Of Emotional Machines and Irrational Men
How AI and Nudge Theory can help us nudge our way out of difficult decisions
There are times when technology precedes scientific understanding.
We would have never gotten the steam engine if we didn’t understand how heat transforms into work. We could not have found a vaccine that eradicates smallpox if we did not know how the variola virus works.
But our society, with its political and cultural structure, evolves irrespectively of our understanding of it. Scientists, faced with a messy world where financial markets and Artificial Intelligence (AI) emerge seemingly without a warning, need to answer all sorts of funny questions:
How rational are humans when making decisions?
Do AI-enabled machines understand emotions?
Where is all this going?
In this blog post we take a look at the practice of nudging applied in today’s online markets and its interplay with consumer decision making. We trace its roots to the fields of game theory and behavioural economics and explain how data-driven nudging AI can help improve current practices.
Rationality, game theory and the birth of nudges
Our financial system is a complex organism whose internal mechanisms are yet oblique to us. As our experience with other complex systems such as the Earth’s climate, the human brain and cities continuously confirms, predicting and analysing its behaviour is an impossible feat, if one is aiming for the whole picture.
But not all hope is gone: financial markets can be understood by isolating smaller parts of the picture and discovering regularities in them. Although simplistic and based on a multitude of assumptions, the rules that describe complex systems become our understanding of how the world works and help us regulate our behaviour.
The foundations of today’s economics were laid by John von Neumann and Oskar Morgenstern in the book The Theory of Games and Economic Behaviour , which gave birth to the field of game theory. Far from studying the messy, complex, real-world markets, game theory tried to describe how humans act when facing very simple decisions, which can be described by tables like this one:
You will encounter this popular game elsewhere under the name Battle of the Sexes and Bach or Stravinsky game, but in this post we gave a pandemic twist to it.
Suppose me and you have arranged meeting this evening. We are living in the same city, currently struck by a pandemic. There is a curfew starting at 6 p.m. and all public places are locked down, which leaves us with two options for our meeting point: my place or your place? Now things get trickier: suppose that we forgot to choose a place, it’s already 5:30 p.m. and, for some imaginary reason, we have no way of communicating.
We both need to make a decision soon, considering that:
- I prefer us meeting at my place to avoid the commute.
- You prefer us meeting at your place for the same reason.
- We both prefer to meet each other rather than stay alone at home.
The numbers in the table capture these preferences, which game theorists call utilities. The first number in the box is my utility and the second yours. If I stay at my place and you join me, I get my highest utility, 3. But if you instead stay at your place, then we both get the lowest utility, 0.
Game theory attempts to answer the seemingly simple question: what do we expect to happen tonight?
To the surprise of scientists, even this very simple setting can be hard to analyse. This realization forced scientists to introduce many assumptions in the analysis of a game:
- we are both conscious of our preferences and are acting rationally and consistently based on them
- we know the preferences of the other person
- we have an unlimited capacity to determine which is the best choice for us
These assumptions have received a lot of criticism since their introduction. In his seminal critique, Arrow questions the assumption of rationality and explains how it can lead to failures of predicting the market.
This is where behavioural economics enter the picture. Opposing classical game theory, this field allows for psychological, cognitive, emotional and cultural factors to enter the decision-making process. A celebrated example of how psychology can enter decision making is the concept of a nudge.
In their book Nudge: Improving Decisions about Health, Wealth, and Happiness, Richard H. Thaler and Cass R. Sunstein gave an intuitive definition:
A nudge, as we will use the term, is any aspect of the choice architecture that alters people’s behaviour in a predictable way without forbidding any options or significantly changing their economic incentives. To count as a mere nudge, the intervention must be easy and cheap to avoid. Nudges are not mandates. Putting fruit at eye level counts as a nudge. Banning junk food does not.
The fathers of nudge theory refer to this approach as libertarian paternalism and view the influencers as choice architects. To them, the rational, omniscient and all-powerful decision makers in classical game theory are not humans but imaginary beings called Econs. In contrast, real humans are known for their:
Loss Aversion: We prefer to not lose rather than gain.
Myopia: We make choices based on short term gain, often ignoring long-term costs.
Conformity: We tend to make choices to conform with the behaviour and expectations of others.
Anchoring Bias: The first information offered to us heavily biases our decisions.
Availability Bias: Our imagination makes us over-estimate the likelihood of something happening.
Over-Confidence: We often base our choices on an erroneous over-confidence in our own abilities.
Over-Optimism: We often make choices based on erroneous over-optimistic expectations.
At first sight, this behavioural view of humans seems to paint a gloomy picture: how can we have any expectations on the future of a society comprised of irrational and biased individuals?
The irrationality of a thing is no argument against its existence, rather a condition of it — Friedrich Nietsche
Nudging our way out of irrationality
Instead of giving up or ignoring human irrationality, the technique of nudging embraces it to guide decision-making by appropriately shaping a person’s environment. Although their objectives may differ, the state, public organisations and private companies are all interested in influencing public behaviour towards a specific direction.
When facing important social problems, such as climate change, a health or financial crisis, policy makers are in need of some understanding and control over the public’s decisions:
Although examples of successful nudging have existed for decades, the proliferation of online shopping has created a new world of opportunities. In the digital age, the principles and techniques of nudging may remain unchanged, but their efficacy and influence are significantly increased.
Price anchoring, reviews, default options, smart notifications and visual cueing are a few of the ways in which online stores can increase their credibility, retention rate and, more generally, shape customer experience to be aligned with the business objectives.
Nudging smarter , not harder
The right practices for nudging are hidden in its definition. In essence, the art of nudging lies in its simplicity; spotting bad practices is easy, as they tend to over-emphasise or misinterpret some of the fundamental properties of nudging, often driving the opposite psychological effects from the intended ones.
Below are a few examples of bad nudging practices and the rules that they violate. Having them in mind, you will surely be able to spot mistakes in your next online shopping spree.
#1 Nudging should have a positive attitude
The label “Must go” triggers negative feelings such as urgency and guilt. Although it can be effective, this kind of manipulation is not a nudge, as these need to be aiming at influencing behaviour through positive reinforcement.
#2 Nudging should be subtle
Nudges should not dominate their environment. When Thaler talked about “cheap and easy-to-implement interventions”, he had the constraints of the physical world in mind. Online nudges on the other hand are in general cheap and easy to implement, which may lead to their over-utilisation; a shopping menu cluttered with labels may fail to convey any information at all.
#3 Nudges should be optional
How can one get pass this pop-up without clicking on the overly negative button?
Manipulating the colours and shape of graphical elements in the user interface is a common way of guiding the selections of users. However, discount offers should be presented as alternatives instead of aiming at shaming the consumer for opting out.
Can AI algorithms understand our emotions?
Data-driven automation has become a cornerstone for successful business planning. From discovering patterns in consumer behaviour to predicting customer demand, AI algorithms are an integral part of the decision making process.
Arguably the biggest difference between nudging in the pre-AI and post-AI eras is the abundance of opportunities to personalise experiences in the latter. Remember the fun butt-litter bins at the beginning of this post? Their efficacy largely relied on the universality of football.
Universality, however, comes with a generality that can appear antiquated in today’s personalised market: a nudge that is effective for everyone is probably not as effective for each individual. On the other hand, personalisation can increase the efforts required to build successful nudging strategies, defeating the original aim for cheap and low-cost interventions. How can we keep the best of both worlds?
AI has come to the rescue of many of our automation problems. This one, however, is of particular interest. Remember that behavioural economics revolutionised our understanding of financial markets by making our viewpoint more human-centred. We replaced the improbable Econs with decision-makers guided by their emotions and personalities. And now, we want to delegate this task to AI algorithms: does this defeat our original purpose of injecting the human element in our analysis?
One common misconception when trying to answer this question is that AI algorithms need to able to feel emotions in order to understand ours. This possibility is often faced with mistrust or even fear.
While the question of artificial emotions is primarily philosophical, AI technology has already proven its ability to recognise emotions. From facial emotion recognition algorithms that can detect victims of bullying, to text-based sentiment analysis used for improving user experience, applications in both the public and private sector are flourishing.
Perhaps public opinion is not ready to acknowledge AI algorithms as emotional or humans as irrational. Even more possibly, focusing on the positive rather than negative implications of these traits will take even more time.
Nevertheless, technological progress is based on technical abilities and societal needs, rather than opinions. It inevitably makes us challenge assumptions that cultural inertia arbitrarily imposes on ourselves and our creations.
First, never underestimate the power of inertia. Second that power can be harnessed — Richard H. Thaler, Nudge
Applied Data Science Partners is a London based consultancy that implements end-to-end data science solutions for businesses, delivering measurable value. If you’re looking to do more with your data, please get in touch via our website. Follow us on LinkedIn for more AI and data science stories!
