Artificial Intelligence? - part the first

It's tough to make predictions, especially about the future.
-- Yogi Berra

Well, Zillow has stepped in it, up to their neck. These endeavors have touched on the folly of AI here and there, but this story is way too juicy to let slide.

Zillow CEO and cofounder Rich Barton explained the shuttering of Zillow Offers by citing "unpredictability in forecasting home prices" that "far exceeds" what the company had expected.
...
Zillow declined a request for an interview with Krishna Rao, the company's vice president of analytics.

Well, no shit Superman!

The essence of the problem, at least from the point of view of longstanding stat practice (first told by Dr. McElhone, IIRC): "you don't make predictions beyond the range of the data". Some stats do that as a matter of course; generally when the client demands an answer for the demand for widgets five years hence, and the like. Any stat who intends to stay in the business will CYA bigly with "these estimates are likely if the historic trends continues for five years", and such.

If The Great Recession taught us all anything (and, apparently, no one in the suites listened), it is that history doesn't continue smoothly into the future. Data results from events, not the other way round. This sort of AI can be made to work, more or less, in the natural world where the rules of engagement (aka, the laws of physics) remain static in the local space in the near future. IOW, you can predict climate pretty accurately until a black swan, like Mt. Tambora going nucular in 1815 messes up the 'model'.

Summer temperatures in Europe were the coldest on record between the years of 1766-2000. This resulted in major food shortages across the Northern Hemisphere.

When it comes to events driven by human decision making (notoriously volatile), betting on AI to supplant the human brain reading the NYT is a losing bet.