Earlier this month, San Francisco banned city agencies from using facial recognition technology in response to a wave of police departments and other agencies in cities around the country adopting a technology that has been shown to be inaccurate and biased. And it’s not just facial recognition: Governmental algorithms that make life-altering decisions about criminal sentencing, homeless housing, and other safety net benefits are on the rise–and often rely upon bad, biased data.
Unfortunately, there are plenty of examples of state governments rushing to implement shady algorithms that hardly work. But there are also several emerging examples of states using automation for good. The San Francisco-based nonprofit Code for America has piloted a project in California that will automatically expunge the criminal records of people whose records are now inaccurate because of changing laws, particularly when it comes to marijuana. The organization is also on the cusp of rolling out its technology to simplify applying for food stamps to the entire state of California. Both examples show the positive impact that simpler kinds of technology can have on government–when it’s utilized with the goal of making it easier for people to access services.
The record expungement project began in 2016 with the goal of making the current process to clear a criminal record more streamlined and easy to understand. The tool the nonprofit built, called Clear My Record, reduced a burdensome application down to a digital form that took only 10 minutes to fill out and then connects the user with an attorney who could oversee their case. The approach has helped 10,000 people in California to leave their criminal records behind.
For Jennifer Pahlka, executive director of Code for America, this was a great step forward. But Pahlka wanted to find a way to help the millions who still didn’t have the relief they needed, especially given the state’s legalization of marijuana, which granted almost anyone who had been convicted of a marijuana crime the chance to clear their record or have their charges reduced. “We knew if we were going to fulfill the intent of the laws, we would have to do something different,” she says.
Rather than requiring each person to submit a petition to have their record cleared, Pahlka and her team worked with the San Francisco district attorney to flip the paradigm: The government would proactively search through its records for people who should have records removed or lessened. To make that happen, they needed an algorithm, one that didn’t even require machine learning (even though it took skilled programmers to implement). “It’s not that different from downloading a bunch of data into an Excel file and using search-and-replace,” Pahlka says.
It was this pilot program in San Francisco, which helped the government identify 9,000 convictions to wipe away, that encouraged California governor Jerry Brown to sign a bill in fall 2018 that would require all counties in the state to proactively start changing people’s records. According to Pahlka, the governor originally wasn’t going to sign the bill because of its cost, which assumed that the government would have to pay a small army of paralegals to pull down rap sheets and expunge them one by one from the state database. “That sounds like an impossible Sisyphean task,” Pahlka says. “They wouldn’t have signed the bill if we weren’t able to show…there’s a role for technology to dramatically reduce the cost.”
Now five counties in California are using Code for America’s automated system–which the organization has adjusted to work for each county’s needs–to clean up their databases and help people whose criminal records have kept them from getting jobs and finding housing. Later this summer, the nonprofit plans to publish an open-source guide to its technology so that any county anywhere in the country can use this automated tool to help people leave their past in the past.
The project points to the vital role of technology in today’s government, but crucially, this algorithm isn’t particularly complex and doesn’t require any cutting-edge (or untested) techniques to function.
“We are a much bigger fan of government getting competent at pretty non-fancy technology than we are of skipping over that level and diving headlong into things like the blockchain or advanced use of AI,” Pahlka says. “If you’re not good at basic technology, you can make mistakes using basic technology. Let’s get good at that first.”
Pahlka also wants to see governments focusing more on service design–the idea that agencies are designing services for real people, and that the services should work for and fit into the lives of these users.
Another of Code for America’s projects, called GetCalFresh, aims to streamline and simplify the onerous application process of receiving food stamps in California. The lengthy, difficult application process has real-world impact: The state historically has had one of the lowest participation rates in the country, and Code for America estimates that 10 million people in California are eligible for the program but don’t take advantage of it.
GetCalFresh, which first piloted in 2014, revamps the old application–that took nearly an hour to complete–into a mobile-friendly webpage that takes just eight minutes to finish. It’s been a runaway success: With 37 out of the state’s 58 counties already using GetCalFresh, more than 390,000 people have successfully used the program to apply for benefits, and there are about 20,000 applications every month. And on June 1, the state of California will mandate that every county use the program.
But the team was still seeing a worrying gap between the number of applications and the number of eligible people who were actually receiving food stamps benefits. So they began to examine other parts of the process, namely the interview and the uploading of verification documents, which were designed around the government’s schedule, making it difficult for users to make it all the way through to receive their benefit. For the interview, applicants receive a letter in the mail with their interview time, which may not work for their job’s hours. If they miss the interview, it’s difficult to reschedule.
That research led to a new feature in GetCalFresh that sends the applicant a text message asking them to call the state’s office whenever they have time so they can do their interview over the phone during a break from work. Other improvements include a text message reminder for the interview, and the ability to take a photo of income verification documents–usually a pay stub–and upload it to GetCalFresh directly from a phone rather than having to scan it or deliver it in person.
In June, California is extending food stamp assistance to anyone who qualifies for Supplemental Social Income, a benefit for low-income people who are 65 or older or have disabilities. That means an additional 360,000 people will be eligible for food stamps. According to Kim McCoy Wade, the director of the CalFresh program at California’s Department of Social Services, the applications are already pouring in–when we spoke earlier this month, the department was already receiving record applications through GetCalFresh.
For Wade, working with Code for America has meant rethinking the way in which technology works within the government. Usually, agencies go through a lengthy procurement process to contract with an outside vendor to build a piece of technology. But that often doesn’t include the idea of iterating on a product–or tweaking it at all, even if it’s not working as well as it could. “We need to assume this is going to change all the time and not set up a change process that is so burdensome,” Wade says. “How do you build something planning for it to change in a matter of days or weeks?”
The challenge of enabling iteration within government services is something that’s top of mind for Wade right now, because the state has commissioned a new version of the paper application process even as the digital version rolls out across every county. “That’s been a learning process: to see how that can help you be more responsive without compromising government rules,” Wade says. “There’s space within the rules to still be responsive to your customers’ needs.”
It’s this kind of mindset that Code for America has been working to instill in its government partners.
“I think in some ways it’s the design mind that has the most impact in government,” Pahlka says. “If you boil all this down, it is an agenda that says government is supposed to work for people. If it’s supposed to work for people, then its core capability and competence should be in human-centered design.”
Ultimately, Pahlka wants to ingrain this service-design focus, as well as the technical ability to carry it out, within government itself–so that agencies don’t need an external organization like Code for America to come in to make these kinds of programs happen.
“Government needs to have a baseline competence with this stuff: an ability to interrogate the algorithm, to understand how it’s working, to change it if it’s wrong,” Pahlka says. She points to a growing narrative that government is simply incompetent at technology, and that technology must be outsourced to vendors. But she hopes to counter that by demystifying what technology means and making it more accessible, like it’s “something an administrative person would do, versus something a super-fancy AI programmer would do.”
When technology is like any other tool and not an untouchable black box, that’s when people can use it more effectively. “Bureaucrats can be designers. I have seen it many times,” Pahlka says. “If you give them the freedom and a few of the tools, they actually very much want to do this. They will hack.”