Using Technology to Improve Crime Analysis with Andy Wheeler

Today's guest is Andy Wheeler, a renowned criminologist, data scientist, and founder of Crime Decoder. Andy is a terrific voice for understanding the role of technology in analyzing and decreasing crime.

We talk through Andy’s work with police departments, including crime trend monitoring,  predictive analytics, focused deterrence, and the challenge of turning data into useful operational decisions. We also discuss the practical realities of artificial intelligence in criminal justice. Rather than focusing on the hype around AI replacing workers or transforming society overnight, Andy walks through how these tools are actually being used today: helping analysts write code, automate repetitive work, build dashboards, and make better use of messy public safety data.

This is a great conversation about how the future of criminal justice analysis is already here.

Andrew P. Wheeler, PhD and founder of CRIME De-Coder, collaborates with police departments across the United States on custom software and data analytics. His work focuses on predictive analytics, operations research, and policy analysis. See his firm and other resources at https://crimede-coder.com/.

Andy also has a recent book, *Large Language Models for Mortals: A Practical Guide for Analysts with Python*, https://crimede-coder.com/blogposts/2026/LLMsForMortals

Transcript

SPEAKER_00 0:00

I'm Jeff Asher, and this is the Jeffalytics Podcast. Artificial intelligence is surrounded by big promises and even bigger questions. Depending on who you ask, it's either about to transform society or destroy it. My guest today spends a lot less time talking about the hype and a lot more time talking about what these tools can actually do, especially in the world of criminal justice. Andy Wheeler is a criminologist, data scientist, and founder of Crime Decoder. His work sits at the intersection of crime analysis, technology, and public safety. In this episode, we talk about how AI is already being used in public safety, where it's helping, where it's falling short, and why the future is likely to be more incremental than revolutionary. We also get into how technology and AI fit into focus deterrence, predictive analytics, and the challenge of making better decisions in an increasingly data-rich world. Technology can provide more information than ever before. Knowing what to do with it is a different challenge entirely. Let's get started. My guest today is Andy Wheeler. Andy, thanks so much for joining me.

SPEAKER_01 0:58

Yeah. Thank you for having me, Jeff.

SPEAKER_00 1:01

Question I start everybody off. And you a little, I think, more mysterious than most. Like, what is your background and what is it that you say that you do? What is it that you do?

SPEAKER_01 1:11

Yeah, so to start, I got my PhD in criminal justice at SUNY Albany. I most of my work was doing number crunching with police departments. So basic research, policy analysis, operations research, and predictive analytics for the most part. I was then a professor of criminology at the University of Texas at Dallas for a few years. And then in 2019, I went over to be a data scientist in the private sector. So day gig, I work uh for a healthcare company that uh examines fraud, waste and abuse for Medicaid claims. But in addition to that, I still do consulting with police departments. And so I have a consulting firm called Crime Decoder, and that's kind of where we probably overlap the most then.

SPEAKER_00 2:01

I would imagine so. Although the fraud stuff sounds fascinating as well. So what sort of motivates you to do all of this work?

SPEAKER_01 2:08

Yeah, I mean, a lot of it's just based on my interactions with police departments and my interest in really technical problems. I view if somebody comes to me and they're like, hey, I have this problem, whether it's monitoring crime patterns or doing this type of intervention or building software, I enjoy working on those types of things, probably the same way that some people enjoy doing crossword puzzles or different types of puzzles like that. I just enjoy doing technical work with police departments.

SPEAKER_00 2:42

And so what kind of problems are you solving with police departments? Is there a typical type or is each one different?

SPEAKER_01 2:48

Yeah, it's a lot of times it may be technical algorithm development. So I worked while I was getting my PhD, I worked as a crime analyst. And so some of those projects were just born out of different projects that I was working on at on the time while I was a crime analyst. So like one of the examples was just simply to tell if I'm monitoring crimes week over week. So if I'm in a, if I was in Troy, New York, so we may have 10 car break-ins in last week and then 20 car break-ins this week, is that a significant increase? Is that at the point where I should go and dig in to those crimes and try to identify if there's like an itinerant group that's committing a bunch of offenses, or if that's just like normal variation over time. So that's one of the projects I worked on is just monitoring how do you how do you develop metrics to monitor crime over time and identify those spikes. Other examples are identifying the best individuals to give focus deterrence messages to. So one of the interventions that police departments do is basically uh gang network interventions, where the focus deterrence model is basically you give a deterrence message to the whole gang. If one gang member messes up, we're gonna go after the rest of the gang. Uh, so I developed an algorithm about who to best prioritize to give those particular messages to. And those are just a couple of examples in my career. So, like I said, a lot of times it's just born out of a particular problem a police department comes to me with, and then I just focus on the best way to solve that problem.

SPEAKER_00 4:30

I want to dig in into the folks' deterrents thing because it's that's kind of really interesting. Did you have any sort of broad conclusions? Is there a general solution to that problem, or is it sort of a very much a case-by-case each agency and each individual is different?

SPEAKER_01 4:46

Yeah. So to back up a little bit on that, folks may, if we're talking about gangs or or groups, a lot of times they're not like uh they're not like bloods and crips, it's just like loosely affiliated groups of mostly younger individuals for a lot of different cities. And so folks may think that groups are entirely connected. So everyone in that gang or that group knows everyone else. In reality, they're not. And so you may have two people affiliated with that group or gang, but do but don't actually like know each other, don't necessarily hang out. And so in the focus deterrence initiative, basically they call in individuals from that gang and deliver that deterrence message with the expectation that you're going to disseminate it to the rest of the group. When I looked at the actual data on who they called in for a few of the different jurisdictions that I was working with at the time in upstate New York, I could tell that they called in people that were very suboptimal to basically spread the message. So if you draw, if you look at a network graph, it would basically be they only called in individuals that were on the periphery of the graph and not like embedded in the graph very well. And so I basically just developed an algorithm about who to prioritize. So if you can only call in five people, who are the best five people to call in? And so it really it probably applies to any department doing that focused deterrence intervention. But honestly, if folks just drew the graphs and then said, like, uh, we should probably call in these people who are more in the middle of the graph as opposed to the outside of the graph, that probably would improve that intervention by for a lot of different departments.

SPEAKER_00 6:37

Do you have any impression as to why they would operate suboptimally from an organizational standpoint? Is it just because these are the easiest people to call in? Or was it throwing darts at a dart bard? Was there any sort of consistency there?

SPEAKER_01 6:50

Yeah, it was mostly just because people didn't look at the information. It's definitely the case that you can't force people to come into those call-ins. And so uh my experience with, and this would have been in Albany or Syracuse, New York, at that particular time. There's basically groups, it may just be a detective who's working with a crime analyst and is saying, Hey, let's call in A or let's call in B, or invite those two individuals to the call-in, I should say. And so it was very ad hoc. So there wasn't just much thought put behind about who to call in at all. And so it was basically just taking the idea about like we should look at the actual network and then prioritize who to do the call-ins for based on actual data, as opposed to, like I said, it was just somebody sitting at a desk somewhere and is like, uh, let's call in Andy Wheeler, as opposed to actually looking at data and making an informed decision.

SPEAKER_00 7:47

And what is the response when you sort of provide this analysis of basically you're doing this wrong? Are people receptive? Or is there sort of pushback on that?

SPEAKER_01 7:56

It's definitely I can definitely see scenarios where there's there's going to be pushback. Honestly, if departments are already working with someone like a researcher like me, they're basically already pretty receptive to it. And so most of the departments that they're self-selected in that they actually want feedback like that. It is a little bit more difficult if, say, you're a crime analyst in a department with sort of an old school chief that, or old school detectives that don't really want that level of feedback. But a lot of times to it improve receptivity, you just need to actually provide useful advice. A lot of crime analysts will provide a bunch of numbers, but not really give useful advice for folks. So if it's very specific, you should call in B instead of A, a lot of folks won't necessarily give pushback to that, especially if it's like you can just give plain reasoning about why that person is better or this or doing this intervention is better. A lot of people, a lot of analysts struggle with that though.

SPEAKER_00 9:00

We got very very into the weeds very suddenly. So I want to take a step back and just talk about sort of when you kind of look at the big picture, what are the overall misperceptions that you see and misconceptions at an agency level, at the public level, around all of the work that you do?

SPEAKER_01 9:15

Yeah, definitely. And so I think probably in terms of public perception, in terms of what crime analysts do, a lot of people don't even realize that police departments are really independent. So I'm in the research triangle area of North Carolina now. People don't realize that the Durham Police Department and the Raleigh Police Department are two totally isolated systems. They basically don't, if something happens in Durham, a crime report, there isn't anybody in Raleigh that can perceive that particular crime incident report. And so I think a lot of people, especially if you're you have concerns about like the surveillance state, a lot of people think police departments are just all interconnected. The FBI can go in and see whatever information that they want. In reality, it's just we have all these little isolated departments doing their own thing.

SPEAKER_00 10:10

And so now I want to talk about AI, because I know that you're a heavy user, and I think uh you've written a ton about this, you do a lot of this, and you talk about this extensively in blog posts and on LinkedIn. So can you walk me through sort of how do you use AI just to start?

SPEAKER_01 10:26

Yeah. And so, like I said, I my day gig, I work as a software engineer. And so that's really the biggest area that I've seen AI use is basically just using AI to help you write computer code to write software. And so with the recent, so you had Chat GPT come out in late 2022, I believe. And so a lot of different consumers that have seen, I like have experience of using that chat application. You can just go into Chat GPT and say, hey, help me write this Python function. And even when it first came out, it did that, it did that very well. And part of the reason is that there's just a lot of computer code on the web and they train the models basically on those historical legacy code. And it's really only gotten better over time. And so if you're a software engineer and you write computer code, just using the tools in a very simple way, hey, I have this function or hey, I have this idea for an app, they worked really well even back in 2023, and they're just getting better over time to do that. And so that's the main application that I have been using AI day to day, both in my job and in my consulting work.

SPEAKER_00 11:45

And so how can this be applied sort of to the crime, criminal justice space?

SPEAKER_01 11:51

Yeah. And so the way that I view it is I expect there to be sort of two different broader use cases of it. One of them is in terms of software applications, different groups will basically bake in some of these different AI applications into their products directly. And so, like there's a little button in Microsoft Word documents to use Copilot. Different software tools, whether it's uh companies like Peregrine that do data analytics or whether it's records management software, they'll basically start having an easy button of, hey, can I use AI to do XYZ? And if we're talking about crime analysis applications, it may be things like, hey, show me the crime trends of burglaries from motor vehicles, or hey, pull out different burglaries that use this particular type of modus operandi. And behind the scenes, the AI just writes the computer code to go do that particular analysis. The other application is just cities themselves. The AI can be used to help, because it can be used to help write software, it really makes it easier for cities themselves, whether it's police departments or other city organizations, to really write software themselves to do different applications. So instead of paying an external firm to have software to do data analysis or to do a dashboard, now it's basically the case that a like a pretty well-motivated crime analyst can go in and ask the AI, hey, help me build this dashboard and write out a decent application on their own without having to pay an external firm to do that.

SPEAKER_00 13:42

Is that something that's happening? Do you have sort of examples of the kinds of things that those like well-motivated analysts are producing?

SPEAKER_01 13:50

Yeah, it's not happening currently all that much now. So I don't have any real great examples besides like little idiosyncratic things analysts are building on the sides of their desk now, which are really not any different than things that they've been doing. So, like, say automating a weekly report, now they're using some of the AI tools to help them write the computer code to do that. So not so nothing real sexy to talk about there. But I do suspect as the tools get more widely integrated and accepted that we'll be seeing more of that type of work. Um Yeah.

SPEAKER_00 14:25

Why do you think that is? Is it just a people are slow to adopt new technology things or bureaucratic?

SPEAKER_01 14:32

It's partly due to speed, but it's also partly due to the technical capabilities of a lot of analysts or city workers for particular departments. So uh one of the things that I do is training for crime analysts. So I'll go in and help analysts get up to speed with Python, for example. And so the biggest struggle with most of the departments that I work with is you'll have a mix of younger folks who are really interested in learning new technology, but then you have a lot of folks who have been working there for 10, 15, 20 years who are just not interested at all and really just want to work at Excel or do whatever sort of work that they've done for the past 20 years. And so part of it is that there's just resistance to change, as well as like technical capabilities of folks. But I do think that it will improve over time, just naturally, but simultaneously it's gonna happen over time just due to like fiscal demands for people to be more productive.

SPEAKER_00 15:34

And thinking about AI from a specifically strictly policing standpoint, I had Ian Adams on a couple of months ago and he talked about his research looking at the report writing technology. Have you thought about the ways like in that these technologies are implemented not from an analyst standpoint, but from a wider policing standpoint and the degree to which they're either effective, or to Ian's um research that shows that it's not inherently saving time right now?

SPEAKER_01 16:04

It's definitely super hard to implement these different AI tools and save time. And so at my day job, one of the things that we're doing is we're building similar tools to basically help nurses do audits of Medicaid claims. And so it's it's really hard. And so a lot of times we're spending significant amounts of engineering to save nurses like five minutes here, six minutes there to do those particular audits. And so they're not a lot of the applications that I see, there's a there's definitely a lot of buzz around AI. And hey, saving a bunch of folks five minutes on something is definitely worth doing. But for a lot of the applications, I don't foresee being able to cut the human out of the loop entirely. And so we're talking more about some of these marginal incremental improvements as opposed to like magically it's gonna cut report times to five minutes for things that took an hour before. And in terms of whether we're talking about the police departments or different types of government agencies, a lot of agencies just don't have the infrastructure in place to be able to really effectively enforce a like even know how long that they spend on particular things. So I I think like a good counterexample are call centers. So if folks are familiar with call centers, those folks are like metrics out the wazoo. So folks know how long folks it take to answer calls, how long they spend on calls, how much of their downtime that they have. And so most State Departments, whether it's we're talking about what police officers do day to day or analysts do day to day, don't have any internal capacity to like determine if an officer saves five minutes on writing a report now. So they don't even have the infrastructure set up to know if it's saving people time. The last part of that is it's really hard to generate savings in terms of labor cost savings with AI, because the only way to actually realize savings is if you save a police officer five minutes of their time, you aren't actually realizing any labor cost savings. You're already basically you're hiring the police officer full time around those hourly wages, essentially, to only if you're actually if your goal is to save money, you actually have to reduce headcount. And so most places, they're not interested in reducing headcount. They're really just interested in doing the job better and being able to do additional work. And so I suspect it in terms of the AI writing, given the advancements of the tools and get like knowledge of the current quality of police officers writing reports, I think probably the better benefit of those tools will be it'll improve report writing over time. It's not there yet, but I think the future state, there's a good chance that using those types of tools can improve report writing for officers, but I don't know if it'll be like massive time savings in that.

SPEAKER_00 19:14

I foresee always needing a human. Can you uh I heard you do this on the Jason Elders crime analyst podcast, whose name I'm suddenly blanking on? I'm sorry. Uh I've been a guest on and it's killing me. But you just sort of walk through the different tools that are available. And can you do that here? Talk through the different AI infrastructure, what tools exist from a like a Claude and and Codex and sort of the differences between them?

SPEAKER_01 19:39

Yeah, definitely. So Jason's podcast, which I'll give a little bit of plugs, was since it was mentioned, Law Enforcement Analyst Podcast Leap. He does a really great job. He's done it for a long time. And absolutely. So yeah, so I recently wrote a book, Large Language Models for Mortals, a practical guide for analysts with Python. And one of the reasons that I wrote that book is like I said, I'm I work in software engineering now. It really, when ChatGPT came out, it massively shifted the software, the focus to not only using those coding, those tools to help you write computer code, but also in applications to use those basically those same models under the hood to do a lot of the stuff I work with. It's people think AI is cool, it's so boring. Like I'm mostly re working to replace fax machines and like document processing and things like that. And so in that book, I basically give an overview of the different types of common applications for those large language models. So one of the most basic ones, like I said, it's super boring, is just what's called like structured output extraction. So imagine you get a PDF or a scan or an image of a check or a police report or basically any type of input document, these tools are very good in ways that were basically a step change over prior over prior tools to be able to just take those PDFs and extract out information from those. So if we're going with a crime analysis example, you can take an officer's just sort of plain text description of what happens at a particular incident and then extract out things like names and dates and addresses and even more technical things like the modus operandi for how somebody broke into an apartment or things like that. So it's basically like instead of writing a specialized tool to pull out addresses, you have like this general tool now. You can just tell it what to extract and it will extract that information. And so it's not a real interesting application, but a lot of different things basically fall into that structured output extraction sort of camp. And so behind the scenes, It's likely the case that the AI report writing tools are doing that at a particular step in the transaction. So Axon has all the body worn camera footage. They basically turn the footage into a transcription and then extract out the information in the transcript to basically fill in the police report. And so a lot of different, pretty boring applications, but things that that help save time sort of look like that. One of the other applications for LLMs are basically they're very good at summarizing documents. And so imagine you're a police officer in the field, and I like I enjoy watching the body worn cam or cops or things like that. And so one of the ones recently was they pulled over somebody in an auto drive Tesla. It was a few years ago, and the officer wasn't, didn't know which citation to give the individual who was sleeping, even though they were on autopilot. So they're like, is this reckless? Is this reckless driving? Like, what is this? One of the example use cases of AI is basically what's called RAG, retrieval augmented generation. And that's just you have your documents and you have a system set up in place where essentially you query the doc, you ask plain text questions like, hey, here's my incident. What's the I have somebody asleep at the wheel using Tesla Autopilot. What's the most likely criminal charge or or traffic citation I should use for this scenario? And so the way that those applications work now is you can go in Chat GPT and ask that and it'll give an answer, but it won't necessarily be an answer based on your department's documents. And so the way to change that to be based on your specific state codes or your particular department uh documents would be you build a system where you essentially scan over your internal documents to find the most relevant ones, and then only feed those relevant documents into the chatbot and then have the chatbot summarize based on your local documents. And so that's what's called RAG. The last application that I talk about in the book is what's called, it's kind of complicated behind the scenes. It's called tool calling. But ultimately, when ChatGPT first came out and you ask a question, so say um it came out in like late November 2022, I think sometime around then. And it was only trained on data like earlier in the year. So if you asked it a question, say, hey, tell me about crime trends in December 2022. So after the data that it was trained on, it wouldn't be able to answer that question. It doesn't internally in the weights and in the way that the model was trained, have that information. The way to solve that though is to do what's called tool calling. So if you do that same question now, what ChatGPT will do is basically under the hood and be like, I know that I don't know this information. I need to actually go to external sources. And it may go to like one of your websites, Jeff, or one of your news articles in the Times or things like that, and go look up those particular crime trends that are more up to date and then return back the information. And it's so it's the same idea as rag before, but it's basically the model can go into multiple steps and get the results of those tool calls to feed back in to further downstream processes. And that's pretty powerful because it can not only be like searching web documents, but it can basically be tool calls to do any intermediate output that you want, like whether it's querying a local table or pulling in different documents, like I talked about in the prior example, or even coders, a lot of times what we're doing is we're having the Claude or Codecs or whatever tool write Python code to do that work. So basically we say, Hey, can you answer this question? And it'll write a little Python code to try to answer that question and then return the response.

SPEAKER_00 26:21

How do you account for hallucinations in the machine that we know that there's imperfections? Uh when ChatGPT first came out, I would like to be like, you know, what has Jeff Asher written? And it would come up with all of these articles. I'd be like, those would be great. You know, those sound exactly like things I would have written. I haven't written them, but and obviously the models have improved significantly since then. But are there techniques or responses to basically just hallucinations and the fact that these things aren't perfect?

SPEAKER_01 26:47

Yeah. It it really depends on the particular application. And so the example I gave with tool calling. So one of the popular ways that people try to critique these models is you can ask it how many R's are in strawberry, and it would give not the correct answers. Or you can ask it to do some simple math questions and it would get the particular math questions wrong. And so I'm not sure there probably is a real technical definition of hallucination, but I'm just broadly saying it's like anytime it gives an answer when the answer is the answer is wrong. Now, if you have a particular situation that there's a known definite output, you can basically use tool calling like I gave the example before. So instead of just having the LLM basically take the text in and give its guest answer for how many R's are in Strawberry, it actually writes Python code to go and check the text, check the number of Rs that are in that word. I know that that's a trivial example, but it basically works that type of instead of the LLM directly answering the question, it writes computer code, and then you can audit the computer code and make sure that the computer code is right to answer the question. That's one of the ways, and that's really the most popular way that people solve that for current software engineering applications now. It's the system, and it's one of the reasons why it's so popular in software engineering, is because we have constrained systems and like we know the correct inputs and outputs. And so you can basically just write what are called tests. You can rewrite tests that are like, I know if I get this input, I should get this in output. And so the agent writes computer code, you have your tests, and if you know you failed your test, like something's wrong in your computer code. Now, for the case of the more general, like summarized information, I kind of liken it to an on-demand Wikipedia currently. Like it'll do the summaries, it'll often give you citations to like external, external sources, and it's pretty good, but it's definitely not infallible.

SPEAKER_00 29:07

And sort of what advice do you have for kind of the non-software engineer user that for everything from sort of the non-technical crime analyst to the policymaker to the just, you know, your relative that wants to use AI and chatbots?

SPEAKER_01 29:21

Yeah, it's definitely the even just the free tools now. I definitely encourage folks to go use them. And like I gave the example, I think it's pretty equivalent to an on-demand Wikipedia with sort of similar quality currently. And so any folks who have been in school recently knows your professors do not like you to cite a Wikipedia page if you're trying to use a point. But Wikipedia overall is pretty high quality. And so it's definitely worthwhile for you to go in. And if you have a general question, to go in and just ask these tools, whether it's the different tools are pretty similar in quality. So it doesn't really matter whether it's ChatGPT or Claude's tool or Google's Gemini. They're all to me, they're very exchangeable, honestly. And so you can go in and ask your question, but you basically need to pay attention to those external resources that they've provided. And in addition to that, one of the biggest problems with the models now is they're what's called psychophantic. And so they'll tend to respond positively if you ask a question. So if you ask, so say you were asking about a medical condition and you're like, hey, I think I have high cholesterol. Here are my symptoms. It's more likely to respond affirmative, yes, I think you have high cholesterol, as opposed to if you just ask the questions and say, here are my symptoms. They tend to basically confirm your suspicions. So you definitely want to be, when you're asking these questions, you want to be critical of the sources and you want to avoid basically telling the machine what you think the answer is to begin with. You kind of want to come into an open mind with those conversations.

SPEAKER_00 31:04

So I keep asking it, I think I'm smart. Do you agree? And it keeps telling me yes.

SPEAKER_01 31:08

So that's you're telling me that that's not I think you're smart, Jeff. So I wouldn't worry about that too much. But basically know that the machines are definitely fallible. And so there are a lot of people that just take it as ground truth. And if you're doing that, if you're at a workplace and you just have the LLM write your email and it's like 10 pages, nobody likes that. So like take a little bit of time to like read the output, sort of understand it. They're really great as teaching tools as well. That's probably one of my most common use cases of them, is like, I don't really understand this. Can you give me like a more a more general explanation and then go deep dive into that, into whatever you're learning, whether it's I'm mostly focused on tech stuff, but obviously most people aren't going to go and ask Python questions or database questions to Chat GPT like I do? But they're really excellent tools to go and help you teach yourself those applications.

SPEAKER_00 32:09

Just know it's like it's fallible, just like any human would be if they're Is there a prediction being made today about AI or crime data or any of this type of work that you think will sort of be seen as ridiculous in a few years? Kind of like if you'd asked yourself four years ago about where ChatGPT would be now or where the this landscape would be now, you never would have guessed it's here.

SPEAKER_01 32:31

I haven't seen it specifically for crime analysts in particular, but to me, I'm really in the camp of AI is likely to be a complement to current analysts or software engineers or basically anybody who does stuff on the computer. So there's a large contingency now that really thinks that AI can really disrupt the industry and and basically take everyone's jobs. And so this the the one of the founders of anthropic, Ariel Amadai, like he's big into that, basically going online and saying, like, everybody's we're gonna take everybody's jobs, essentially. Which doesn't endear the technology to Yeah, and so I'm really being like a professional software engineer, working with the types of things that analysts do, I really don't, it's really hard to cut humans out of the loop. And so I'm really not, I'm really bearish on that. I think that there's a much easier path, though, to basically have folks who do work on the computer have them use these AI tools to really be complement to their work now. And so I do suspect it will be like into one of the common things that analysts sort of get stuck doing are open records requests. So I know a bunch of analysts who say the majority of things I do with my job are open records requests. And so that job is it should almost be automated. It should almost be you build a tool that basically takes in the records requests, builds the basically the query language, the SQL, to generate it, and then just give the results and the human do a quick pass to make sure if it makes sense or not. And so that analyst, even if they did that for 90% of their job before, there's gonna be other work for them to do now. They're gonna be able to go do more interesting work, basically, with their jobs. So definitely bearish on AI taking a bunch of jobs. But that said, I think it can be a real boon for basically everybody who works on a computer now. So it will be transformative, but probably more on the margins and less on the it's gonna take everybody's job.

SPEAKER_00 34:44

Aaron Powell And how do you try to account for some of the the clear downsides of AI, talking about you know the environmental downsides, the sort of injecting computers into art and things like that really provoke strong reactions sort of against the technology?

SPEAKER_01 35:00

Aaron Powell Yeah, there's definitely one of the areas I've thought about it the most is actually in writing. So there's a lot of folks using using these different tools to the term of art now is slop. So to produce AI slop, essentially. And so there will always be there's a lot of things like art or writing that have a lot of objective or or not objective, but subjective, like what is good or what is bad. That said, a lot of folks can look at particular art or look at particular writing and basically use the Potter Stewart approach and saying, like, this is good or this is bad. And so the that subjective perceptions of the work are not going to go away. And people will have to learn how to use these tools to put quality work. And a lot of things in terms of like subjective, you can ask ChatGPT to generate a blog post and like first sort of this superficial pass. A lot of times it seems like really good, but then when you pay a little bit more attention to it, it's pretty vapid, it's like filled with a bunch of platitudes, doesn't really say specific information. And so there will still be individuals who basically can tell, okay, this is bad, but how do I use these tools to help me do my job better or faster or smarter? And so I think in it's the complement perspective. So instead of like taking a human entirely out of it, how can we use these tools to do what we're doing now, but do them better? I think that's the path forward that the most successful individuals will end up being in really the near future, not even that long-term of future.

SPEAKER_00 36:54

And so I guess that takes me to my last question is sort of what does the future look like? The short term, the medium term, the long term with this technology.

SPEAKER_01 37:01

Yeah. I think uh for the short term, there's really just going to continue to be incremental improvements into these models that come out. And so for the day-to-day folks, there's not going to be a new anthropic or open AI model that comes out that like fundamentally that like takes your job tomorrow. That's just not going to happen. They're going to be these incremental improvements to the models, which even at this point, it's hard for me to tell like the massive them the improvements to the models in my particular day-to-day work. And so they've been very helpful, but the difference between the Sonnet 4.1 model and the Sonnet 4.6 model is not all that, not that big of a change in my day-to-day work for helping me write software, essentially. And so the biggest sort of medium term that I'm not quite sure how it's going to shape out is these models now are getting more and more expensive. And so there's this idea of induced demand. And so a lot of times they use it for building highways on the road. So, like if you build more highways, it tends to people who took surface roads before will take the highways now. So it's basically a free good. And so a lot of people who weren't using that before are using it now. The AI models aren't necessarily a free good, like they cost money. And so those particular in the medium term, these different vendors, I'm not quite sure what the pricing models are going to look like. So now you pay 20 bucks to be able to use some of these tools on a regular basis. I think that they're going to get more expensive, but there's going to be some more, especially for companies or enterprises rolling these out to like 5,000 individuals at once, there's going to be a lot of competition among OpenAI and Anthropic and Google about pricing. So I do expect that these models to get cheaper over time. And in terms of like really far out, I mean, maybe not super far out, but maybe like say in the next 10 years, there's really most of the work currently right now has really been focused on text in and text out. And so there's definitely some of the tools now, you can also, you can also include images and say, like, hey, describe what's in this image. A lot of the work for model improvements, I see in terms of uh there there needs to basically be those same types of level of innovations. And it's starting to happen now, but it's really like it's really years behind where it is for text for video and audio and images, and that's where a lot of the innovations are probably going to come over in the criminal justice side. So being able to basically say to Google's model, hey, here's my body worn camera footage, like pull out, pull out the the important points that I want to know. Or monitoring a CCTV camera, hey, give me an alert when somebody fires a gun and we see a muzzle flash. The models right now, the like the big those are mostly being developed by sort of these like small idiosyncratic teams. I suspect that the models from Google or OpenAI anthropic will eventually be able to compete with those. And that really opens it up for like just a local department to build that type of software themselves, as opposed to needing like several million dollars to be able to develop those applications.

SPEAKER_00 40:51

Crazy frontier.

SPEAKER_01 40:55

Yeah, so go to my website crimede-coder.com is is probably the best place to look me up. I'm also on LinkedIn, and so I have a crime decoder page on LinkedIn where I where I post the most frequently as well.

SPEAKER_00 41:15

Well, thank you so much for coming on. I appreciate it. I know I learned a lot, and hopefully others did as well.

SPEAKER_01 41:21

Yeah, thank you very much, Jeff.

SPEAKER_00 41:23

Thanks for listening to the Jeffalytics Podcast. Be sure to subscribe and to learn more, head on over to ahdatalytics.com for more information and previous episodes. If you like what you heard, please leave a glowing review, which will help others to discover the show. Until next time, I'm Jeff Asher.