability environment sitting in the home office, how will you work with those units which have money and similar sort of approaches to make sure you're not treading on each other's toes and the actual outcomes are approved?
- [06:27]
Al Murray
Yeah, so we use ACE all the time. The Accelerator capability environment. They're a great agency. When you've got a need, they reach into the market and say, you know, I need five solution architects. Okay, we can get you those, or I've got a problem and they'll dive into the industry and help you with that. We work very closely with Paul Taylor. He's got the covenant for AI, which all chief constables have signed up to, but, you know, no capability to work on AI. And we've recruited temporarily one of a significant leader in Dylan Aldrich from OPSCA into police AI. So there's only complementarity there. I don't think. There's no sort of duplication. I don't think so.
- [07:07]
Katherine Levin
That's all about structure. So let's go back to the topic of responsible AI. You've used that phrase a few times. Do you have a definition for that?
- [07:15]
Al Murray
Yeah. So, I mean, this is hugely important and it comes back to legitimacy and the right the public give to the police to do what they do. And, you know, it's a real cliche, but we survive on public consent. And again, like, the criminological literature is really good on this. If you think I'm legitimate as a police officer, you're much more likely to comply with what I say, to cooperate, to provide witness statements, or if I do something, for you, to trust my motives. So how do we build AI in a way that embraces that trust from the community? Because I think there's sort of three ways you can get AI into the hands of policing quick. One is let's just engineer in artificial intelligence and you can get that in really quick. But that's not responsible. That won't have been tested and evaluated. You won't understand the bias, it won't be transparent or explainable. So in actual fact, we want to deliver a responsible AI. That means you've got to do it a bit slower, but it means that the public can be reassured that we're not just throwing AI out willy nilly that we've understood the bias, we've tried to mitigate it, we've understood the data it's trained on, we've understand the outputs, we understand the motivations, the models, and we've tested it and then put it out into policing. And that way, you know, people can trust our approach. And then, then there's a sort of even. There's a third way which has to be even slower, which is when it touches the criminal justice system, because everybody needs to trust the criminal justice system. And you can't have AI in a court of law or being used in artifacts that go to a court of law without it being absolutely transparent. As you can imagine. You know, if you're going to use AI to write a statement, you need to say you've used AI to write that statement and it needs to be able to be cross examined and the use of that AI. So that's even a third stage where we work very closely with the Crown Prosecution Service on ensuring any AI within the criminal justice system has every I dotted and T crossed.
- [09:19]
Katherine Levin
So when you're using that approach to perhaps improve productivity in policing, so use of redaction or whatever it is, will people be able to see on the outside of policing what you've done to improve that productivity with that product so that they are reassured that it makes sense, that it does get their confidence, it is legitimate. And all the words that you use in policing, can you do that when it comes to productivity, what does that look like?
- [09:46]
Al Murray
Yeah, well, a couple of things to say on that. Firstly, we have to rigorously baseline where we are without it, look at what the consequence of using it is and understand the difference so we can demonstrate value for money, for the taxpayer and there has to be real integrity in that. And secondly, we want a registry, a public facing registry of pertinent AI that people can go on and see, okay, you're using AI in this context and we can be assured that it's tested and evaluated. So we have had to say to a couple of forces, you can't use this form of AI to write these forms of statement because it hasn't been through the due diligence. But similarly, we can use AI in redaction and we can talk about some of the use cases that we're considering building for AI in a bit. And you can, as long as, as long as you can say, this is how we got there, this is how we understand how it works and this is how we've tested and evaluated it and that's what we can broadcast on an open source registry.
- [10:45]
Katherine Levin
What's your timeline for registry?
- [10:47]
Al Murray
Well, I want to get it out at this year and there's, there's sort of different levels here, I think, and you use AI and your listeners will use AI all the time without even knowing it, whether it's in Google Maps or, you know, the Spotify playlist. There's some things that you just expect to be used or that are integrated in software as a service applications that are just there. I don't think we're talking about AI in this space. I think we're talking about AI that people want to understand, would be concerned about and would be concerned if we hadn't been diligent on its rollout, that it shouldn't be used. So we, for example, Catherine, want to use AI in image recognition. So, you know, a crime takes place very quickly. You've got terabytes worth of CCTV you need to analyze and AI can have a hugely beneficial role there. But the market is full of different types of AI in this space. Some are good and some are bad. So we would want to put that on a registry and put that it's been evaluated and this is the best in class and this is how the human in the loop operates or similarly. And another use case that we're working on is child abuse image classification. You know, we're seizing again, terabytes and terabytes worth of horrific imagery. And AI can help us understand within an ecosystem of 20 laptops that have been seized, where is the child abuse imagery and what type of child abuse imagery is it? So we can get there fast, remind somebody in custody if necessary, or if they're completely innocent, you know, let them go home. So that's the type of thing where again, people would want to be reassured that we've got the best thing in the market and then it's been tested and evaluated and then we've got some basic productivity tools that we're also working on that we can dive into, if
- [12:36]
Katherine Levin
you like, just on that point about officers being exposed to images in the way that you've just described, you've described in a systems way, and it's to identify them within laptops and speed it all up. But I wonder, just in terms of the well being of officers whose job it is to look at those images, whether it's child abuse or anything, that's just a worry about police officers who are exposed to this. Is there a sense of AI being able to reduce the exposure and trauma for staff and improve well being? And it's part of the baseline that you mentioned earlier. Would that include well being as well?
- [13:09]
Al Murray
Yeah. So 100% viewing this material is incredibly difficult and we need to get enough into a court of law to justify our prosecution of an individual. And if AI can assist with that and help in the classification of that, I don't think it's ever going to remove the necessity to view this imagery. But you don't have to see all the imagery. You can see some and you can be directed straight perhaps to the most harmful. See some, classify it and AI can identify the rest. And again, we'd have a human in the loop here. It would help with the triage, if we ever wanted to use it in a court of law, to say, for example, there's 20,000 images here, 20% of this category, and AI has identified that it would be absolutely right, if we ever got that far, to be able to demonstrate to the prosecution and to the defense this is how it's worked, this is how it's been tested, this is how it's been evaluated and the court can make a view on whether it's beyond reasonable doubt or not. But even prior to going to court, I think AI does already, and there's quite a lot of AI in this field already, but it can remove quite a lot of that exposure. And then to your point on wellbeing, yeah, I'd love to have a metric for baselining wellbeing before and after. We don't have it at the moment, but it's a really good point.
- [14:27]
Katherine Levin
I also have spoken to people in policing about this point around the control room. So if AI has been used in the control room to improve efficiency, improve productivity, and actually have people talking to people on call without having to spend all the time doing the basic admin. Actually what it does is it actually refines the exposure of those callers, control handlers, to the most difficult calls on 999. And so I just wondered whether the opposite effect would happen as well. Actually, you reduce so much AI that actually you expose people just to the really hard stuff and that makes it worse as well, because they're not Getting any relief? Is there any discussion going on about that in policing?
- [15:05]
Al Murray
Yeah, well, digital public contact, really good leaders in this field and AI is a really important decision making tool. And I don't think we'd ever want to replace the 999 call handler with an AI, but with an AI working in the background going, have you seen this type of offense has been highlighted. Here's the correct policy in this place, here's the correct criming pathway, here's the correct vulnerability. All of that is working through. It's incredibly effective. And similarly, in the west midst, there's an AI called Andy Ezra that's operating at the moment. Again, really effective, where it answers a 101 calls and can write reports for a caller. So if I call up with SMIDS and I say I want an update on the robbery or my car theft, it will write a message to the officer in the case and saying, can you respond to this person? You don't need a human to do that. But if it detects that you're Vulnerable in your 101 call, it pushes you straight to the front of the queue and you can speak to a human and at every possible time it says, look, do you want to speak to a human? So we don't want this time where, you know, you and I all been on long queues, you just say to the machine, I want to speak to a human because you don't understand my case. That's not what's happening in West Mids. In fact, it's making it much more effective. And this is the nuance with smart AI, I think it's not all or nothing, it's always human. And I think.
- [16:32]
Katherine Levin
And that's presumably the same as the system in Thames Valley with Bobby.
- [16:37]
Al Murray
Yes, Bobby is a Salesforce product and again, can really help people on their victim's journey.
- [16:44]
Katherine Levin
Yeah, we do need to think about the victim, don't we, in all of this? So it's all talked about response bailout, talked a lot about systems, talked about the staffing. But at the heart of this is the victim getting a better experience from policing that journey that you've talked about. Even if it's the exhibitor simply saying, this is what's happening, we are listening to you, things are happening. So in terms of the system sitting behind all of that, you Talked in your LinkedIn post about agencies that carrying technical debt. Now this comes up quite a lot in conversations I have with people in technology about spending so much time keeping the lights on actually they have time to innovate and space headspace, money, whatever. How will you in your work within police AI be able to help forces manage that and balance things out so they can do the innovation and perhaps use the AI to deal with the technical debt that they're carrying?
- [17:29]
Al Murray
Yeah, this is, this is something that does keep me awake at night more than anything else. It's one thing showing that the AI application is effective and it can be scaled. The engineering required then to get it into 43 different force it infrastructures is challenging. And then getting staff to use it within their workflows, within those forces to realize the benefit is a real challenge. And we need to look at a number of models there and we talk about AI at the edge, AI in the cloud, AI on prem, and which is the best model for which force and for where they are from a tech stack point of view. So that's a really important point here that probably a lot of people miss from an AI point of view, which is just normal solution architecture. And that is a real challenge for us.
- [18:22]
Katherine Levin
So it's a challenge. Where does it sit? We created quite a pile of things here in this conversation. You Talked in your LinkedIn post about where to focus first. So is the focus there first? Is it on the innovation? Is it on the really super duper exciting things over at the edges? Where do you decide where to put your effort?
- [18:38]
Al Murray
So that's a really good question. So it's not on much as we'd love to be hyper innovative, in fact I allow forces. Forces can be hyper innovative here and they can build products of responsible AI. The only thing I time I think I'd intervene was if they took responsible out of it and they can try things out. Where we come in is we think okay, that looks really good. How do we build it and evaluate it and test it and scale it. So I think our job is to take what's really good and that could be an off the shelf product from a company. We'd evaluate it, we'd assure that it's responsible and then we'd create a mechanism for getting it into police forces. So we are not at the experimental, not at the hyper innovative. We are at the scale, test, evaluate and engineer. If again, I think quite a lot of institutions around the world are at the hyper innovative. But like with any new business, the failure rate is so high that we could spend our whole time being innovative and not get any use cases into the hands of police officers and police staff. So we don't and we don't want to do that. So we'd encourage forces to be innovative and then we take what they do, evaluate it and roll it out.
- [19:51]
Katherine Levin
It's interesting. I think in public sector they are. People are afraid to fail and I think that's just sort of in the DNA. And it's interesting, you use a phrase in your post, you said there's jeopardy and we need to get used to that. So how do you encourage forces within leadership teams, within technology teams, or simply those at the front line who've got an idea to actually go, let's just try it out. Can we be safe in doing that? How are you going to create that kind of environment for people?
- [20:17]
Al Murray
That's a really interesting question. So there's sort of two sides to that and I think quite a lot of chief constables are incredibly progressive in this area and I see some of the products around the country being developed and I think that's really good. And that's because someone's been given permission to develop it. And actually, I don't see the jeopardy there so much because it's quite easy to show a proof of concept. Where I personally feel it is that if you go, if I'm in the NCA or if I'm in a police force, I slot into a job and there's history and precedent in that job, you know what you're going to do. Yes, critical incidents come along and can affect that force or that institution, but ultimately it will succeed. Very rarely does it fail. I think with police AI, we are talking about some significant challenges of getting significant AI capability into the hands of online staff. We are investing in use cases and there's a genuine chance that it might not be scalable or it might not be deliverable in that tech architecture. So I think police AI will succeed, but some of the big bets that we make will not come off. And that is different for policing. So it feels like we're a startup and we're very small and we are trying to deliver some big products and some of those big products won't work and we will have to. And the ecosystem and the, you know, my peers will have to go, yeah, okay, you tried your best, we understand, and we've spent some money on it, but it didn't work. But the other thing that you're delivering is, and I don't think policing is as used to that as it will need to be, I think that is the nature and AI changes every day is not a challenge for us. So whilst we are working on delivering something, we've seen it haven't we? Every day the rate of change gets faster. Mustafa Suleiman, who's one of the co founders of DeepMind, he said the rate of change will never be as slow as it is today. So we already think it's gone out of control and really fast. Well, tomorrow it just gets even faster. So the other thing is you have this sort of paralysis analysis or analysis paralysis where you go, well it's changing so let's wait. And then you never deliver anything. So you know, this is that this is the space we're in and it's okay. But people need to understand that, I
- [22:40]
Katherine Levin
think, and your political masters as well, because they're the ones with the cash that they give you this money and the expectation is that you're going to deliver and give results. So when you don't you tell them along the way, do you tell them at the end who's kind of looking over your Charlie fan going, are you going to succeed Al in your three years with your however much money you've got every year for this?
- [23:00]
Al Murray
As you'd expect and been a taxpayer yourself, you would want there to be quite a strong grant agreement. And there is and I most definitely have my key performance indicators, targets, goals, monthly reporting requirements and as a taxpayer myself, I'm pleased, you know, I don't want to waste money, but we actually work very closely with that team and sort of co produced a business case and I feel confident that they understand the risk associated with this. It feels a bit like a one team approach. Yes, they are the person who's our grant holder and they will ultimately the Home Secretary wants to see delivery but they understand the predicament and I think working in a transparent way on a monthly basis and we talk to the team in the Home Office every day, there's quite a maturity in the relationship that means it's going to be okay. And they also understand some things may not work.
- [23:56]
Katherine Levin
That's reassuring. I wonder if I could just quote one of the comments to your post on LinkedIn. Timber traffic is quite relevant to this break. A man, somebody called Lee A, he said we need to make sure that AI is used for problems where it is needed and usable rather than AI being attached to things which are easily achievable but unnecessary. Is that a conversation or having at the bunch?
- [24:20]
Al Murray
Yeah, this is a really good point. And we've taken the AI covenants which has been written by the Office of the Police Chief Scientific Advisor that it's got quite high adjectives. So it says, you know, must be Explainable, must be transparent. What does that mean? So we've worked very closely with Muffy Calder and Marion Oswald, who are very good AI ethicists in Northumbria University, to create a responsible AI checklist that takes those sorts of words and translates them into meaningful questions that you ask of your AI. And one of the things, one of the points it really points to, your comment, is just because you can does not mean you should. And if I had an opportunity to describe to you, I use cases, hopefully I would allay the fears of the person who posted that on LinkedIn, because you would see that I think the vast majority of officers and staff would go, yes, I'm glad you put an AI on that. So child abuse, image classification, image identification, disclosure. You know, it really hurts investigators when they've got to spend two weeks populating disclosure schedules. And you can auto populate a disclosure schedule which you then check. You still need to make sure it's done well, or file build. You know, AIs are very good at checking things. And so if you're a detective inspector and you have to check 50 crime investigations every day, you know, it just takes all the fun out of the day. And then you have to report back and say, this isn't good enough. Well, an AI can do that. It can't do it as well as a seasoned investigator, but it can go, look, look, di, there's no CCTV inquiries here. There's no, one's done any house to house. They haven't explored this hypothesis or updated the victim. So immediately you can see, you don't have to check it yourself. The AI can point. So these are things that I think the majority of officers would go, oh, thank goodness. Or crime. Crime identification, classification and allocation and closure. There's 500 pages of home office counting rules on how you write a crime or how you classify crime. You know, most people think it'd be relatively simple, and it is if there's a murder or robbery. But if you're a domestic abuse survivor and you're phoning up and you're describing behavior over 20 years of your domestic abuser, it's quite easy to miss rape or to miss a coercive control. And I can look all of that and go, right, here's, here's where we think you should be climbing from what he or she's just said. I mean, who would argue against that? These are things to the, to the person from LinkedIn, I think they go, okay, yeah, that will make sense because
- [27:02]
Katherine Levin
you're essentially what you're doing there is you're changing your capacity. So instead of it being spent on things which are mundane, then they can be using it on things where actually you do need the human. And when you talk about your baseline earlier, will you be sort of looking to see how many hours I am, I'm saving as a result of doing this? And then these hours are now being put into this activity and we are seeing this improvement. Is that going to be the process of working out your success factors?
- [27:25]
Al Murray
Yeah, absolutely. And it can be this, this is where you can be tempted to cut corners because you're not actually delivering AI. But the measurement of impact is so key. And I, you know, I come back to my interest in evidence based practice here. If you don't baseline effectively, you can never demonstrably show whether it's been beneficial or not and value for money or not. And there's been some really interesting experiments in this field where products that you think are really effective when you come out in a way where you can genuinely measure impact, like using randomized control trials, it hasn't. And by using that type of methodology, you can save quite a lot of license fee money, for example. So there was an experiment where we were using a AI tool to pixelance and some response shifts got it, some response shifts didn't get it. The response shifts that got it. Loved it. But because there was really thorough baselining, it showed that there was no improvement in the speed of statement taking. So you didn't have to actually spend the taxpayers money buying that licensed product because it didn't do what it said on the tin. So that baseline is hugely important.
- [28:36]
Katherine Levin
And also wonder about the suppliers in all of this conversation. So then there are so many AI solutions out there. Look on LinkedIn or look on AI supplier and policing, there'll be loads of things. How are you going to make sure that policing is able to work with suppliers in a way which genuinely makes a difference rather than buying something which actually doesn't make a difference? It doesn't do that baseline that you've just talked about, but it just wastes money along the way. So how do you kind of get that balance right and have those sensible conversations with suppliers?
- [29:06]
Al Murray
Yeah, I sort of feel it both ways. I do not envy anyone trying to sell into the leasing market. You know, who do you speak to? How do you sell? It's really difficult. Do I speak to the chief, Do I speak to the head of procurement? Do I speak to a Chief Superintendent? It's really difficult. But similarly for us speaking with tech Companies, of course, every tech company will say they do everything and just ask us what you want and we'll provide it. You often get handed from a sales team to the actual deliverers who go, oh, no, we can't do it like that. And it's always much more complicated and expensive than you thought. So there's a happy medium here. And one of the significant pillars of least AI that we are building is an AI laboratory function that will, for example, if we go back to the image recognition, huge amounts of providers in here saying, I will ingest CCTV and I will find you the black gulf. Well, the AI lab will have a ground truth set of data and will test all those claims. It will do it once on behalf of the country and say, here's the product to buy and in fact, here's the product to buy and here's a framework to buy it. On the blue light commercial which we've negotiated. And here's how you get into your system. You can imagine, with 43 different forces. I mean, it's a fine example, Catherine, of why police AI has been built to prevent that exact problem. You describe detection, video and audio profoundly different. Huge amounts of product in the market, and they are hugely variable in how effective they are. And all sorts of claims will be made. And with an independent lab to say, okay, let's test more for policing, it
- [30:45]
Katherine Levin
works well, I like that idea. I don't worry, though, that the pace problem will cause an issue for you. So you talked earlier about how it's just developing all the time. When do you, you know, get. When do you get off the train? I think was your metaphor right at the beginning. So when you're doing your laboratory work and then you've got to go through all the procurement process, that's just adding time, time, time, and then everything else is moving on as well. So at what point do you make the decision? And kind of, yes, that's the right point, even though things are moving on, but we still have to go through this due diligence. We still have to be accountable, we still have to make good use of public money. How do you know when that point is?
- [31:21]
Al Murray
Yeah, that's a really good point. And like, the way of doing things historically cannot be the way that we do things in the world of AI for that reason. So there has to be sprints, scrums, rapid decision making, rapid evaluations, and we have to accept that everything is not like it used to be. There will be some, some things will be suboptimal as a result of that and, but we need to speed up. And it, you know, it's a real live, live issue you've highlighted. But, you know, the old idea of, well, it goes through this process, then it will go to a board, then it will go to another board, then it'll be evaluated, then we'll do this, and then if we operate like that, we won't get AI in terms of frontline staff. So decision making fast, evaluation fast, but let's make it effective as well. We just need to think a bit like a business in this space, rather than have the corporate governance that slows everything down to the point where we become ineffective. And that's the responsibility of everyone here. Both commercial providers, me and Police AI, chief constables, partners. Things do have to change.
- [32:30]
Katherine Levin
And then the three years that you, you have now for your £75 million a year over three years. At the same time, the National Police Service may be being developed and running in parallel with you. Do you see opportunities for, in terms of the reduction in number of forces that comes as part of that idea of changing the way that police are structured. Will that help you in terms of delivering some of this pace that you're looking for?
- [32:53]
Al Murray
Yeah, I mean, Police AI was announced in the white paper that announced police reform. And I think we will be one of the first teams to be in the new National Police Service. And so, yes, having that functionality in the center will help. There's always a good debate to be had in relation to economies of scale. And there's two arguments. You know, 40, is it easier to sell into 43? Was it easier to sell into 12? And is it easier to engineer into 43 or to engineer into 12? I mean, the examples from Police Scotland are really interesting. Where you had, I think it was eight forces, wasn't it, that became one, and they became one before the legacy IT in the eight was off. So they're still operating off different IT platforms. And I don't think for a second the merger of the forces immediately means the merger of the technical and digital capability within those forces. That is a huge piece of work that will operate concurrently. So it's going to be complicated. Let's not beat about the bush. It's going to be really difficult. The tech side is a real challenge. So I worked in West Mercia. I'm not saying anything that's inappropriate, but West Mercia merged with another force, then separated with another force. And the technology challenges associated with that marriage, then divorce, were significant and expensive. So any change in this area is difficult and the investment needed to eradicate that technical debt is significant. But sometimes, you know, bigger is not better, Sometimes decision maker making is slower in larger entities and larger businesses. So there's always a balance, I think.
- [34:29]
Katherine Levin
Okay, that's useful. Just before we come to end our interview, we're coming to time, unfortunately, although it's been fascinating, I just wanted to ask you about the policing problem book. So we saw that published maybe the
- [34:41]
end of last year.
- [34:42]
And in terms of the 13 problems set out in there, will the AI work that you're doing be used to help deal with some of the problems set out there or is it completely separate your work?
- [34:53]
Al Murray
No. So luckily I've got quite a lot of people who I link into officially. So the Science and Innovation Coordination Committee that sort of commissioned the problem book under the leadership of Jeremy Fawn, the Chief of South Wales, I have a dotted line into and I explained to that team what we didn't we used when we looked at our use cases that I've described today, we used the problem book and it's really helpful and we say, okay, here's the main problems we what's the AI solution in some of these problems? Is it scalable? Yes, it is. Okay. We're going to invest in building image recognition, for example. So yeah, very close relationship with that problem.
- [35:30]
Katherine Levin
Yeah, I'm glad there's some connections there because I often spend time talking to people from the DDAT committee, the Digital Data and Technology Committee and there's the Science and Technology Committee and now there's you. So just trying to understand it all fit scale. So it sounds like you're all talking to each other, which is always good to hear. So I think you've got your work cut out. I think this is a huge challenge for you and your team. Excited to see how it develops over the next three years. And I hope that you continue to write these posts on LinkedIn so that people can understand all of these difficult questions that you've been through about responsible AI. It is a steep learning curve. Where do you focus? How do you get leaders? How do you move people beyond the sort of risk averseness that they might feel when they don't want to take, you know, the step forward, waste money, how do you turn into a startup? I mean, these are big questions in policing. It's a, it's amazing to hear that it's happening.
- [36:22]
Al Murray
Yeah. So it's a fascinating time to be in policing and to be in AI and policing, not just from a technological point of view, but from a cultural point of view as well. So yeah, it's good. And thanks very much for having me on.
- [36:35]
Katherine Levin
Wish you all the best. Thank you so much for talking to me.
- [36:38]
Al Murray
Cheers Katherine. All the best.
- [36:40]
Katherine Levin
There's a lot here about responsible AI. Al says we are delivering responsible AI into the hands of police officers. It's not a theoretical idea, it's the reality. And there's a real pace to the leadership here, along with a pragmatism about getting on, trying stuff out, being okay if it doesn't work. It feels like a start up, he says. This isn't language you hear in the public sector. It's really very refreshing. But where to put the effort? How not to be dazzled, puzzled by exciting things that don't drive the improvement that the public would want to see? It's hard, I think, for people to do that. And yet ironically, it's the legacy systems that keep him up at night. It's good to hear lots about benchmarking and impact, and it all leads nicely back to what he says about evidence based policing. My thanks to Al for talking to me. That's it from me for now. Episodes of this series drop every Tuesday. This series is brought to you in partnership with the Emergency Tech show and the Police Digital Service. The show takes place on the 16th and 17th of September at the NEC in Birmingham. Registration is now open and the link is in the show notes. If you're interested in finding out more about Police AI, one of Al Murray's team, Lewis Gordon Sinclair, will be with us on stage to bring us up to date on what they've been up to. So be sure to share like and if you haven't already, subscribe to For Every Response, the podcast from Emergency Services Times.