Jump to content
I Forge Iron

Pricing algorithms learn to collude


Recommended Posts

I've recently come across a few articles on Artificial Intelligence research which suggests that independent price setting algorithms have "learned" to collude with one another to maximize profitability.  Researchers are still working out whether this is happening in the real world.

One of the most distressing comments in the article is that the researchers mentioned that these algorithms are capable of colluding without direct communication.  The AI programs learn to respond to their digital competitors price in a way that maximizes the "punishment" to whoever gives the lower price.  Within a few iterations, the programs "learn" that it's mutually beneficial to set the price just below monopoly level.  This gives the impression of competitive market without the risk of accepting less profit.  The researchers ran tests with more AI "competitors" which didn't change the results.  All the AI programs learned to avoid competitive pricing.  Human nature being what it is, I think it's very possible that plausible deny-ability is potent incentive.  "We didn't teach it to cheat", isn't the same as "we taught it right from wrong".  The article suggests this AI behavior doesn't meet the current legal definition of collusion.  Even if it did, proving real-world examples of this are going on may be very difficult.  

Interestingly, studies done on human subjects suggests that people were unable to achieve the same level of price fixing without direct communication. 

I think that's an erroneous conclusion.  People programmed the algorithm that taught itself to cheat.  The AI is just a tool allowing people to bypass a human limitation.

One of the more potent aspects of this tool may be the way it illustrates how people often separate stated intentions from measured outcomes to obscure what they're really trying to achieve.

 

Link to comment
Share on other sites

Hmmm, deliberate or unintended consequence? AI isn't, its just really complex sophisticated programming right now and software has been "learning" for a long time. If the intent is to maximize profit then isn't it basic logic different systems weigh the results of actions and adjust? Other pricing software is just another variable on the landscape to be dealt with. They interact and reach the least energy state, like any other dynamic system. 

Seems pretty predictable to me.

Frosty The Lucky.

Link to comment
Share on other sites

  • 1 month later...
On 2/13/2019 at 4:59 PM, rockstar.esq said:

Within a few iterations, the programs "learn" that it's mutually beneficial to set the price just below monopoly level.  This gives the impression of competitive market without the risk of accepting less profit.  The researchers ran tests with more AI "competitors" which didn't change the results.  All the AI programs learned to avoid competitive pricing.

Speaking as a long time computer nerd, that computers don't really learn anything. Learning implies understanding. Understanding implies a certain amount of applied intuition. A human would be able to set the price of an object based on that understanding without any data from the market and probably be able to be competitive. The AI won't ever reach that level of understanding. On the other hand the NYSE selling floor has been replaced by a server room, because AI's *can* chew through data much faster, recognize patterns in said data and initiate actions in response to those patterns. but some human with a nominal understanding of the stock market programed the pattern into it, and told it what to do through a very long If-then (and/or/) processes.

What most AI-developers are now calling 'learning' is really giving another layer, instructing it to adjust itself according to a few parameters. The only thing here is that maybe everybody's going to need AI's to shave off that few extra cents worth of profit. Keep in mind, when businesses talk about 'competitive pricing' they're not saying it's the lowest. they're saying it's the most profitable. Because there will always be those guys that go, "aha! that thingamajig costs a thousand dollars more! it MUST be better"

On 2/13/2019 at 5:05 PM, Frosty said:

If the intent is to maximize profit then isn't it basic logic different systems weigh the results of actions and adjust? Other pricing software is just another variable on the landscape to be dealt with. They interact and reach the least energy state, like any other dynamic system. Seems pretty predictable to me....

There's another intent- to generate more profit for the company using the algorithms. Enough 'more profit' that company are willing to buy their services. But yep. this is exactly what AI does.

 

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...