5 ms·

Different companies using the same AI for pricing support purposes could easily tip into illegal practices if there is a feedback loop from company to model. Th

by wjnc 4y ago

Different companies using the same AI for pricing support purposes could easily tip into illegal practices if there is a feedback loop from company to model. That’s just plain old collusion via a middleman. Lo and behold: “To arrive at a recommended rent, the software deploys an algorithm — a set of mathematical rules — to analyze a trove of data RealPage gathers from clients, including private information on what nearby competitors charge.” Dark times if the FTC isn’t picking this up.

acdha 4y ago
I’d also add that this is going to fuel the importance of explainability in these systems. Even if your motives are completely fair, not being able to say why the system produced a particular output can look really bad in court. There isn’t an easy way to distinguish “they have no idea how this AI system works” from “they built it to avoid leaving a paper trail”.
boole1854 4y ago
> Different companies using the same AI for pricing support purposes could easily tip into illegal practices if there is a feedback loop from company to model. While there are anti-competitive concerns with this kind of system, one of the points I was trying to make is that a problem requires more than there be simply a 'feedback loop'. Even in a market with 'perfect' competition, there can be feedback loops: e.g., an individual farmer might price their grain by checking the current market price which is tracked by a third party, and that third party determines the 'current market price' by querying the prices set by all farmers. The reason this isn't necessarily anti-competitive and can be compatible with 'perfect' competition is due to the other characteristics of the market (all participants are price takers, etc), not due to a lack of coordination between producing firms and market analytics firms.
- wjnc 4y ago
  I agree. I was thinking about granularity / specificity but wasn't able to form a coherent train of thought. I'll give it a try. In my part of the world rents are often very naively indexed, say with CPI or CPI+. That is currently an issue, but usually pretty well functioning. Any individual has no influence on the CPI. So the CPI has a feedback loop, but no characteristic of influence. In an AI world that characteristic of influence is indeed not clear at first. I can only expect this to be a pretty dark AI considering the 'Great Succes' it is for the owners. I expect it to vacuum data from brokers in order to calculate a "willingness to pay" combined with a "propensity to pay" to calculate individual and portfolio optimal increases. How is this (/ could this be) collusion? Because the rent increase on property X1, owner Y1 (that is private information, at least I presume for the US rental contracts are 1-on-1 and not public information) is input for all the distributions over properties X and owners Y. I think your point that other market characteristics have a large impact on collusion (yes / no / degree) is subtle as well.

by wjnc 4y ago

acdha 4y ago

I’d also add that this is going to fuel the importance of explainability in these systems. Even if your motives are completely fair, not being able to say why the system produced a particular output can look really bad in court. There isn’t an easy way to distinguish “they have no idea how this AI system works” from “they built it to avoid leaving a paper trail”.

boole1854 4y ago

> Different companies using the same AI for pricing support purposes could easily tip into illegal practices if there is a feedback loop from company to model. While there are anti-competitive concerns with this kind of system, one of the points I was trying to make is that a problem requires more than there be simply a 'feedback loop'. Even in a market with 'perfect' competition, there can be feedback loops: e.g., an individual farmer might price their grain by checking the current market price which is tracked by a third party, and that third party determines the 'current market price' by querying the prices set by all farmers. The reason this isn't necessarily anti-competitive and can be compatible with 'perfect' competition is due to the other characteristics of the market (all participants are price takers, etc), not due to a lack of coordination between producing firms and market analytics firms.

wjnc 4y ago
I agree. I was thinking about granularity / specificity but wasn't able to form a coherent train of thought. I'll give it a try. In my part of the world rents are often very naively indexed, say with CPI or CPI+. That is currently an issue, but usually pretty well functioning. Any individual has no influence on the CPI. So the CPI has a feedback loop, but no characteristic of influence. In an AI world that characteristic of influence is indeed not clear at first. I can only expect this to be a pretty dark AI considering the 'Great Succes' it is for the owners. I expect it to vacuum data from brokers in order to calculate a "willingness to pay" combined with a "propensity to pay" to calculate individual and portfolio optimal increases. How is this (/ could this be) collusion? Because the rent increase on property X1, owner Y1 (that is private information, at least I presume for the US rental contracts are 1-on-1 and not public information) is input for all the distributions over properties X and owners Y. I think your point that other market characteristics have a large impact on collusion (yes / no / degree) is subtle as well.