We let AI run our real estate industry at our peril

The debacle at Zillow shows that AI will keep going wrong in real estate settings

Max Chesapeake
4:59 PM, Apr 21, 2022

Artificial Intelligence written on a typewriter
Photo by Markus Winkler on Unsplash

Back in 2018 Elon Musk proclaimed that Tesla would be self-driving in the next year, I recall that we were told that you could call the car from the other side of the country and it would arrive in front of you the next day.

4 years on and it is, well, not even close. I mean, they now have self-driving of sorts, but if clinging on in sheer terror while constantly correcting the car is autonomous driving then I will give it a miss.

What does that have to do with real estate? Well for a while a Zillow the online real estate service was valuing houses by AI. With that cost chosen it was making purchases.

The software, constantly fed with house data, images, valuations and sale data was pattern matching the current house with what had sold recently. Had sold, which is the crucial part. The software was acting on slightly out of date data. Just far enough behind the curve to make a difference. When the market mood changed and the trend just started to go off on another direction, the AI didn't pick it up.

We in real estate constantly talk about the feel, the gut instinct, what our heart is saying, we know when we get it and we know when buyers and sellers are getting it. We also know that the mood is contagious, it travels between people, especially when it is anxiety.

Last year the market got a little anxiety, and everyone knew it, but the software doesn't have feelings, I mean we are all told that AI can detect this, but it didn't. For about 6 months it overspent by $500 million on a bunch of houses. In the end Zillow had to can 25% of the staff and sell 7000 properties to recover.

As we advance in this technological age we place more and more emphasis on AI being the tool to manage our ever expanding complexity of life. From food production and delivery, driving our cars to how we should spend our free time with recommendations on Netflix.

But we are placing these systems at the heart of problems they are I'll equipped to solve. We don't actually need an AI system to value a house, me need a market of buyers and sellers, they themselves will choose the value and that value, which outsiders can disagree with, is the right value because it was one agreed on by seller and buyer.

I've worked in financials where nobody wants to make a decision anymore, they need the computer to work it out, and when the value looks wrong they get the engineers to explain the outcome, that outcome is then accepted. Even if it is 2 plus 2 is 5. Nobody wants to say that the outcome is wrong because nobody wants to go against the computer.

There will of course be another attempt at this which will end in failure. The engineers will say that the problem can be solved with more logic, more training and more data, but it will not. Because the AI cannot model the feel.