Preview Mode Links will not work in preview mode

Apr 14, 2022

In Episode 5 of Series 7 of The Rights Track, Todd is in conversation with Amrit Dhir, Director of Partnerships at Recidiviz – a team of technologists committed to getting decision makers the data they need to drive better criminal justice outcomes. 


Todd Landman  0:00 

Welcome to the Rights Track podcast, which gets the hard facts about the human rights challenges facing us today. In series seven, we're discussing human rights in a digital world. I'm Todd Landman, in this episode, I'm delighted to be joined by Amrit Dhir. Amrit is the Director of Partnerships at Recidiviz, a team of technologists committed to getting decision makers the data they need to drive better criminal justice outcomes. He has previously spent over a decade at the intersection of technology and new business development, working, for example at Sidewalk Labs, Google for Startups and Verily. Today, we'll be exploring the practical uses of technology and data in the criminal justice system. So Amrit, it's great to have you on this episode of the Rights Track. Welcome from California.

Amrit Dhir  0:44 

Thank you so much, I'm really glad to be here.

Todd Landman  0:46 

It's great for you to join us. And I want to start with a simple question. We had a guest - Sam Gilbert - on our last episode, we made this distinction between the sort of data for good and data for bad and there's a very large sort of argument out there about surveillance capitalism, the misuses of data, you know, behavioural microtargeting and all these sorts of issues. And yet I see that where you're working at Recidiviz there's a kind of data for good argument here around using technology and data to help criminal justice systems and the healthcare sector. So just briefly, could you tell us about this data for good and data for bad distinction?

Amrit Dhir  1:19 

Yeah, well, as with most things, I think it's difficult to pigeonhole anything into one of those camps, everything it seems, can be used for good or bad. And so data itself is not one or the other. I think it's about the use, I think that's what Sam was getting at with you as well. With Recidiviz, you know, what we've understood is that data that's been collected over a long period of time, especially in the context of the United States, and our unfortunate kind of race to mass incarceration, from basically the 1970s until about mid-2010s. We've collected a lot of data along the way, and we're not actually using or understanding that data. And so what we do at Recidiviz is we bring that data together, so make it something that can be better understood and better utilised, to help reduce prison populations to help drive better outcomes. So we're focused on taking data that's been, again, collected over quite a long period of time and consistently collected, but also making it better understandable.

Todd Landman  2:17 

So this sounds like big, messy, disparate, fragmented data, is that correct?

Amrit Dhir  2:22 

Most of those things, most of the time. It's definitely fragmented most of the time, it's not always necessarily what we'd call big. Because, you know, coming from Google, I think of big in the terms of, you know, search query type volume. So in corrections, it's not necessarily that big, but it is certainly messy, and it is certainly fragmented.

Todd Landman  2:42 

You know we had a guest on Rights Track, some while back, David Fathi from the American Civil Liberties Union, he explained to us the structure of the American sort of prison system, not justice in itself, but prison system with, you know, 50 state prison systems, plus a federal prison system and a mix of public and private prisons. So it's a mixed picture in terms of jurisdiction, the use of incarceration and of course, the conditions of incarceration. So what's the sort of data that's being collected that you find useful at Recidiviz?

Amrit Dhir  3:13 

Yeah, I'll actually add a piece of that as well, you're exactly right to say, you know, every one of the 50 states has a different system, the federal system is itself separate. But then there's also county jails. And those systems are running completely separately from even the states that they're in. So it is messy. And the data also extends, by the way, so we're talking about what we consider the back half of the system. So once someone has already gone to prison, we think of that as the back half. Whereas there's a front half of the system as well, which is the courts, your prosecutor and defence attorneys, and up to policing. And so all of those different segments have their different datasets as well. At Recidiviz we're starting at the back half, largely, because we think there's a lot more impact to be had there, at least for now. And the data extends to many things. So it can be first of all, admissions data. When someone comes into a facility, what sentence did that person come in with? Where is that person going to be in the facility? As in like, where's that bed? And then, as often happens, there are transfers between prisons, within prisons. That's another set of data. There are programmes that the person may be participating in. Some of these are built with the spirit of rehabilitation and reintegration into society. Those are important and knowing how they work and when they work, and if they work is important. And then when someone gets out of prison, that's not the end either. We've whole infrastructure of supervision. And broadly, those are grouped into two categories - parole and probation. And someone may be back out in their community and still under a degree of supervision that's more than what someone who has not been in prison goes through. They have to check in with their parole officer. They have certain requirements, they have certain restrictions. All of those are data points as well. How are you checking in with your parole officer? Did you have to take a drug test? Did you ask for permission to leave the state, all of those things. And as you can imagine, even just by the list I've given you, which is just a very small percentage of it, all of those are sitting in different data silos and are interacted with by different people within the system and it gets pretty tricky.

Todd Landman  5:21 

And you collect data on the sort of sentencing? So you know an analysis of that plus demographic makeup of the prison population, time served? And also, the use of the death penalty and or deaths in custody - is that data that you can collect?

Amrit Dhir  5:37 

Yes, so we can do all that. And I'm glad you pointed out racial and demographic data, because that's a big part of what we do and what we highlight, because you may not be surprised to hear that in the US, there are like pretty severe disparities when it comes to race, ethnicity. And these are things that departments of corrections. So those are the executive agencies within each state, we usually call them department of corrections, although they'll have different names in different states. They have this data, and they want to make better sense of it. Their stakeholders want to understand it better. So generally, these agencies report to the governor, but they're also accountable to the legislature. So there's a degree of sharing that data or better unpacking that data that's important. Then we also have, I would broadly, categorise, and we say these kinds of things a lot where there's broad categorizations and then there's also much more detailed ones. But broadly, you can think of this as public data, and then departments of corrections data. So the public data is what's available anyway - we can go out there and find without any data sharing agreement with any agency. As these are government agencies where this data is required to be public. And so you'll find researchers and universities and different organisations accessing this data and publishing it or analysing it, we do that also. But we also get data sharing agreements directly with departments of corrections, and help them unpack that as well. So there's a kind of complimentary interaction there between the two datasets.

Todd Landman  7:09 

I understand. And how do you actually reduce incarceration through data analysis? I'm perplexed by that statement you made quite early on when you were talking to us.

Amrit Dhir  7:18 

There's a couple things and I'll categorise this. My broad categories into three categories. There are leadership tools, line staff tools, and then public tools. So let me start with public tools, because I think that's more related to what we just talked about in the previous question. The public tools are ones that are available to you and me. And so there's two that you can look on our website and find right now. One is a public dashboard that we call spotlight. As of the date of this recording there are two that have been published one for North Dakota and one for Pennsylvania. I encourage everyone to go check those out. If you just Google, you know our name Recidiviz and Pennsylvania, you'll see it come up as the first result. And there you can see that all the data in a accessible way. So the 'viz' in Recidiviz stands for data visualisation. We worked with the Pennsylvania Department of Corrections, to better represent the data that they have, so that the public can see it. And you can see the breakdown, by ethnicity, by district, by sex by other filters, and really get in there in some detail and see what's happened also over time. So that's one that's the public dashboard. That's largely to raise awareness. And it's something that when you talk to departments of corrections, you learn that they have lots of FOIA requests, which are Freedom of Information Act requests, so requests from media, from researchers, from the public, but also from the legislature. And so that's one thing that we do that just broadens the conversation. Another are what we call policy memos. If you go to our website, and or if you just type in, these are one-page memos that we have our data scientists put together that assess the impact of a particular administrative or legislative policy proposal. So imagine that you are looking to Pennsylvania for example, wanting to make a change to geriatric parole, or if you want it to end the criminalization of marijuana, we can then and we have gone in there and analyse the data that's publicly available. And sometimes we also access our data with collaboration with the DOC. And we can tell you what the both impact on the number of basic liberty person years that are returned. How people will get out of prison earlier or not go to prison at all, as well as how much money the state in these cases will save. And so that's a great way to inform policymakers to say hey, this is actually a good policy or a bad policy, because it's going to get people out of prison and it's going to also save you money.

Todd Landman  9:57 

Yeah the concept it's like a variable called liberty person years that you use. And then of course, it's almost like a time series interrupted model where if you get new legislation, you can look at that liberty person years before the legislation and after to judge the degree to which that legislation may or may not have made a difference, right?

Amrit Dhir  10:16 

Exactly right. And I encourage folks just to go check, check some of those memos out, there's probably like 50 on there now. And they're very easy to understand, very easy to access. They're all one page. They're all very beautifully visualised. Because you can take this very, as you said, messy and fractured datasets, but actually come to some pretty simple insights. And I would say simple and actionable. And so that's what we do there. And that was a long description of public data, but I can go into the other two, if you're ready for it.

Todd Landman  10:43 

Yes, please.

Amrit Dhir  10:44 

Okay. So working backwards, we'll go to line staff tools. And so this line staff, meaning people who are working within corrections or on supervision. And let me take the example of supervision first, because one thing that's interesting and that I actually learned only while at Recidiviz is that half of prison admissions in the US every year, come from supervision. Meaning people who are getting their parole or probation revoked and are going back to prison. That's half of the emissions we get every year. And that's a huge number.

Todd Landman  11:15 


Amrit Dhir  11:15 

And so this, you can think of this as the back end of the back end, it's the very last piece. And so for Recidiviz we were kind of assessing where we should start, that seemed like the right place to do so because the impact was just so great. Now, put yourself in the shoes of a parole officer. These folks have pretty difficult jobs in that they often have, you know, up to 100 and sometimes more, we've seen up to 120 people that they are I'll use a verb 'serving' as a parole officer. So the idea is you got people that have been returned to the community, they've been in prison, they now are trying to get jobs, they're trying to get job training. They're trying to reintegrate into their communities, and the parole officer is there to help them do that, and keep track of how they're doing. Now, that's one thing to do if you got 20 people, you want to keep track of and help and connect to the right resources, but if you've got 100, and you're supposed to meet with them every month, it becomes impractical. And that ends up meaning sometimes that parole officers aren't doing as good a job as they'd like to do. Because it's just too hard, just too much to manage.

Todd Landman  12:22 

You need a structured database approach.

Amrit Dhir  12:24 

Exactly. So that's where data can be very useful, because we can automate a lot of what a parole officer needs to do. And rather than having to check, you know, we've heard up to 12 different datasets to figure out where are the programmes my the people I'm serving are have available to them? When do I know if I need to do a home visit? Where do I find a list of employers that I can send them to? Where are housing options for them? All these are in different places, but we at Recidiviz, bring them all together, give them an easy-to-use tool, so that we can actually service them even you know, on their smartphones, in an app, to show them, hey, did you know that this person is actually eligible to be released from parole if they just upload a pay stub? And hey, do you want to just take a pay stub with your phone, and we can do it for you? I mean, how much easier that is than you having to go through all 100, figure out who's eligible based on your own recall or some other antiquated system and kind of struggle to try to help people. We can help you do that. And that's a big thing that we've done.

Todd Landman  13:22 

I mean it's almost like an E-portfolio approach that there's this way to archive parolees meeting certain milestones and conditions. And it makes the management of those cases so much more straightforward. Whilst there's also a record of that management that makes it easier for the parole officer to serve the people that they are serving.

Amrit Dhir  13:42 

Exactly. You got it exactly right. And by the way, there's, you know, a degree of nudging that can be done in this as well, if you're familiar with like the Cass Sunstein and others, behaviour psychology, but how, you know, instead of saying, hey, this person needs a drug test, and have that'd be the first thing that you prioritise. I mean you can say, hey, this person needs help finding a job. And here are some resources, here's some employers in the area that we know employ people who are formerly incarcerated. It's a great way to actually not only automate and make the life of the parole officer easier and better, but also to kind of encourage the better behaviours within those communities.

Todd Landman  14:16 

Now that makes sense. So what's the third channel then?

Amrit Dhir  14:18 

Ahhh the third one is leadership tools. And this is for the directors and their deputies, the most senior people in a department of corrections, they may come in. And actually what we're seeing now is that a lot of the people who are coming in today and are sitting in these roles are reformers. They believe that the size of our criminal justice system in the United States is just too large. And they are motivated to improve outcomes. And they're focusing on things like recidivism, which is a term for people coming back to prison after being released. And that's a number you want to have low naturally. But historically, what happens - actually not even you know what historically -what happens today. He is that these recidivism reports will come out maybe every three years. Yeah. So if you're a director, and by the time they come out, they're almost three years old. So you're almost like because the six year timelines, and you want to know, hey, I instituted this new reform this new programme, I want to know if it's been successful, you won't know until a couple years out whether it worked. And so what we do instead is to give you real time data, we can tell you what's happening on your team and in your agency on a real time basis. And also project out based on what we're seeing with some meaningful kind of population projections as well. So that helpful.

Todd Landman  14:34 

That's fascinating. And let me ask you just another technical question. So when people are released from prison, is it typical for them to also have a sort of GPS tag on their leg for a certain period of time? And does that form any of the data that you look at?

Amrit Dhir  15:52 

So it depends? It's a very good question. And it's one of the more controversial topics today in this space, and especially in the Reform Movement, there's a concern that we may be heading towards, from mass incarceration to mass incarceration, and that people will be monitored and supervised within their communities. And I think that is a very meaningful concern that we need to be careful of, because we don't want that to happen. But to broadly answer your question about the state of this today, it depends on where you are, it depends on the county depends on the state depends on all those things, in terms of whether you are wearing a device that electronically monitors, you know, we don't track that ourselves, that's something that we do or want to do. Our approach is to helping people get off of supervision and get into programmes and other kinds of initiatives that help them on their way.

Todd Landman  16:43 

Excellent. So this discussion really opened up into, you know, the bad side of the question, I guess, you know, you just have to go into this with our eyes open, I suspect that you're triangulating a lot of data. You're providing that in real time on dashboards, a lot of it's in the public domain. What are the risks around this? What are the pitfalls? What's the risk of re-identification? What's the risk of, you know, lapsing into kind of credit scoring philosophies? And just, as you said about the tags, there's worry about that kind of, you know, E-surveillance and E- carceration. Equally, someone could backward engineer some of your data and actually profile people. So, what's the downside of this approach?

Amrit Dhir  17:21 

Yeah, that was a great list. So there's certainly a concern of bias entering any analysis of a dataset. And we are very careful about that. So one thing to note is that everything that we do is open source. So it's open to the technology community to take a look at what's kind of under the hood. And that's important, because we would do want to make sure that we are not only participating and contributing to the broader ecosystem that are, in this case, tech and criminal justice ecosystem, but that we're also held accountable to them. So that's the first thing that we do, we also are very mindful and transparent about our data ethics policies, and how we handle those kinds of questions and sometimes ambiguities. So if you look at, for example, the spotlight dashboard that I mentioned that you'd find for Pennsylvania, North Dakota, you will see in the methodology that we explain what happens when there's a question. So for example, if someone puts down three different ethnicities, how do we manage that in the data visualisation that just shows them as one. Our approach there is transparency and engagement.

Todd Landman  19:31 

 Have you done any links with the ACLU on this? Because they're quite interested in prison conditions. They're interested in incarceration, sentencing, etc. Do you do any kind of briefing with the ACLU?

Amrit Dhir  20:16 

Yeah, so two things actually on that. I will take them in reverse order. So first of all, we do work with ACLU. If you look at our website, on the policy page, which again, are those one page memos, the ACLU has requested a number of those. And there's naturally different chapters of the ACLU in different states in different parts of the country. And we work with different stakeholders within the ACLU as well on those. The other piece, though, get one of the back to what you said about three strikes. There's another piece of that I think people may not be as familiar with it. I certainly wasn't, which is this issue of technical revocations. So if you're on supervision, like I said, half of prison emissions every year are from revocation of your supervision, meaning you're going back to prison, from parole or probation. But half of those, so a quarter of all emissions every year are from technical revocations. And those are when someone breaks a rule, that is not a law for the rest of us. Right. So it's not that they stole something, it's not that they got caught breaking a law, that they broke a rule of their parole, and sometimes these are ones that you and I would feel horrified to learn of. So that, you know, we've got examples of people, for example, going to an open mic night where there was alcohol presence, and that person wasn't allowed to be around alcohol. Being in the wrong County. Being out past curfew. All of these things that are, and you know there are anecdotes all over the place of the kinds of things that send people back to prison that we as society would not tolerate. And those are also some of what we're reducing.

Todd Landman  21:49 

That's amazing, that sort of distinction to draw between, you know, breaking a rule and breaking the actual law, I guess the rules follow from the law. But I get your point in terms of, you know, how would somebody know if they crossed the county line, particularly if they're at an area they don't know well. So this has been a fascinating exploration with the ways in which you have triangulated datasets, made them more visible, put them into real time, and I have to reflect on what you said. I mean, I grew up in Harrisburg, Pennsylvania. So I'm going to immediately read all your Pennsylvania data. I actually grew up near a prison in Camp Hill, Pennsylvania, you know, so it'd be interesting to see how things have moved on from the time that I lived there many moons ago. I won't tell you how long ago that was. But, but this is a really good conversation for us to have around some of the ways in which different types of data can be leveraged for good. And also some of the challenges of that, or the misuse of that information, as well as the sort of things that you don't collect, you know, the fact that you don't collect data on these tags. And that that varies, of course, and the variation you see in terms of the population that you're collecting data on varies because of the fragmentation of the US prison system and the sort of federal system that the US is structured in, but also data that no one really brought together in one place before. And I think that when we hear this data for good argument, we hear a lot of people saying we're actually bringing datasets that haven't been brought together before in order to derive insights from those data and do something that is for good and brings about positive social changes result. So I just think this tour that you've given us today is absolutely fantastic. And on behalf of the Rights Track thanks so much for being on this episode with us today.

Amrit Dhir  23:25 

Oh, thank you for having me. It's been fun. Thank-you.

Christine Garrington  23:29 

Thanks for listening to this episode of The Rights Track, which was presented by Todd Landman and produced by Chris Garrington of Research Podcasts with funding from 3DI. You can find a detailed show notes on the website at And don't forget to subscribe wherever you listen to your podcasts to access future and earlier episodes.