Skip to navigationSkip to content
Close
Pricing algorithms can learn to collude with each other to raise prices

Pricing algorithms can learn to collude with each other to raise prices

Read more on MIT Technology Review

Contributions

  • Engineering constraints on software applications used to provide a protective moat. Apps could not do much, so they could not do much harm. For the past decade, apps have operated without material limits on processing power, memory, storage, and bandwidth, enabling giant leaps forward in functionality

    Engineering constraints on software applications used to provide a protective moat. Apps could not do much, so they could not do much harm. For the past decade, apps have operated without material limits on processing power, memory, storage, and bandwidth, enabling giant leaps forward in functionality. This has introduced new issues for software developers. Coding skills are not longer enough to ensure success. If the past two years have shown us anything, it is that technology can inadvertently do great harm when those who create it fail to anticipate side effects and unintended consequences.

    In this case of colluding algorithms, the outcome may have been unexpected, but it might also have been welcome at Amazon. This is one reason why I believe it is essential that all AI and algorithms be transparent. They must be able to explain their choices, demonstrate efficacy and safety, and their freedom from bias. This can and should be done now. It will slow things down a bit, but with great benefits to a large number of people.

  • AI Algorithms will finally sorts of harmful, illegal and unethical things if we don't set up the parameters and guidelines for which they need to operate within. This should be seen as just an extension to free market regulation but for machines.

  • We need to demand that algorithms adhere to the same ethical codes we hold humans to.

    Technology can be used to liberate us from human corruption, or it can invisibly institutionalize bias, making it impossible to undo. It is a critical time in history to ensure it is the former and not the latter.

  • And the machines learned that higher prices are good cuz...? Algorithmic bias always starts with people. Please stop debating this. Discuss what the appropriate response/intervention is instead. Lots of gray area there.

  • This brief piece reminds me of Michael Eisen’s blog post ‘Amazon’s $23,698,655.93 book about flies’ which I read about in a book (http://www.michaeleisen.org/blog/?p=358 ), as well as O’Neil’s Weapons of Math Destruction. Whether it is collusion or simply unintended consequences, these situations are

    This brief piece reminds me of Michael Eisen’s blog post ‘Amazon’s $23,698,655.93 book about flies’ which I read about in a book (http://www.michaeleisen.org/blog/?p=358 ), as well as O’Neil’s Weapons of Math Destruction. Whether it is collusion or simply unintended consequences, these situations are good examples of how difficult it is to think through all of the possible outcomes and their consequences, which is the higher level point here. The reach of our actions might be exceeding our ability to think about and anticipate their consequences. (Johnson’s book Farsighted discusses this topic with some depth.)

  • Interesting that the algorithms priced - colluded - almost to the monopoly price, but not quite. Why? Is that a human or an AI optimization?

  • Very interesting “emergent” behavior. Not unlike how complex patterns can “emerge” out of things as simple as cellular automata.

  • There’s nothing we can’t ruin.

  • Bad algorithms is how AI will ruin the world... not some Terminator-style show down

  • Can a framework be established? how do u ask a machine to behave it self if you allow it self learn and improve? I am a novice!

  • Learning from the, er, best...

  • The fundamental problem is that consumers do not have the ability to create algorithms that react in kind. This represents an opportunity to me. A disruptor could produce technology that would allow consumers to work in concert to manage demand to manipulate prices as well.

    The federal trade laws have

    The fundamental problem is that consumers do not have the ability to create algorithms that react in kind. This represents an opportunity to me. A disruptor could produce technology that would allow consumers to work in concert to manage demand to manipulate prices as well.

    The federal trade laws have clearly not kept up with technology though. Virtually every major product is priced fixed these days. Go to any vendor anywhere in the country online or not and find a given product offered below a Brand's designated minimum advertised price. The Federal Trade Commission is virtually useless these days in protecting consumers.

    Don't get me started on the patent abuses... the most obvious variations on a design are awarded patents with no real innovation involved. When lawyers spend more energy on design description wording than inventors do coming up with ideas you are not protecting ideas, you are simply creating a hindrance to competitive products on the market.

  • Powerful