GIST: "Writing about AI for an appreciable amount of time is, in my experience, enough to make any reasonable person concerned about the future of humanity. But I worry the focus of that concern is too often directed at the relatively distant future, which could lead to unforeseen consequences in the present. Headlines from the past few months illuminate how bad things can get. Consider the cases of the self-driving Uber that killed Elaine Herzberg in Tempe, Arizona and that of the Apple engineer who was killed when his Tesla, driving on Autopilot, plowed into a traffic barrier on the highway. You’re probably aware of the content suggestion algorithms from Facebook and YouTube, which have been implicated in the spread of fake news and extremist views. Then there was a story last week about how companies and cities use Palantir’s analytics for corporate security and predictive policing, with potentially disastrous results. One man interviewed in the story claimed that he isn’t involved with the Eastside 18 gang in Los Angeles but said the LAPD has him in their database as an associate and officers have been stopping him as a result. None of these cases involved AI that’s advanced enough for Elon Musk to call it an existential threat to humanity. But that didn’t stop people from getting hurt. If we’re not careful, this will happen more frequently in the coming years. It’s easy to worry about a catastrophic future for bank tellers, truck drivers, or some other profession that’s being told today their jobs will go away in the future. Thanks to Westworld, Battlestar Galactica, Iain M. Banks’ Culture novels, and other media, we can confidently picture how artificial superintelligence could upend our lives. Don’t get me wrong — that’s all very concerning. But we can’t let our anxiety about the distant future blind us to what’s going on right in front of us. When AI systems fail, they can do so in ways we don’t expect. Research and anecdotal evidence have shown us that those systems are often biased against minorities. That bad news turns dire when those systems become critical components of our infrastructure. Consider the story of the Southern State Parkway in New York that Robert Caro laid out in The Power Broker. According to Caro’s interviews with Sidney M. Shapiro, urban developer Robert Moses decided to build overpasses for the parkway that were too short for buses to pass under, which would limit the access people of color had to it. Those overpasses still stand. It’s a story that worries me about the future, considering that we’re using AI to build far more powerful systems that can make predictive decisions about all manner of things. This isn’t an idle concern: Law enforcement agencies are using AWS’ Recognition service to do image recognition today as part of their work. A recent study of competing facial recognition APIs showed they were less accurate identifying the gender of people with darker skin, especially women. There’s good news: Changing an algorithm is far easier than building a new train line or adjusting the height of a bridge. Our codification of bias need not be set in stone. But we’re already developing technologies like algorithms that claim to provide predictive policing capabilities. Companies are already testing the use of AI for investing, money lending, and other tasks. We need to be working on this now. In my view, there are a few things we can do. First, I’d like to see a non-industry body develop a system of auditing algorithmic bias, which would allow governments and businesses to restrict the use of algorithms that don’t meet a particular bias metric. With that in place, I’d like to see governments identify situations where the biases of those algorithms must be made available to the public. Law enforcement, consumer finance, and health care seem like good candidates from the get-go. Second, I believe governments and corporations should be required to disclose when an algorithm influenced the outcome of a decision. For example, if police arrest someone based in part on the evaluation of a predictive policing system, the suspect should have the right to know that.
I don’t have high hopes any of that will happen, given the current political climate. But a boy can dream. It’s the only way I’ll be able to sleep at night, anyhow."

The entire commentary can be found at: