This proposal lets the actor with the keenest interest and best knowledge of the property set a value, and lets the 'challenge' occur in the marketplace, by someone who's willing to buy it at that price. It lets the "brokers" and other experts of the world set the assessed value, by their actions (or potential actions), at a higher resolution than the current processes (which wait for actual sales or later bureaucratically-settled re-assessments).
"... people are not experts at property valuation, and would have to spend a significant amount of time and mental effort figuring out what self-assessed value to put for their house, and they would complain much more if they accidentally put a value that’s too low and suddenly find that their house is gone."
The solution: users would choose an AI to generate real time valuations.
In the property/personal home context, I think this is a bad idea, because if the AI gets it wrong, and you're forced to sell your house, that is bad. However, this idea could work very well in other types of property markets (this was all discussed in the article).
Then there is clearly the problem with current market set up, which basically allows people to trade with infinite length property rights, especially on land. This gives incentive to buy and never sell, because you know that new generations will have to live somewhere and you will be the one able to provide this service. Esentially leading back to pure feudalism.
A middle ground solution could be to make the land market with land tax not immediate, ie. sell whenever someone else outbids you, but as auction for the property every, let's say, 7 years.
This would give you enough time to consider investing into the property withou losing it right away due to trivial underpricing. It would also solve the optimal property allocation, just on longer term. Another advantage for agricultural land would be that a piece of land couldn't enter the market unless it underwent the seventh sabbatical year of resting, which you could actually enforce under this system.
Sadly this isn't a solution, it just adds another layer of complexity to the decision making process. AI isn't magically unbiased, there's a multitude of ways for human bias/ particular interests to enter the model, from feature selection to model layout up to interpretation of the results.
So now people have to be both experts at assessing housing and neighborhood values and AI to make an informed decision.
That's the big problem of market-based social designs: There's always the fundamental asymmetry between a professional, better-equipped actor (a corporation specialized in that particular market, or a rich person like the bezos example, who can outsource that to employees) and a normal person.
Regulation, society's solution to that problem, tries to level the field: People knowledgable in the problem domain think about possible negative externalities and risks to citizens, and implemenent barriers for abuse.
I'd also add that using an AI also doesn't necessarily mean something opaque. A hand-made decision tree would be preferable to a neural net, for example. The decision tree could prompt you for your human decision if it encounters an outlying, uncertain instance.
Constantly offering to the market is market making. It's a difficult, risky and specialized domain in any asset classes. Most markets cannot support real-time market making for the simple reason that price discovery is expensive and intrinsically tied to liquidity.
Valuing an asset takes work. The process consumes resources. There is a reason we have brokers for assets where one needs someone standing ready to auction. Making this proposal belies a fundamental misunderstanding of the history of markets.