iiinova — where your digital self keeps evolving

Alex Chohlas-Wood @alexchohlaswood.com · 19/04/2026

This week at @brookings.edu, me, @chiraagbains.bsky.social, and Katie Kinsey argue for why states should take the lead on the regulation of AI in criminal justice settings, a high-stakes setting where effective regulation could not be more important.

121

Alex Chohlas-Wood @alexchohlaswood.com · 19/04/2026

That's why it's critical to allow all fifty states the flexibility to choose their own regulatory path — so that we can scale those that work, and move on from those that don't.

110

Alex Chohlas-Wood @alexchohlaswood.com · 19/04/2026

In an age of rapidly evolving AI, we are *just* starting to learn how we can properly regulate these powerful new technologies. While many new regulations will be well-intended, we won't know what really works until we try many different approaches.

130

Alex Chohlas-Wood @alexchohlaswood.com · 01/04/2026

.@counciloncj.org just released a comprehensive guide to help justice agencies make the right decision about whether and how to adopt AI. This guide reflects many productive discussions we've had on the Task Force for AI over the last year—kudos to CCJ for distilling this into a practical guide!

010

Reposted by Alex Chohlas-Wood

Council on Criminal Justice @counciloncj.org · 04/03/2026

How can AI improve the New York Police Department and the justice system as a whole? A perspective from CCJ Task Force on AI member @alexchohlaswood.com in @vitalcitynyc.bsky.social explores the potential opportunities, and risks, of the new technology: www.vitalcitynyc.org/how-to-deplo...

vitalcitynyc.org

How To Deploy AI To Improve Policing in New York

Powerful new technology has the potential to improve the City’s justice system, but to get it right, the NYPD has to address serious risks.

022

Alex Chohlas-Wood @alexchohlaswood.com · 23/02/2026

Police departments nationwide are racing to adopt AI. Yet the NYPD hasn't permanently adopted a new AI tool since 2016.* That stagnation is a choice. And it's NYC's most vulnerable residents who pay the price. In @vitalcitynyc.bsky.social this week, I lay out a better path forward. (Link below)

A 911 call dispatcher working in front of five open screens. CC By 2.0. Source: https://www.flickr.com/photos/143513894@N04/27188584211

131

Alex Chohlas-Wood @alexchohlaswood.com · 30/10/2025

Thanks Nikhil! I have a copy if it’s taken down!

020

Alex Chohlas-Wood @alexchohlaswood.com · 30/10/2025

Learn more about how you can use 911 data to understand police deployment in our short blog post for ASA's Committee on Law and Justice Statistics: community.amstat.org/lawandjustic...

community.amstat.org

ASA Community

The ASA Community is an online gateway for member collaboration and connection.

011

Alex Chohlas-Wood @alexchohlaswood.com · 30/10/2025

In a new blog post for the @amstatnews.bsky.social, John Hall and I make novel use of the city's 911 data to show that overnight train patrols more than doubled after the city announced its new policy in January.

A time series of overnight subway patrols between September 2019–June 2025. Subway patrols more than doubled after the NYPD's policy announcement, from about 7,500 train patrol calls per month before the announcement to over 15,000 calls per month after the announcement.

120

Alex Chohlas-Wood @alexchohlaswood.com · 30/10/2025

But it turns out the information we need is already public, in the city's "Calls for Service"—a.k.a. 911—dataset!

A screenshot of NYC's Open Data portal listing NYPD's "Calls for Service (Historic)" dataset, with nearly 50 million rows.

100

Alex Chohlas-Wood @alexchohlaswood.com · 30/10/2025

You might think that you'd need detailed police deployment records to answer this question. Getting this information via a public records request could take months, and may be denied by the department altogether.

100

Alex Chohlas-Wood @alexchohlaswood.com · 30/10/2025

In January, the NYPD said it would put two officers on every late-night subway car in New York City. How can we know whether the NYPD kept its promise?

Police officers face away from the camera toward a New York City subway train. The photo is taken through a fare gate. Photo credit: Metropolitan Transit Authority via CC 2.0.

230

Reposted by Alex Chohlas-Wood

Computational Policy Lab at Harvard Kennedy School @comppolicylab.bsky.social · 20/10/2025

The Computational Policy Lab is growing! We’re hiring software engineers to build technology that makes government, education, and social programs more fair, effective, and data-driven. Know someone who'd be a great fit? Learn more: careers.harvard.edu/job/software...

careers.harvard.edu

Software Engineer

001

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

Learn more about our study in this thread from two years ago: bsky.app/profile/alex...

100

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

Reminders alone are unlikely to dramatically reduce overall jail populations, as jail stays for missed court dates are often short. But stacking them with other small, common-sense reforms could substantially improve our justice system.

100

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

Reminders also help everyone by saving precious public resources instead of paying for these wasteful jail stays. The reminders themselves cost about 60¢ per case—less than the costs of paying for someone’s arrest and incarceration, even for a single night.

100

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

We found that a broad swath of clients appeared to benefit from reminders—even people facing low-level charges. Think of a DUI case: make a bad mistake one night, then forget a court date, and suddenly you’re in jail for a few days. A minor case just became much more serious.

100

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

Court date reminders may seem small, but they make a big difference. Our study was conducted with public defender clients who can’t afford a lawyer. These nudges are a huge boon for low-income clients, helping them avoid the high costs of a disruptive stay in jail.

100

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

Our study has now been peer reviewed—and it's officially out today in Science Advances! It’s open access, which means it’s free for anyone to read. www.science.org/doi/epdf/10....

science.org

Automated reminders reduce incarceration for missed court dates: Evidence from a text message experiment

You have to enable JavaScript in your browser's settings in order to use the eReader.

120

Alex Chohlas-Wood @alexchohlaswood.com · 01/10/2025

Have you ever forgotten an important date—like a birthday for a loved one? Now imagine if forgetting meant ending up in jail. Two years ago, we ran a randomized experiment that found that text message reminders reduce jail stays for missed court dates by over 20%.

A hand holding a smartphone displaying a court date reminder on screen.

1102

Alex Chohlas-Wood @alexchohlaswood.com · 03/09/2025

The ASA's Law & Justice Statistics committee is hosting an upcoming webinar featuring George Mohler. Join us on September 30 from 1–1:30pm ET! Register here: amstat.zoom.us/webinar/regi...

010

Alex Chohlas-Wood @alexchohlaswood.com · 04/08/2025

Calling all professors—we're enrolling new courses in our ongoing randomized study of AI in education! Get free access to a customized virtual tutor, and receive an honorarium at the end of the semester if you participate. Details below.

010

Alex Chohlas-Wood @alexchohlaswood.com · 29/07/2025

Come join us and teach a great class this fall!

020

Alex Chohlas-Wood @alexchohlaswood.com · 14/02/2025

(And I'll be teaching a course like this next spring at NYU so I'd love to hear what else you find!)

110

Alex Chohlas-Wood @alexchohlaswood.com · 14/02/2025

And another from Jochen Hartmann here: cms.mgt.tum.de/fileadmin/mg...

cms.mgt.tum.de

110

Alex Chohlas-Wood @alexchohlaswood.com · 23/01/2025

Job alert! We're hiring a clinical (teaching-based) Assistant Professor of Applied Statistics for Social Science Research at NYU! Application review begins on February 10, and the position would start on September 1. Apply here: apply.interfolio.com/162021

Two students collaborating on a plot on a whiteboard.

041

Reposted by Alex Chohlas-Wood

Hannah Li @hannahli.bsky.social · 09/01/2025

Yes to evaluating the *outcomes* of these systems rather than as a standalone algorithm! This is something that's been bothering me for a while about ML assisted decisions

001

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

Ultimately, we’ll likely achieve better outcomes if we think of algorithms as *policies* — and design them in a way that aims for the specific policy goals we desire. (17/17) bsky.app/profile/alex...

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

Learn more in our open-access paper, “Learning to be Fair: A Consequentialist Approach to Equitable Decision Making”, with @madisoncoots.com, Henry Zhu, Emma Brunskill, and @5harad.com! pubsonline.informs.org/doi/10.1287/... (16/)

pubsonline.informs.org

Learning to Be Fair: A Consequentialist Approach to Equitable Decision Making | Management Science

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

Many studies have framed fairness as a mathematical problem, proposing axioms without considering the consequences. In contrast, our approach: - Focuses on outcomes - Devises a computational framework for learning to be fair in an efficient and cost-effective manner (15/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

We use data from the Santa Clara County Public Defender to show that this approach would result in higher utility: - During the learning phase AND - After we stop learning! Of course, this approach applies in any resource-constrained setting, not just for rides to court! (14/)

A chart of regret vs. iteration. Random assignment methods accumulate a lot of regret over the course of an experiment, whereas bandit methods like UCB and Thompson sampling accumulate minimal regret.

A chart of performance vs. iteration. Bandit methods like UCB and Thompson sampling are quick to learn good policies, while other methods learn good policies more slowly.

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

The framework we designed uses contextual bandits and optimization to: - Learn how people respond to rides, and then provide rides to people who need them—even while we’re still learning - Equitably allocate rides by modeling preferences as parameters in a convex objective (13/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

But randomized controlled trials are costly in a couple ways. First, people who would really benefit from a ride might be excluded if they’re randomized to a control arm. Second, we might waste money on rides for people who don’t need transportation assistance. (12/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

How could we make decisions like this in the real world? One approach would be to run a randomized controlled trial to learn how people respond to rides. We could then estimate the tradeoffs at hand, and choose an tradeoff that best reflects our preferences. (11/)

A Pareto curve, showing the tradeoff between helping people get to court and the proportion of people in a target group who get the benefit of rides. The chart also shows a slider bar indicating a survey response, with the selection corresponding to the point of maximum utility.

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

This suggests that there’s no one-size-fits-all definition of fairness. Instead, we should make decisions in a way that reflects our preference for how to make difficult tradeoffs. (In practice, one could run a survey like the above to elicit preferences from people.) (10/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

To illustrate, we asked 300 Americans how they would make this tradeoff. After explaining the problem, we let them choose their preferred outcome. Most people preferred an outcome other than demographic parity—even people in the same political party! (9/)

The graphic we showed survey respondents. There are five options presented, with various amounts of average spending per Black and White person. Each option also has an associated number of people who avoid jail, with the highest number associated with more spending per White person than Black person.

A bar chart of survey responses, split by party affiliation.

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

Fortunately, we don’t have to follow only one of these approaches. We could instead balance between these approaches in a way that feels most fair. But “what feels fair” is ultimately a matter of personal preference, and depends on the exact tradeoff in question. (8/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

This would reduce disparities in who gets a ride. But there would be real drawbacks! - We’d pay for longer rides to court, so - We’d provide fewer rides overall, so - More people would go to jail for missing court. In other words, there’s an inherent tradeoff at play. (7/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

It might seem wrong to exclude most Black and Hispanic residents like this. So what should we do? We could instead provide a ride to an equal share of court attendees from every neighborhood. (For researchers in ML fairness, this is akin to imposing demographic parity.) (6/)

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

But there’s a catch! Take Boston as an example. In prioritizing cheap + short rides, imagine that we drew from people who lived close to the courthouse (like in the map). By trying to be efficient with our budget, we’d exclude many Black + Hispanic residents of Boston. (5/)

A map of the Boston area, showing residents in different colors by their race/ethnicity. A star is in the center of the map surrounded by a circle; most of the people in the circle are white residents, with some Asian and some Hispanic residents as well. A large population of Black and Hispanic residents sits south of the city center and is outside of the circle.

100

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

So we’d have to choose who gets a ride. A natural way to do this would be to select people who: a) Wouldn’t attend normally, but would attend if provided a ride b) Live close to court, so the ride is cheap This would maximize court appearances given our budget. Seems good! (4/)

110

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

What if we wanted to improve court attendance rates even more? One possible initiative would be providing people with free rides to court. But with a limited budget, we wouldn’t have enough funding to give everyone a ride. (3/)

An image of two women appearing to shake hands outside of a car.

110

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

Imagine a government initiative that aims to improve court attendance rates. In other research, we’ve already shown that sending automated court date reminders can increase court appearance rates and reduce pretrial incarceration. (2/) bsky.app/profile/alex...

110

Alex Chohlas-Wood @alexchohlaswood.com · 08/01/2025

NEW in Management Science! My coauthors and I came up with a new consequentialist approach to designing equitable algorithms. Instead of imposing fairness criteria on an algorithm (like equal false negative rates), we aim for good outcomes. More in the 🧵 below! (1/)

A screenshot of the first page of our paper, Learning to Be Fair, showing the title and abstract.

1167