State and local governments are widely using algorithms to automate decisions. These tools are powerful, but can easily backfire. Their applications are growing fast, which means now is the time for Philadelphia and Pennsylvania lawmakers to start tracking and studying their use.
We need to ensure algorithms are effective and equitable, so we can harness the efficiency of automation without abusing the rights of its citizens.
What’s the big deal? You may realize we interact with algorithms whenever we’re online. These software routines decide whose profiles we see on dating apps, filter spam from our inboxes, and tag and sort our photos. More subtly, algorithms also shape our offline experiences: they streamline bus routes, provide insurance quotes, set bail, evaluate students, decide credit card limits, and help the government apportion public funds.
The humans behind these computer programs are usually well-intentioned, creating them to help government agencies make faster decisions or lower companies’ bottom line — ideally so savings can be passed on to taxpayers and customers.
But rampant use of AI is creating problems that go unseen, obscured by complex statistics until it’s too late.
Biases creep into algorithms easily. After all, most engineers who design them aren’t paid for perfection; they’re paid for cheap and efficient solutions. And after algorithms are deployed, it can be almost impossible to detect bias. Few systems build in audits of the decisions — “A computer did it, how could it be unfair?” — and subtle trends that indicate unfairness are buried in heaps of data that can be spread across an organization and are often kept private for proprietary reasons.
So algorithms continue on, cementing into computer code the racial and gender biases that lurk beneath the surface of society. They are automating harmful and discriminatory practices, which will continue to grow if left unchecked.
Here are a few examples of how Pennsylvania uses algorithms to make high-risk, sensitive decisions about citizens:
- In Philadelphia, algorithms assign “risk scores” to defendants that can impact their bail decision or probation terms.
- In Pittsburgh, algorithms help set bail and choose the highest-priority targets for child abuse interventions.
- School systems use algorithms to evaluate students and teachers, such as the statewide Pennsylvania Value Added Assessment System.
These are all areas in which it’s important to get decisions right. As such, they present an opportunity to lead by example.
The Pa. General Assembly and Philadelphia City Council can both take action now: they can establish a task force to keep track of state and local use of algorithms. This low-cost first step will empower Pennsylvania to develop a smart regulatory approach to automation, allowing government agencies and the public to reap the benefits of this new technology while minimizing the risks to residents.
Already in Pa., algorithms are being applied to important decisions without proper safeguards. This is true when government agencies or researchers design algorithms, but is even more evident when private companies are contracted to build and run them, a common practice. The system used to evaluate students and teachers in Pennsylvania public schools, for example, is contracted from a company called SAS Institute Inc. Researchers recently found that this and other similar tools are racially biased, perhaps even more than other tools in the market.
Intentional policymaking around algorithm use is so important because bias and errors in algorithms tend to sneak through, with disastrous consequences.
Even small biases present in an algorithm can have huge effects when scaled up to make thousands of decisions about people’s lives every second. Their consistency and the scale with which they’re applied means harms can easily balloon, and even become systemic, erecting more barriers for communities already marginalized.
How can we rein in the damages that algorithmic decision systems cause? Start by examining and learning more about the systems we’re already using.
Lawmakers in Harrisburg could pass bills modeled after similar measures proposed in the New York and Washington state legislatures. Councilmembers in Philly can build on the work of a similar task force in New York City. These task forces would test for bias, find best practices, and publish reports to inform the public and suggest future legislation. This would add much-needed oversight, increase transparency, and facilitate research to continue to make algorithms more fair.
Leaders need to take action now to establish proper oversight and ensure equitable treatment for all residents.