Outsourcing judgement, the AI edition

With every new technology, people try to do two things with it: communicate with others and rate people1. AI is no exception and HR and communications professionals should expect it to show up in two places: social media analysis and candidate screening.

Over the past 13 years, I’ve become an expert in many different ways to rate people, from the academic citation analysis tools on which universities spend millions to dating apps, and I’ve used a number of tools to monitor social media. The tools are dangerous to your business if you don’t know what you’re doing. You absolutely cannot assume social media is an accurate reflection of actual customer or consumer sentiment2. Social media monitoring tools will show you thousands of mentions from accounts with names like zendaya4eva and cooldude42 and the tools roll everything up into pretty dashboards that summarize the overall sentiment for you. There’s just one problem, and it’s that social media sentiment analysis sucks. Posts aren’t long enough for the algorithms to get enough signal and they can’t detect sarcasm or irony. You’re better off just looking at a sample of posts than using a sentiment dashboard. Analytics vendors know this and they’re working on building AI into the tools to make this better, but if you’re looking at social media sentiment because it’s easier to get than data on actual customers, you’re like the proverbial drunkard, looking for your keys where the light is better rather than where you actually lost them, and no amount of AI can fix that.

Candidate screening tools make some of the same promises. We can analyze the social media history of a candidate and flag areas of concern! I’ve written social media policies3 for several organizations and never have I ever seen a hiring or firing decision depend on a social media post that required a tool to flag. It’s very tempting to outsource our judgment. Thinking is hard and people aren’t always very good at it. You might think it’s better to have an objective process that eliminates conscious or unconscious bias4, but when you do this, you’re taking agency out of the hands of HR and the hiring manager. Hiring is a hard, multi-factorial decision and the last thing you want to do is outsource judgment here5 .

Amazing Science – Mind Control Edition

Sometimes an article comes out with a title that makes me think, “Wait a second. It’s not even April 1st yet.”

Joe Tsien’s NR2B overexpression experiment and John Chapin’s “Rats control robots with minds” were pretty amazing articles, and now Dalle Molle Institute for Perceptual Artificial Intelligence comes out with a technique that can detect “whether you are thinking about a calculation, a place, a colour or even what you want to eat for dinner…but it’s not good enough yet to detect exactly what colour you’re thinking of.” I believe they’re using Bayesian analysis, a great statistical learning technique which I’ve seen being used more and more often, to look for “EEG patterns embedded in the continuous EEG signal associated with different mental states.” Here’s a summary .pdf describing the technique.