The consequences of cosy AI collusion

“People of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices.”

That, of course, is Adam Smith in the Wealth of Nations, noting the ever-present threat that collusion poses to competition.

Price fixing is not in the public interest, which is why governments intervene to prevent it. But what is a fair price? One yardstick might be that pricing should be consistent – the vendor selling the same item at the same price irrespective of who’s buying it. However, that’s complicated by what you mean by the same item – for instance, is a taxi ride at rush hour or chucking-out time the same as one taken at 11 o’clock in the morning? Arguably, these are different services that justify different prices. If demand exceeds supply at certain times, then rationing is inevitable by one means or another.

In the first century, Publilius Syrus said that “everything is worth what its purchaser is willing to pay for it”. In supermarkets, you can see how some items, such as premium brands of chocolate and ice cream, regularly cycle between two price points. This doesn’t appear to be about the balance of supply and demand, but about selling to two different types of consumer – those who’ll pay a modest premium for an occasional treat and those who’ll pay what it takes to get what they want when they want it. Then there’s the fact that in some markets people simply expect to pay more for the same item than those in other markets: the point at which consumers make the distinction between a reasonable mark-up and an outrageous rip-off is not consistent and thus neither are prices.

Looking ahead, we can expect a whole extra dimension of complexity – given that prices are increasingly set by computers not humans. E-commerce systems now gather so much relevant information that only a pricing algorithm can process it all. But to what end?

In a somewhat alarming article for VoxEU, Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolò, Sergio Pastorello describe their research into the implications:

“Already in 2015, more than a third of the vendors on Amazon.com had automated pricing… and the share has certainly risen since then – with the growth of a repricing software industry that supplies turnkey pricing systems, even the smallest vendors can now afford algorithmic pricing.”

Moreover, the most advanced programs aren’t like the “the traditional revenue management systems long in use by such businesses as airlines and hotels”, but are much more autonomous and unpredictable – using AI technology to learn the most effective pricing strategies (effective at extracting the most money from the consumer, that is).

So, here’s the big question: if these AI systems are free to discover the most profitable pricing strategies without human supervision, will they collude with one another to fix prices?

Today’s AI systems, though impressive, are basically trial-and-error engines, lacking even the slightest shred of consciousness. Surely, then, it would be beyond their capabilities to spontaneously organise themselves into a price-fixing cartel – especially under the conditions of a complex and competitive marketplace.

Calvano and his colleagues decided to put that assumption to the test – and here’s what they discovered:

“…we construct[ed] AI pricing agents and let them interact repeatedly in controlled environments that reproduce economists’ canonical model of collusion, i.e. a repeated pricing game with simultaneous moves and full price flexibility. Our findings suggest that in this framework even relatively simple pricing algorithms systematically learn to play sophisticated collusive strategies.”

Uh-oh. Adam Smith was right – pricing strategies on the part of multiple market players do tend toward cosy collusion not cut-throat competition. If the mental befuddlement of “merriment and diversion” is no obstacle, that’s because consciousness isn’t required at all – mindless calculation is all you need.

That collusion can spontaneously arise without criminal intent makes it all the more of a regulatory nightmare:

“…the algorithms leave no trace of concerted action – they learn to collude purely by trial and error, with no prior knowledge of the environment in which they operate, without communicating with one another, and without being specifically designed or instructed to collude. This poses a real challenge for competition policy.”

The authors conclude that those calling for laws to restrict the use of such algorithms have a point.

I would add that online retailers should also be calling for regulation. In the short-term, most of the the AI firepower may on their side, but how long before ordinary consumers have their own AI agents – algorithms designed to find the ideal price at which to buy rather than sell?

One can easily envisage a digital assistant that could trawl through online retail sites to find the lowest price for a particular item. Various price comparison sites already exist of course, but imagine an AI agent capable of learning from experience and adapting its methods. Now, what if collusion turns out to be the most effective strategy for the AI agents on our side? Instead of the spontaneous organisation of price-fixing cartels, we’d have the spontaneous organisation of consumer boycotts to force prices down.

Without the help of AI, the coordination costs of collective action and asymmetric access to information tend to put consumers at a disadvantage to the businesses we buy from. Individually, we are small and they are big. We, therefore, depend on unimpeded competition in market place and/or effective regulation to get a fair deal.

Retailers should think very carefully before using price-gouging algorithms to undermine the status quo. Any additional advantage they gain could be lost (and then some) if they end up unleashing the power of consumer AI.

---