AI Learns Behavioral Science - ChatGPT, Please Buy Me a Party Dress

A Perfect Marriage

One of the consumer applications of AI and specifically LLMs (large language models), which has sparked much discussion, is their potential to take over our shopping. LLMs will know everything we need, and, of course, everything that is available. Would it not lead to a perfect marriage, providing just the right products to consumers and leaving them perfectly satisfied? Behavioral science research indicates that this marriage requires strong counseling, informed by insights into consumer shopping behavior and attitudes.

Marital Discord

Unfortunately, this prediction rests on several fallacies that are bound to lead to significant marital discord, in this marriage of LLMs and consumers.

Firstly, it assumes that consumers view shopping as a chore. It may be so in certain developed markets and for some routinely bought products.  But in developing markets, shopping more often than not is an engaging activity that is a source of entertainment and even joy. This has become even more pronounced in the age of online shopping, and consumers in China, India, and Southeast Asia spend endless hours browsing and salivating over products and services for sale.

Increasingly, it is also a social activity when they buy on social media sites from their friends or influencers, seek their friends’ opinions about how they will look wearing a particular dress or which color is likely to suit them the best. If LLMs intend to deprive consumers of this rich exploration and social engagement, they are starting on the wrong foot.

Secondly, while consumers often engage in elaborate exploration, this does not necessarily imply that they are engaging in a systematic evaluation or using what economists call a rational choice model. Research indicates that most consumer decisions are driven by spontaneous, instinctive System 1 rather than by evaluative, rational, and elaborate System 2. LLMs are designed to conduct in-depth research and analyze the findings systematically. Research indicates that when consumers make choices in this elaborate way, they are often less satisfied with their choices later than when they make choices quickly and spontaneously.

Thirdly, this spontaneity in our behavior is not a conscious preference, but is a result  of years of evolutionary development. Evolution has programmed our minds in ways most conducive to our survival and propagation. For instance, consumers make quick decisions based on System 1, as such decisions were important for survival in the savannah.

Gad Saad, in his book, The Consuming Instinct, states that, “the great majority of consumption acts can be mapped onto one of four Darwinian overriding pursuits, namely, survival (preference for the fatty smoked meat), reproduction (offering flowers as part of an elaborate courtship ritual), kin selection (buying a gift for my nephew), and reciprocity (organizing the bachelor party).” How will LLMs, which have been programmed to argue with data and logic, mimic this Darwinian consumer behavior?

Fourthly, stemming from these Darwinian instincts and other factors, consumer behavior is actually largely driven by a set of short-cuts that the behavioral scientists refer to as heuristics. Common heuristics discussed by scientists are the availability heuristic, the representative heuristic, and the anchoring heuristic. Using heuristics means that we ignore many facts and detailed specifications of the choices we have and use a quick shortcut to make the choice.

As an example, after detailed evaluation of tens of automobile models, the consumer may finally opt for one that is the most popular (availability), or comes from Germany (provenance), or is available in a unique shade of red, which makes the consumer feel royal and special (affect heuristic). These heuristics are often contextual, highly influenced by social interactions and state of mind. LLMs lack this context, and their recommendations are likely to be clinical and dry, which may not resonate with consumers.

Lastly, behavioral economist Richard Thaler says in his book, Misbehaving, there are two types of utilities that the consumer derives from shopping. The obvious one is the acquisition utility - the joy of buying a product which is relevant for the consumer needs and adds value to them. But then there is also a transaction utility, the satisfaction derived from concluding a purchase to your benefit, where you feel that you made a smart decision, that you managed to not only get a good product, but also secured a good deal and made a smart choice. Transaction utility makes the consumer feel good and proud of her ability and good fortune. If the joy of transaction is taken away by LLMs, the consumer is surely likely to feel cheated.

So How Bo LLMs Fit In?

Clearly, the starting point is that LLMs should enhance the consumer shopping experience, including the final satisfaction consumers feel after the purchase. Hence, removing the pleasurable and meaningful aspects of the shopping process is not a sound starting point. The trick is for LLMs to integrate themselves in the shopping process and enhance the key sources of meaning and joy. If social interaction enhances the shopping experience, LLMs should explore how to further integrate it. 

If browsing a range of attractive products is engaging, LLMs should offer their advice on top of the browsing, not instead of it. Such integrations should be guided by a sensitive exploration of consumers’ feelings and experiences. And, of course, finally the achievement of having secured a wonderful product at a bargain, which is the neighbour’s envy, needs to be entirely the consumers’ - LLM can be seen to assist, no substitute.