What impact, if any, does machine intelligence (MI) have on your day-to-day life? Well, when you booked your last flight the price you paid was likely calculated using some form of a machine-driven algorithm. In fact, next time you go to an all-inclusive resort, it may be worth asking other guests what they paid for their vacation package. We all accept that we may be paying slightly different prices, but you could find another family from the same city that booked at the same time yet, your prices differ substantially. Is this fair? It’s likely that MI was used to group guests and generate tiered-pricing depending on certain factors, which may mean that the cost of your vacation was more (or less) than that of the family sitting next to you at the beach – a fact that many consumers would not be comfortable with. How far can companies take MI to optimize pricing?

What is MI anyway?

Before we examine ethics, it’s important to understand what we’re talking about when we use the phrase MI. You’re probably familiar with artificial intelligence (AI), a term that is used interchangeably with machine intelligence (MI). Machine intelligence is a branch of computer science that focuses on building machines that can perform tasks that have, to date, required a level of human intelligence to complete. Machine intelligence sometimes relies on a technique called machine learning (ML). With ML, computational algorithms are trained to interpret, learn from, use, and act on data.

Machine learning takes two forms – supervised (when the output can be compared to an expected result) and unsupervised (when the output cannot be anticipated). Most of the applications that affect a person’s day-to-day lives rely on supervised machine learning.

However, unsupervised machine learning is more closely aligned to how humans learn – through observation, experience, and analogy like pricing, where there is a lot of heavy, complex data to consider, unsupervised machine learning offers a lot of potential, but it’s important to move forward cautiously and ethically to protect consumers and our society.

MI in Pricing Practice

The airline industry is an interesting example of how MI is currently used to adjust prices, sometimes dozens of times in a day to maximize revenue. The industry has used algorithmic black boxes for years, which is why most airlines can’t always isolate why their pricing is set as it is – it’s simply not possible to analyze the algorithm in isolation and as such, difficult to know why or how the output price happened. With an algorithm optimizing prices throughout the day, you and your neighbor can end up paying different prices for the same flight. However, the practice is not necessarily aimed at gouging consumers. There are examples of travel companies that use pricing algorithms to ensure that their customers get the best deal.

[1] Booz Allen Hamilton. The Artificial Intelligence Primer – Distinguishing Hype from Reality In Our New Technological Era. Retrieved from: https://www.boozallen.com/s/insight/thought-leadership/the-artificial-intelligence-primer.html.

Concerns around PD and TC

The debate surrounding the use of MI algorithms in pricing is focused on preventing damage through price discrimination (known as Algorithm Price Discrimination or PD) and collusion (known as Algorithmic Tacit Collusion or TC). Both PD and TC practices raise significant concerns for consumers.

With MI-enabled pricing algorithms, the more a company knows about its consumers the more that company can offer personalized pricing. Personalized pricing refers to the practice of charging different consumers different prices for the same or similar products to maximize both revenues and margins.

As consumer data becomes more heavily tracked and accessible, there is concern that MI algorithms could exploit this information to generate accurate profiles of consumers and acquire a deeper understanding of their purchase behaviors, which will enable price discrimination (PD).

With price discrimination, algorithms help to segment customers into customer groups and can accurately estimate each group’s willingness to pay (WTP), information which a business can then use to determine how much to charge those consumers. Since the consumers in each group have similar traits, they are likely to purchase a product or service at a given price. Pricing models that accurately determine the price consumers are willing to pay consistently [lead] to an increase in revenue (and therefore, profits) compared to a single best price strategy.[1]

The more data and more powerful the algorithms that are used, the more accurate this consumer segmentation can be. Using different tools online allows companies to profile consumers and display different prices accordingly (by zip code, for example).

PD has also allowed for product ranking – in which personalization algorithms are used to provide different search results to certain online users. Home Depot recently did this by steering mobile users towards more expensive products. For consumers, this naturally raises the question “How can I protect myself?” and there’s no easy answer. The fact is that price discrimination is happening, and it’s only going to get worse.

[1] Gautier, A., Ittoo, A. & Van Cleynenbreugel, P. AI algorithms, price discrimination and collusion: a technological, economic and legal perspective. Eur J Law Econ 50, 405–435 (2020). Retrieved from: https://doi.org/10.1007/s10657-020-09662-6

How Regulators Can Address or Subvert Collusion

Price discrimination is a reality, and although pricing collusion is more complex and difficult to accomplish, as our technological capabilities increase, it is a future concern. There cannot be a one-fix solution to prevent pricing collusion. For instance, the more consumers are pushed to deal directly with price-bots (to thwart the transparency that allows rival sellers to collude), the more MI algorithms will learn about the characteristics of individual customers, which opens the door to prices tailored to each customer’s willingness to pay, a profitable strategy for sellers.

However, there are ways that regulators (and ethical companies) can prevent market collusion:

• Test price-bots in a “collusion incubator” to see how market conditions might be tweaked to make a price-fixing deal less likely or less stable.
• A “maverick” firm, with to the incumbents might have a lasting impact; an algorithm programmed to build market share, for instance, might help break an informal cartel.
• Explore whether bots that are forced to deal directly with consumers could be enticed to undercut rivals.
• Test to see if imposing speed limits on responses to changes in rivals’ prices hampers collusion.

The Limitations of MI on Price Discrimination and Collusion

In practice, pricing algorithms can lead to collusion and price discrimination. Cases have certainly been observed, but such examples are a result of traditional pricing algorithms and not a result of true ML algorithms. While it is possible in theory, there are some serious limitations to machine learning-based pricing algorithms. MI algorithms are very costly, and the design stage is error-prone, expensive, time-consuming, and dependent on a point in time that is constantly changing. In addition, there are other limitations:

• Collaboration would be required for collusion to occur (especially in the exchange of data)
• Powerful computational abilities are needed to implement ML algorithms
• A shift from supervised data learning to unsupervised data learning would be necessary
• Human errors would need to be eliminated
• It is unlikely that sophisticated machine learning-based pricing algorithms can be effectively deployed in highly competitive markets, with large numbers of differentiated products and low barriers to entry (at least anytime soon).

Conflicting goals between competitors present a challenge for MI. While MI algorithms for pricing may be possible in theory, we’re unlikely to see this used as a widespread practice because at the core of MI is the assumption that all companies have the same goals. In reality, every company has subtle differences and nuances that would be very difficult to anticipate on a wider scale.

While we’re not there yet, we should keep the possibility of MI-driven collusion and price discrimination top of mind. A study conducted by the Commonwealth Scientific and Industrial Research Organisation’s (CSIRO) Data61 found that AI can influence human behavior and exploit the vulnerabilities of an individual’s habits and patterns. The human behavior research involved having humans play three games against a computer – a simple experiment that allowed the machine to learn the player’s behavior and patterns and then act as their trustee.

We’ve also seen this happen with product ranking – where traditional algorithms learned consumer behavior through cookies and web searches and then ranked products and services considering that data. Thus, if machines can understand context, explain their reasoning, and learn from less data, they can allow companies to deploy finer-grained price discrimination and collusion; and maybe influence human’s willingness to pay. This is the next step – from machine learning to machine intelligence influencing human behavior; and it raises ethical concerns, thus imposing an even bigger obligation on how companies use consumer data.

This leads us back to the question of how consumers can protect themselves. One option is to rely on government regulators for protection. For example, the Federal Trade Commission (FTC)’s BOTS ACT is one piece of legislation designed to protect consumers from ticket bots, software that buys up big blocks of tickets and then resells those tickets to consumers for exorbitant prices. Of course, this practice is different from price discrimination, which although unethical, is not illegal. This means that consumers need to look beyond government regulations for protection.

Here are some tips for consumers:

• Shop for products anonymously by browsing in private mode or incognito
• Clear all browser caches and cookies
• Search for products using different browsers to test prices
• Search for products using different devices to test prices
• Shop on a Windows computer instead of a Mac. Research has shown that Mac users tend to have higher incomes and spend more online, which has prompted some companies to charge those Apple users more. This was tested by The Wall Street Journal, which found that Orbitz displayed higher-priced hotels to Mac users than to PC users.
• Ask a friend or family member in a different location to search for the same product and see if they get a different price
• Use price trackers to track the prices of products over time. These trackers often alert consumers when a price drops.
• Apps like Paribus help users get money back from retailers who are displaying different prices. In exchange for a cut of the refund, Paribus negotiates a refund on behalf of the consumer.

While the technology may not be in place to make price collusion a reality, price discrimination is certainly happening and is something that consumers should be aware of. To avoid being duped into paying higher prices because of location, net worth, past shopping history, etc., consumers must shop wisely, protecting as much of their purchase behavior as possible. Now more than ever, it’s important to take the time to research and compare in order to get the best possible price.