Like any tool, behavioural insights can be used for harm as well as good. Companies often use the concept of friction to their advantage, making it very easy to sign up to their services and very difficult to unsubscribe. Governments too may introduce this kind of “sludge”, implementing administrative burdens that discourage certain groups from accessing public services, for example.
Critiques of behavioural insights can be boiled down to two main charges: one, that the approach is “manipulative”; and two, that it is “paternalistic”. These are legitimate criticisms and our goal is not to rebut or minimise them; we explore them in more depth in the book.
Here, we focus on offering a basic framework to guide the responsible application of behavioural insights. This framework involves four factors, two of which relate to manipulation (i.e., how an intervention works):
Control: How easy is it for the relevant actor to resist the intervention, considering the context and their capabilities?
Transparency: How reasonable is it to expect that the target of an intervention will perceive its intention, at the moment of decision or later? The other two relate to paternalism (i.e., what behaviour is being influenced): extent of consequences. What are the likely harms and benefits of adopting or not adopting the behaviour, and where do they fall?
Strength of preferences: How strong and settled are the preferences of the people being influenced? What is the strength of evidence informing this judgment?
The diagram below shows how the two factors in each group inform each other. The level of transparency influences how much control people have, while the consequences of the behaviour need to be weighed against the strength of intentions involved. Considering these 4 factors offers a more structured way to assess ethical issues when applying behavioural insights.
In the book, we give more detail on the various aspects of this framework. In this excerpt, we cover some of the core principles.
One aspect of control deals with the basic choices offered to someone. These are often presented as a “ladder” that starts at the bottom with just providing information, and moves up to enabling or guiding choices, providing incentives, then disincentivising, restricting choice, coercion, and eliminating choice. Importantly, the intensity of influence can vary even on the same rung of this ladder, which represents the other aspect of control.
For example, providing information could range from posting nutritional data on a government website, to a policy that uses all the resources from behavioural insights to make sure that consumers cannot avoid knowing what’s in their potential purchases. The key aspect is how easy it is to resist the intervention. How far we are up the ladder obviously matters, but so does how easily someone can resist their automatic responses to the intervention. We need to make a realistic assessment of this second aspect, using the evidence from behavioural science, plus taking into account factors such as context, history, stress, motivation, and cognitive pressure (where possible).
Obviously, a crucial factor for resisting influence is to be aware of it in the first place, which brings us to transparency. The main question here is how easy it is for someone to become aware that they are being influenced in a particular direction. To be more precise, we can see 3 levels of transparency, which build on each other: someone is aware of the existence of influence, the intent of the influence, and the mechanism(s) it uses. The table below gives some examples.
At this point, we need to note that a core tenet of the behavioural insights approach is that humans have two main ways of thinking that influence their behaviour. One is controlled, slow, deliberative, reflective, and self-aware. It requires focused effort from us, and therefore we may have limited capacity or inclination to make decisions this way. We call this the Reflective System. The other process is uncontrolled, fast, intuitive, and unconscious. Since it occurs outside our awareness, it also requires little to no effort from us, and therefore we can make many decisions this way without fatigue. We called this the Automatic System.
Returning to our “ladder”, it is important to note that these levels do not map directly onto the Reflective and Automatic systems. It is not the case that the more the Reflective System is involved, the more transparent things are. For example, the system of traffic lights works because of our automatic association of red with danger and caution. We may instinctively brake (or speed up) when the lights change. But this reliance on the Automatic system, plus the intention involved, is all completely transparent. On the other hand, we may get influenced in non-transparent ways even when our Reflective system is mostly engaged, as the magazine subscription example in the table shows.
We spend less time talking about consequences and intentions, because they relate to paternalism in general rather than behavioural insights specifically. But they are still important. The starting point for considering consequences is whether the harms or benefits from a behaviour fall on the person who acts, or other people. The former is paternalistic; the latter gives government more scope to act. But even for paternalism, the level of harm involved is important. For example, in many countries there has been controversy around government attempts to improve diets by reducing access to large portions of high-sugar food items (e.g., soft drinks). While many people may agree they would like to have a healthier diet, a strong theme in the objections is that people’s liberty is being removed and their intentions disrespected.
Compare these policies to a 1998 law in the UK that banned the sale of paracetamol (acetaminophen) in packs larger than 32 tablets for pharmacies, and 16 tablets for non-pharmacies. This change reversed the upward trend of suicides and poisonings, saving an estimated 765 lives over the following decade. The impact on freedom of the food and painkiller policies is very similar: in both cases you can still obtain the product, if you make an additional purchase. But we suggest that most people would perceive them quite differently because of the level of direct and immediate harm involved. One prevents a possible indirect contribution to obesity, the other prevents an immediate and direct contribution to loss of life. Consequences matter.
Yet so do intentions. Obviously, the ethics of suicide become more complicated as the strength and consistency of intent increases, as debates about assisted dying demonstrate. Our final point is that we need to find reliable ways of understanding intentions concerning a behaviour. After all, behavioural science shows that our stated intentions vary greatly according to how choices are presented. People may reverse their intentions when it comes to action. Indeed, we may be presented with an ethical dilemma, whereby we need to act to confirm the strength of people’s intentions. In the case of the painkiller packaging, for a substantial number of people, the intention to self-harm was not strong enough to overcome the relatively small amount of effort required to go to another store.
Ethics need careful consideration. But we want to end with a reflection from a group of academics who, by their own admission, started out in opposition to behavioural insights. They write: “As we became more aware of [its] diverse nature… we found ourselves increasingly advocating some of its insights to policy-makers.” They suggest a “caricatured critique” of behavioural insights “underestimates both its diversity of insights and its potential significance while overestimating some of its moral implications.”
We are inclined to agree.
Behavioural Insights was published in September 2020 by MIT Press. Click here to read more and buy the book.