Making intelligence smart
Not too long ago, the phrase “building out systems to oversee government artificial intelligence uses” would have seemed plucked from the pages of a cyberpunk novel. These days, the tools are not only very real, but it’s perfectly reasonable to expect our leaders to develop policies to ensure that the AI that we are increasingly delegating to works as intended.
Most people still think of an AI as a fully sentient, HAL 9000-style machine that can just as easily play chess as converse about philosophy. That’s close but remains in the realm of science fiction. Yet for years now, much more specialized and quotidian AI tools have been in use by researchers, corporations and governments, often in ways that help speed functions for the better, but sometimes to disastrous effect.
In one infamous example, machine learning algorithms used to help make bail and sentencing decisions were found to have imported the racial biases embedded in the criminal justice datasets it was fed, giving a veneer of scientific infallibility to a process that was fundamentally racist. It is just this sort of AI-gone-wrong circumstance that state Comptroller Tom DiNapoli is hoping to prevent in calling on NYC agencies to build out rules and auditing procedures for the many AI systems they lean on.
In an audit published last week, his office looked at the use of AI by the Administration for Children’s Services, the Department of Buildings, the Department of Education and the NYPD. While thankfully no horror stories emerged, the report found inconsistent practices and poor record-keeping that could hamper the ability to even examine the systems for inefficiency, poor performance or bias.
Used correctly, AI can make government more nimble and improve outcomes for everyone, but just like we don’t let officials make decisions without oversight or performance metrics, and we shouldn’t let machines do so either, especially when they’re helping make crucial decisions. It doesn’t have to be cumbersome; agencies can adopt general guidelines and audit procedures for all their AI tools. Now that’s intelligent.