Wolf has fun with Zillow
Wolf is having fun with Zillow's stupid AI-based failure. They bought 35k houses at the peak of a wildly distorted bubble, and now they're trying to unload 7k of the houses at a serious loss.
Zillow decided to go into the house-flipping business by remote control. They trusted their AI to tell them what each house should bring, and then engaged flipping crews by remote control.
There are plenty of lessons in this failure. Some of them are so universally obvious they don't even count as lessons. You're supposed to buy low and sell high, not buy high and sell low. Remote control doesn't work.
The computer stuff is obvious to programmers.
Silicon Valley arrogance trusts software more than ANYTHING else.
If you have software you don't need knowledge or common sense.
AI, which is just a very big abacus, is supposed to acquire magical powers from its size. Programmers know that the abacus can only do what you order it to do. Zillow made a HUMAN decision to buy high, and then programmed their abacus to justify their HUMAN decision. A simulation or model can be pushed and pulled in any desired direction by choosing the assumptions and setting the weights. Above all, a simulation that predicts unlimited exponential growth is IMPOSSIBLE.
As usual, one commenter hits the mark perfectly and tersely:
I just can’t get over the irony of a business built on predicting home prices attributing its setbacks to “home pricing unpredictability.”Labels: AI point-missing, Emersonian justice, SES