There's been a lot written about AI in recent years, but it's rare to find an article that explains the basics in non-technical language, without dumbing down the concepts. It's definitely worth the time to read this article by Yonatan Zunger: Asking the Right Questions About AI. It explains the processes used to build AI systems, and how the technology and — most importantly — the data used to build them can make AI-enabled applications and devices do unexpected or ethically dubious things.
For example: why image searches in 2016 for "three black teens" and "three white teens" produces the following results:
The bias displayed here comes with data:
What happened here wasn’t a bias in Google’s algorithms: it was a bias in the underlying data. This particular bias was a combination of “invisible whiteness” and media bias in reporting: if three white teenagers are arrested for a crime, not only are news media much less likely to show their mug shots, but they’re less likely to refer to them as “white teenagers.” In fact, nearly the only time groups of teenagers were explicitly labeled as being “white” was in stock photography catalogues. But if three black teenagers are arrested, you can count on that phrase showing up a lot in the press coverage.
Read the full article at the link below. There's an audio version available too.
Medium: Asking the Right Questions About AI (via Thomas Lumley)